F5 tts (#59)

* init tts/F5-TTS

* epic hax to download ckpts

* clear default reference audio/text from inference vars

* allow blank inputs for inference-cli to just pull checkpoints without running inference

* f5-tts github workflow to publish docker images

tts/F5-TTS/model/__init__.py

@@ -0,0 +1,7 @@
+from model.cfm import CFM
+
+from model.backbones.unett import UNetT
+from model.backbones.dit import DiT
+from model.backbones.mmdit import MMDiT
+
+from model.trainer import Trainer

tts/F5-TTS/model/backbones/README.md

@@ -0,0 +1,20 @@
+## Backbones quick introduction
+
+
+### unett.py
+- flat unet transformer
+- structure same as in e2-tts & voicebox paper except using rotary pos emb
+- update: allow possible abs pos emb & convnextv2 blocks for embedded text before concat
+
+### dit.py
+- adaln-zero dit
+- embedded timestep as condition
+- concatted noised_input + masked_cond + embedded_text, linear proj in
+- possible abs pos emb & convnextv2 blocks for embedded text before concat
+- possible long skip connection (first layer to last layer)
+
+### mmdit.py
+- sd3 structure
+- timestep as condition
+- left stream: text embedded and applied a abs pos emb
+- right stream: masked_cond & noised_input concatted and with same conv pos emb as unett

tts/F5-TTS/model/backbones/dit.py

@@ -0,0 +1,158 @@
+"""
+ein notation:
+b - batch
+n - sequence
+nt - text sequence
+nw - raw wave length
+d - dimension
+"""
+
+from __future__ import annotations
+
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+from einops import repeat
+
+from x_transformers.x_transformers import RotaryEmbedding
+
+from model.modules import (
+    TimestepEmbedding,
+    ConvNeXtV2Block,
+    ConvPositionEmbedding,
+    DiTBlock,
+    AdaLayerNormZero_Final,
+    precompute_freqs_cis, get_pos_embed_indices,
+)
+
+
+# Text embedding
+
+class TextEmbedding(nn.Module):
+    def __init__(self, text_num_embeds, text_dim, conv_layers = 0, conv_mult = 2):
+        super().__init__()
+        self.text_embed = nn.Embedding(text_num_embeds + 1, text_dim)  # use 0 as filler token
+
+        if conv_layers > 0:
+            self.extra_modeling = True
+            self.precompute_max_pos = 4096  # ~44s of 24khz audio
+            self.register_buffer("freqs_cis", precompute_freqs_cis(text_dim, self.precompute_max_pos), persistent=False)
+            self.text_blocks = nn.Sequential(*[ConvNeXtV2Block(text_dim, text_dim * conv_mult) for _ in range(conv_layers)])
+        else:
+            self.extra_modeling = False
+
+    def forward(self, text: int['b nt'], seq_len, drop_text = False):
+        batch, text_len = text.shape[0], text.shape[1]
+        text = text + 1  # use 0 as filler token. preprocess of batch pad -1, see list_str_to_idx()
+        text = text[:, :seq_len]  # curtail if character tokens are more than the mel spec tokens
+        text = F.pad(text, (0, seq_len - text_len), value = 0)
+
+        if drop_text:  # cfg for text
+            text = torch.zeros_like(text)
+
+        text = self.text_embed(text) # b n -> b n d
+
+        # possible extra modeling
+        if self.extra_modeling:
+            # sinus pos emb
+            batch_start = torch.zeros((batch,), dtype=torch.long)
+            pos_idx = get_pos_embed_indices(batch_start, seq_len, max_pos=self.precompute_max_pos)
+            text_pos_embed = self.freqs_cis[pos_idx]
+            text = text + text_pos_embed
+
+            # convnextv2 blocks
+            text = self.text_blocks(text)
+
+        return text
+
+
+# noised input audio and context mixing embedding
+
+class InputEmbedding(nn.Module):
+    def __init__(self, mel_dim, text_dim, out_dim):
+        super().__init__()
+        self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)
+        self.conv_pos_embed = ConvPositionEmbedding(dim = out_dim)
+
+    def forward(self, x: float['b n d'], cond: float['b n d'], text_embed: float['b n d'], drop_audio_cond = False):
+        if drop_audio_cond:  # cfg for cond audio
+            cond = torch.zeros_like(cond)
+
+        x = self.proj(torch.cat((x, cond, text_embed), dim = -1))
+        x = self.conv_pos_embed(x) + x
+        return x
+    
+
+# Transformer backbone using DiT blocks
+
+class DiT(nn.Module):
+    def __init__(self, *, 
+                 dim, depth = 8, heads = 8, dim_head = 64, dropout = 0.1, ff_mult = 4,
+                 mel_dim = 100, text_num_embeds = 256, text_dim = None, conv_layers = 0,
+                 long_skip_connection = False,
+    ):
+        super().__init__()
+
+        self.time_embed = TimestepEmbedding(dim)
+        if text_dim is None:
+            text_dim = mel_dim
+        self.text_embed = TextEmbedding(text_num_embeds, text_dim, conv_layers = conv_layers)
+        self.input_embed = InputEmbedding(mel_dim, text_dim, dim)
+
+        self.rotary_embed = RotaryEmbedding(dim_head)
+
+        self.dim = dim
+        self.depth = depth
+        
+        self.transformer_blocks = nn.ModuleList(
+            [
+                DiTBlock(
+                    dim = dim,
+                    heads = heads,
+                    dim_head = dim_head,
+                    ff_mult = ff_mult,
+                    dropout = dropout
+                )
+                for _ in range(depth)
+            ]
+        )
+        self.long_skip_connection = nn.Linear(dim * 2, dim, bias = False) if long_skip_connection else None
+        
+        self.norm_out = AdaLayerNormZero_Final(dim)  # final modulation
+        self.proj_out = nn.Linear(dim, mel_dim)
+
+    def forward(
+        self,
+        x: float['b n d'],     # nosied input audio
+        cond: float['b n d'],  # masked cond audio
+        text: int['b nt'],     # text
+        time: float['b'] | float[''],  # time step
+        drop_audio_cond,  # cfg for cond audio
+        drop_text,        # cfg for text
+        mask: bool['b n'] | None = None,
+    ):
+        batch, seq_len = x.shape[0], x.shape[1]
+        if time.ndim == 0:
+            time = repeat(time, ' -> b', b = batch)
+        
+        # t: conditioning time, c: context (text + masked cond audio), x: noised input audio
+        t = self.time_embed(time)
+        text_embed = self.text_embed(text, seq_len, drop_text = drop_text)
+        x = self.input_embed(x, cond, text_embed, drop_audio_cond = drop_audio_cond)
+        
+        rope = self.rotary_embed.forward_from_seq_len(seq_len)
+
+        if self.long_skip_connection is not None:
+            residual = x
+
+        for block in self.transformer_blocks:
+            x = block(x, t, mask = mask, rope = rope)
+
+        if self.long_skip_connection is not None:
+            x = self.long_skip_connection(torch.cat((x, residual), dim = -1))
+
+        x = self.norm_out(x, t)
+        output = self.proj_out(x)
+
+        return output

tts/F5-TTS/model/backbones/mmdit.py

@@ -0,0 +1,136 @@
+"""
+ein notation:
+b - batch
+n - sequence
+nt - text sequence
+nw - raw wave length
+d - dimension
+"""
+
+from __future__ import annotations
+
+import torch
+from torch import nn
+
+from einops import repeat
+
+from x_transformers.x_transformers import RotaryEmbedding
+
+from model.modules import (
+    TimestepEmbedding,
+    ConvPositionEmbedding,
+    MMDiTBlock,
+    AdaLayerNormZero_Final,
+    precompute_freqs_cis, get_pos_embed_indices,
+)
+
+
+# text embedding
+
+class TextEmbedding(nn.Module):
+    def __init__(self, out_dim, text_num_embeds):
+        super().__init__()
+        self.text_embed = nn.Embedding(text_num_embeds + 1, out_dim)  # will use 0 as filler token
+
+        self.precompute_max_pos = 1024
+        self.register_buffer("freqs_cis", precompute_freqs_cis(out_dim, self.precompute_max_pos), persistent=False)
+
+    def forward(self, text: int['b nt'], drop_text = False) -> int['b nt d']:
+        text = text + 1
+        if drop_text:
+            text = torch.zeros_like(text)
+        text = self.text_embed(text)
+
+        # sinus pos emb
+        batch_start = torch.zeros((text.shape[0],), dtype=torch.long)
+        batch_text_len = text.shape[1]
+        pos_idx = get_pos_embed_indices(batch_start, batch_text_len, max_pos=self.precompute_max_pos)
+        text_pos_embed = self.freqs_cis[pos_idx]
+
+        text = text + text_pos_embed
+
+        return text
+
+
+# noised input & masked cond audio embedding
+
+class AudioEmbedding(nn.Module):
+    def __init__(self, in_dim, out_dim):
+        super().__init__()
+        self.linear = nn.Linear(2 * in_dim, out_dim)
+        self.conv_pos_embed = ConvPositionEmbedding(out_dim)
+
+    def forward(self, x: float['b n d'], cond: float['b n d'], drop_audio_cond = False):
+        if drop_audio_cond:
+            cond = torch.zeros_like(cond)
+        x = torch.cat((x, cond), dim = -1)
+        x = self.linear(x)
+        x = self.conv_pos_embed(x) + x
+        return x
+    
+
+# Transformer backbone using MM-DiT blocks
+
+class MMDiT(nn.Module):
+    def __init__(self, *, 
+                 dim, depth = 8, heads = 8, dim_head = 64, dropout = 0.1, ff_mult = 4,
+                 text_num_embeds = 256, mel_dim = 100,
+    ):
+        super().__init__()
+
+        self.time_embed = TimestepEmbedding(dim)
+        self.text_embed = TextEmbedding(dim, text_num_embeds)
+        self.audio_embed = AudioEmbedding(mel_dim, dim)
+
+        self.rotary_embed = RotaryEmbedding(dim_head)
+
+        self.dim = dim
+        self.depth = depth
+        
+        self.transformer_blocks = nn.ModuleList(
+            [
+                MMDiTBlock(
+                    dim = dim,
+                    heads = heads,
+                    dim_head = dim_head,
+                    dropout = dropout,
+                    ff_mult = ff_mult,
+                    context_pre_only = i == depth - 1,
+                )
+                for i in range(depth)
+            ]
+        )
+        self.norm_out = AdaLayerNormZero_Final(dim)  # final modulation
+        self.proj_out = nn.Linear(dim, mel_dim)
+
+    def forward(
+        self,
+        x: float['b n d'],     # nosied input audio
+        cond: float['b n d'],  # masked cond audio
+        text: int['b nt'],     # text
+        time: float['b'] | float[''],  # time step
+        drop_audio_cond,  # cfg for cond audio
+        drop_text,        # cfg for text
+        mask: bool['b n'] | None = None,
+    ):
+        batch = x.shape[0]
+        if time.ndim == 0:
+            time = repeat(time, ' -> b', b = batch)
+
+        # t: conditioning (time), c: context (text + masked cond audio), x: noised input audio
+        t = self.time_embed(time)
+        c = self.text_embed(text, drop_text = drop_text)
+        x = self.audio_embed(x, cond, drop_audio_cond = drop_audio_cond)
+
+        seq_len = x.shape[1]
+        text_len = text.shape[1]
+        rope_audio = self.rotary_embed.forward_from_seq_len(seq_len)
+        rope_text = self.rotary_embed.forward_from_seq_len(text_len)
+        
+        for block in self.transformer_blocks:
+            c, x = block(x, c, t, mask = mask, rope = rope_audio, c_rope = rope_text)
+
+        x = self.norm_out(x, t)
+        output = self.proj_out(x)
+
+        return output

tts/F5-TTS/model/backbones/unett.py

@@ -0,0 +1,201 @@
+"""
+ein notation:
+b - batch
+n - sequence
+nt - text sequence
+nw - raw wave length
+d - dimension
+"""
+
+from __future__ import annotations
+from typing import Literal
+
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+from einops import repeat, pack, unpack
+
+from x_transformers import RMSNorm
+from x_transformers.x_transformers import RotaryEmbedding
+
+from model.modules import (
+    TimestepEmbedding,
+    ConvNeXtV2Block,
+    ConvPositionEmbedding,
+    Attention,
+    AttnProcessor,
+    FeedForward,
+    precompute_freqs_cis, get_pos_embed_indices,
+)
+
+
+# Text embedding
+
+class TextEmbedding(nn.Module):
+    def __init__(self, text_num_embeds, text_dim, conv_layers = 0, conv_mult = 2):
+        super().__init__()
+        self.text_embed = nn.Embedding(text_num_embeds + 1, text_dim)  # use 0 as filler token
+
+        if conv_layers > 0:
+            self.extra_modeling = True
+            self.precompute_max_pos = 4096  # ~44s of 24khz audio
+            self.register_buffer("freqs_cis", precompute_freqs_cis(text_dim, self.precompute_max_pos), persistent=False)
+            self.text_blocks = nn.Sequential(*[ConvNeXtV2Block(text_dim, text_dim * conv_mult) for _ in range(conv_layers)])
+        else:
+            self.extra_modeling = False
+
+    def forward(self, text: int['b nt'], seq_len, drop_text = False):
+        batch, text_len = text.shape[0], text.shape[1]
+        text = text + 1  # use 0 as filler token. preprocess of batch pad -1, see list_str_to_idx()
+        text = text[:, :seq_len]  # curtail if character tokens are more than the mel spec tokens
+        text = F.pad(text, (0, seq_len - text_len), value = 0)
+
+        if drop_text:  # cfg for text
+            text = torch.zeros_like(text)
+
+        text = self.text_embed(text) # b n -> b n d
+
+        # possible extra modeling
+        if self.extra_modeling:
+            # sinus pos emb
+            batch_start = torch.zeros((batch,), dtype=torch.long)
+            pos_idx = get_pos_embed_indices(batch_start, seq_len, max_pos=self.precompute_max_pos)
+            text_pos_embed = self.freqs_cis[pos_idx]
+            text = text + text_pos_embed
+
+            # convnextv2 blocks
+            text = self.text_blocks(text)
+
+        return text
+
+
+# noised input audio and context mixing embedding
+
+class InputEmbedding(nn.Module):
+    def __init__(self, mel_dim, text_dim, out_dim):
+        super().__init__()
+        self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)
+        self.conv_pos_embed = ConvPositionEmbedding(dim = out_dim)
+
+    def forward(self, x: float['b n d'], cond: float['b n d'], text_embed: float['b n d'], drop_audio_cond = False):
+        if drop_audio_cond:  # cfg for cond audio
+            cond = torch.zeros_like(cond)
+
+        x = self.proj(torch.cat((x, cond, text_embed), dim = -1))
+        x = self.conv_pos_embed(x) + x
+        return x
+
+
+# Flat UNet Transformer backbone
+
+class UNetT(nn.Module):
+    def __init__(self, *,
+                 dim, depth = 8, heads = 8, dim_head = 64, dropout = 0.1, ff_mult = 4,
+                 mel_dim = 100, text_num_embeds = 256, text_dim = None, conv_layers = 0,
+                 skip_connect_type: Literal['add', 'concat', 'none'] = 'concat',
+    ):
+        super().__init__()
+        assert depth % 2 == 0, "UNet-Transformer's depth should be even."
+
+        self.time_embed = TimestepEmbedding(dim)
+        if text_dim is None:
+            text_dim = mel_dim
+        self.text_embed = TextEmbedding(text_num_embeds, text_dim, conv_layers = conv_layers)
+        self.input_embed = InputEmbedding(mel_dim, text_dim, dim)
+
+        self.rotary_embed = RotaryEmbedding(dim_head)
+
+        # transformer layers & skip connections
+
+        self.dim = dim
+        self.skip_connect_type = skip_connect_type
+        needs_skip_proj = skip_connect_type == 'concat'
+
+        self.depth = depth
+        self.layers = nn.ModuleList([])
+
+        for idx in range(depth):
+            is_later_half = idx >= (depth // 2)
+
+            attn_norm = RMSNorm(dim)
+            attn = Attention(
+                processor = AttnProcessor(),
+                dim = dim,
+                heads = heads,
+                dim_head = dim_head,
+                dropout = dropout,
+                )
+
+            ff_norm = RMSNorm(dim)
+            ff = FeedForward(dim = dim, mult = ff_mult, dropout = dropout, approximate = "tanh")
+
+            skip_proj = nn.Linear(dim * 2, dim, bias = False) if needs_skip_proj and is_later_half else None
+
+            self.layers.append(nn.ModuleList([
+                skip_proj,
+                attn_norm,
+                attn,
+                ff_norm,
+                ff,
+            ]))
+
+        self.norm_out = RMSNorm(dim)
+        self.proj_out = nn.Linear(dim, mel_dim)
+
+    def forward(
+        self,
+        x: float['b n d'],     # nosied input audio
+        cond: float['b n d'],  # masked cond audio
+        text: int['b nt'],     # text
+        time: float['b'] | float[''],  # time step
+        drop_audio_cond,  # cfg for cond audio
+        drop_text,        # cfg for text
+        mask: bool['b n'] | None = None,
+    ):
+        batch, seq_len = x.shape[0], x.shape[1]
+        if time.ndim == 0:
+            time = repeat(time, ' -> b', b = batch)
+        
+        # t: conditioning time, c: context (text + masked cond audio), x: noised input audio
+        t = self.time_embed(time)
+        text_embed = self.text_embed(text, seq_len, drop_text = drop_text)
+        x = self.input_embed(x, cond, text_embed, drop_audio_cond = drop_audio_cond)
+
+        # postfix time t to input x, [b n d] -> [b n+1 d]
+        x, ps = pack((t, x), 'b * d')
+        if mask is not None:
+            mask = F.pad(mask, (1, 0), value=1)
+        
+        rope = self.rotary_embed.forward_from_seq_len(seq_len + 1)
+
+        # flat unet transformer
+        skip_connect_type = self.skip_connect_type
+        skips = []
+        for idx, (maybe_skip_proj, attn_norm, attn, ff_norm, ff) in enumerate(self.layers):
+            layer = idx + 1
+
+            # skip connection logic
+            is_first_half = layer <= (self.depth // 2)
+            is_later_half = not is_first_half
+
+            if is_first_half:
+                skips.append(x)
+
+            if is_later_half:
+                skip = skips.pop()
+                if skip_connect_type == 'concat':
+                    x = torch.cat((x, skip), dim = -1)
+                    x = maybe_skip_proj(x)
+                elif skip_connect_type == 'add':
+                    x = x + skip
+
+            # attention and feedforward blocks
+            x = attn(attn_norm(x), rope = rope, mask = mask) + x
+            x = ff(ff_norm(x)) + x
+
+        assert len(skips) == 0
+
+        _, x = unpack(self.norm_out(x), ps, 'b * d')
+
+        return self.proj_out(x)

tts/F5-TTS/model/cfm.py

@@ -0,0 +1,279 @@
+"""
+ein notation:
+b - batch
+n - sequence
+nt - text sequence
+nw - raw wave length
+d - dimension
+"""
+
+from __future__ import annotations
+from typing import Callable
+from random import random
+
+import torch
+from torch import nn
+import torch.nn.functional as F
+from torch.nn.utils.rnn import pad_sequence
+
+from torchdiffeq import odeint
+
+from einops import rearrange
+
+from model.modules import MelSpec
+
+from model.utils import (
+    default, exists, 
+    list_str_to_idx, list_str_to_tensor, 
+    lens_to_mask, mask_from_frac_lengths,
+) 
+
+
+class CFM(nn.Module):
+    def __init__(
+        self,
+        transformer: nn.Module,
+        sigma = 0.,
+        odeint_kwargs: dict = dict(
+            # atol = 1e-5,
+            # rtol = 1e-5,
+            method = 'euler'  # 'midpoint'
+        ),
+        audio_drop_prob = 0.3,
+        cond_drop_prob = 0.2,
+        num_channels = None,
+        mel_spec_module: nn.Module | None = None,
+        mel_spec_kwargs: dict = dict(),
+        frac_lengths_mask: tuple[float, float] = (0.7, 1.),
+        vocab_char_map: dict[str: int] | None = None
+    ):
+        super().__init__()
+
+        self.frac_lengths_mask = frac_lengths_mask
+
+        # mel spec
+        self.mel_spec = default(mel_spec_module, MelSpec(**mel_spec_kwargs))
+        num_channels = default(num_channels, self.mel_spec.n_mel_channels)
+        self.num_channels = num_channels
+
+        # classifier-free guidance
+        self.audio_drop_prob = audio_drop_prob
+        self.cond_drop_prob = cond_drop_prob
+
+        # transformer
+        self.transformer = transformer
+        dim = transformer.dim
+        self.dim = dim
+
+        # conditional flow related
+        self.sigma = sigma
+
+        # sampling related
+        self.odeint_kwargs = odeint_kwargs
+
+        # vocab map for tokenization
+        self.vocab_char_map = vocab_char_map
+
+    @property
+    def device(self):
+        return next(self.parameters()).device
+
+    @torch.no_grad()
+    def sample(
+        self,
+        cond: float['b n d'] | float['b nw'],
+        text: int['b nt'] | list[str],
+        duration: int | int['b'],
+        *,
+        lens: int['b'] | None = None,
+        steps = 32,
+        cfg_strength = 1., 
+        sway_sampling_coef = None,
+        seed: int | None = None,
+        max_duration = 4096, 
+        vocoder: Callable[[float['b d n']], float['b nw']] | None = None,
+        no_ref_audio = False,
+        duplicate_test = False,
+        t_inter = 0.1,
+        edit_mask = None,
+    ):
+        self.eval()
+
+        # raw wave
+
+        if cond.ndim == 2:
+            cond = self.mel_spec(cond)
+            cond = rearrange(cond, 'b d n -> b n d')
+            assert cond.shape[-1] == self.num_channels
+
+        batch, cond_seq_len, device = *cond.shape[:2], cond.device
+        if not exists(lens):
+            lens = torch.full((batch,), cond_seq_len, device = device, dtype = torch.long)
+
+        # text
+
+        if isinstance(text, list):
+            if exists(self.vocab_char_map):
+                text = list_str_to_idx(text, self.vocab_char_map).to(device)
+            else:
+                text = list_str_to_tensor(text).to(device)
+            assert text.shape[0] == batch
+
+        if exists(text):
+            text_lens = (text != -1).sum(dim = -1)
+            lens = torch.maximum(text_lens, lens) # make sure lengths are at least those of the text characters
+
+        # duration
+
+        cond_mask = lens_to_mask(lens)
+        if edit_mask is not None:
+            cond_mask = cond_mask & edit_mask
+
+        if isinstance(duration, int):
+            duration = torch.full((batch,), duration, device = device, dtype = torch.long)
+
+        duration = torch.maximum(lens + 1, duration) # just add one token so something is generated
+        duration = duration.clamp(max = max_duration)
+        max_duration = duration.amax()
+        
+        # duplicate test corner for inner time step oberservation
+        if duplicate_test:
+            test_cond = F.pad(cond, (0, 0, cond_seq_len, max_duration - 2*cond_seq_len), value = 0.)
+            
+        cond = F.pad(cond, (0, 0, 0, max_duration - cond_seq_len), value = 0.)
+        cond_mask = F.pad(cond_mask, (0, max_duration - cond_mask.shape[-1]), value = False)
+        cond_mask = rearrange(cond_mask, '... -> ... 1')
+        step_cond = torch.where(cond_mask, cond, torch.zeros_like(cond))  # allow direct control (cut cond audio) with lens passed in
+
+        if batch > 1:
+            mask = lens_to_mask(duration)
+        else:  # save memory and speed up, as single inference need no mask currently
+            mask = None
+
+        # test for no ref audio
+        if no_ref_audio:
+            cond = torch.zeros_like(cond)
+
+        # neural ode
+
+        def fn(t, x):
+            # at each step, conditioning is fixed
+            # step_cond = torch.where(cond_mask, cond, torch.zeros_like(cond))
+
+            # predict flow
+            pred = self.transformer(x = x, cond = step_cond, text = text, time = t, mask = mask, drop_audio_cond = False, drop_text = False)
+            if cfg_strength < 1e-5:
+                return pred
+            
+            null_pred = self.transformer(x = x, cond = step_cond, text = text, time = t, mask = mask, drop_audio_cond = True, drop_text = True)
+            return pred + (pred - null_pred) * cfg_strength
+
+        # noise input
+        # to make sure batch inference result is same with different batch size, and for sure single inference
+        # still some difference maybe due to convolutional layers
+        y0 = []
+        for dur in duration:
+            if exists(seed):
+                torch.manual_seed(seed)
+            y0.append(torch.randn(dur, self.num_channels, device = self.device))
+        y0 = pad_sequence(y0, padding_value = 0, batch_first = True)
+
+        t_start = 0
+
+        # duplicate test corner for inner time step oberservation
+        if duplicate_test:
+            t_start = t_inter
+            y0 = (1 - t_start) * y0 + t_start * test_cond
+            steps = int(steps * (1 - t_start))
+
+        t = torch.linspace(t_start, 1, steps, device = self.device)
+        if sway_sampling_coef is not None:
+            t = t + sway_sampling_coef * (torch.cos(torch.pi / 2 * t) - 1 + t)
+
+        trajectory = odeint(fn, y0, t, **self.odeint_kwargs)
+        
+        sampled = trajectory[-1]
+        out = sampled
+        out = torch.where(cond_mask, cond, out)
+
+        if exists(vocoder):
+            out = rearrange(out, 'b n d -> b d n')
+            out = vocoder(out)
+
+        return out, trajectory
+
+    def forward(
+        self,
+        inp: float['b n d'] | float['b nw'], # mel or raw wave
+        text: int['b nt'] | list[str],
+        *,
+        lens: int['b'] | None = None,
+        noise_scheduler: str | None = None,
+    ):
+        # handle raw wave
+        if inp.ndim == 2:
+            inp = self.mel_spec(inp)
+            inp = rearrange(inp, 'b d n -> b n d')
+            assert inp.shape[-1] == self.num_channels
+
+        batch, seq_len, dtype, device, σ1 = *inp.shape[:2], inp.dtype, self.device, self.sigma
+
+        # handle text as string
+        if isinstance(text, list):
+            if exists(self.vocab_char_map):
+                text = list_str_to_idx(text, self.vocab_char_map).to(device)
+            else:
+                text = list_str_to_tensor(text).to(device)
+            assert text.shape[0] == batch
+
+        # lens and mask
+        if not exists(lens):
+            lens = torch.full((batch,), seq_len, device = device)
+        
+        mask = lens_to_mask(lens, length = seq_len)  # useless here, as collate_fn will pad to max length in batch
+
+        # get a random span to mask out for training conditionally
+        frac_lengths = torch.zeros((batch,), device = self.device).float().uniform_(*self.frac_lengths_mask)
+        rand_span_mask = mask_from_frac_lengths(lens, frac_lengths)
+
+        if exists(mask):
+            rand_span_mask &= mask
+
+        # mel is x1
+        x1 = inp
+
+        # x0 is gaussian noise
+        x0 = torch.randn_like(x1)
+
+        # time step
+        time = torch.rand((batch,), dtype = dtype, device = self.device)
+        # TODO. noise_scheduler
+
+        # sample xt (φ_t(x) in the paper)
+        t = rearrange(time, 'b -> b 1 1')
+        φ = (1 - t) * x0 + t * x1
+        flow = x1 - x0
+
+        # only predict what is within the random mask span for infilling
+        cond = torch.where(
+            rand_span_mask[..., None],
+            torch.zeros_like(x1), x1
+        )
+
+        # transformer and cfg training with a drop rate
+        drop_audio_cond = random() < self.audio_drop_prob  # p_drop in voicebox paper
+        if random() < self.cond_drop_prob:  # p_uncond in voicebox paper
+            drop_audio_cond = True
+            drop_text = True
+        else:
+            drop_text = False
+            
+        # if want rigourously mask out padding, record in collate_fn in dataset.py, and pass in here
+        # adding mask will use more memory, thus also need to adjust batchsampler with scaled down threshold for long sequences
+        pred = self.transformer(x = φ, cond = cond, text = text, time = time, drop_audio_cond = drop_audio_cond, drop_text = drop_text)
+
+        # flow matching loss
+        loss = F.mse_loss(pred, flow, reduction = 'none')
+        loss = loss[rand_span_mask]
+
+        return loss.mean(), cond, pred

tts/F5-TTS/model/dataset.py

@@ -0,0 +1,257 @@
+import json
+import random
+from tqdm import tqdm
+
+import torch
+import torch.nn.functional as F
+from torch.utils.data import Dataset, Sampler
+import torchaudio
+from datasets import load_dataset, load_from_disk
+from datasets import Dataset as Dataset_
+
+from einops import rearrange
+
+from model.modules import MelSpec
+
+
+class HFDataset(Dataset):
+    def __init__(
+        self,
+        hf_dataset: Dataset,
+        target_sample_rate = 24_000,
+        n_mel_channels = 100,
+        hop_length = 256,
+    ):
+        self.data = hf_dataset
+        self.target_sample_rate = target_sample_rate
+        self.hop_length = hop_length
+        self.mel_spectrogram = MelSpec(target_sample_rate=target_sample_rate, n_mel_channels=n_mel_channels, hop_length=hop_length)
+        
+    def get_frame_len(self, index):
+        row = self.data[index]
+        audio = row['audio']['array']
+        sample_rate = row['audio']['sampling_rate']
+        return audio.shape[-1] / sample_rate * self.target_sample_rate / self.hop_length
+
+    def __len__(self):
+        return len(self.data)
+    
+    def __getitem__(self, index):
+        row = self.data[index]
+        audio = row['audio']['array']
+
+        # logger.info(f"Audio shape: {audio.shape}")
+
+        sample_rate = row['audio']['sampling_rate']
+        duration = audio.shape[-1] / sample_rate
+
+        if duration > 30 or duration < 0.3:
+            return self.__getitem__((index + 1) % len(self.data))
+        
+        audio_tensor = torch.from_numpy(audio).float()
+        
+        if sample_rate != self.target_sample_rate:
+            resampler = torchaudio.transforms.Resample(sample_rate, self.target_sample_rate)
+            audio_tensor = resampler(audio_tensor)
+        
+        audio_tensor = rearrange(audio_tensor, 't -> 1 t')
+        
+        mel_spec = self.mel_spectrogram(audio_tensor)
+        
+        mel_spec = rearrange(mel_spec, '1 d t -> d t')
+        
+        text = row['text']
+        
+        return dict(
+            mel_spec = mel_spec,
+            text = text,
+        )
+
+
+class CustomDataset(Dataset):
+    def __init__(
+        self,
+        custom_dataset: Dataset,
+        durations = None,
+        target_sample_rate = 24_000,
+        hop_length = 256,
+        n_mel_channels = 100,
+        preprocessed_mel = False,
+    ):
+        self.data = custom_dataset
+        self.durations = durations
+        self.target_sample_rate = target_sample_rate
+        self.hop_length = hop_length
+        self.preprocessed_mel = preprocessed_mel
+        if not preprocessed_mel:
+            self.mel_spectrogram = MelSpec(target_sample_rate=target_sample_rate, hop_length=hop_length, n_mel_channels=n_mel_channels)
+
+    def get_frame_len(self, index):
+        if self.durations is not None:  # Please make sure the separately provided durations are correct, otherwise 99.99% OOM
+            return self.durations[index] * self.target_sample_rate / self.hop_length
+        return self.data[index]["duration"] * self.target_sample_rate / self.hop_length
+    
+    def __len__(self):
+        return len(self.data)
+    
+    def __getitem__(self, index):
+        row = self.data[index]
+        audio_path = row["audio_path"]
+        text = row["text"]
+        duration = row["duration"]
+
+        if self.preprocessed_mel:
+            mel_spec = torch.tensor(row["mel_spec"])
+
+        else:
+            audio, source_sample_rate = torchaudio.load(audio_path)
+
+            if duration > 30 or duration < 0.3:
+                return self.__getitem__((index + 1) % len(self.data))
+            
+            if source_sample_rate != self.target_sample_rate:
+                resampler = torchaudio.transforms.Resample(source_sample_rate, self.target_sample_rate)
+                audio = resampler(audio)
+            
+            mel_spec = self.mel_spectrogram(audio)
+            mel_spec = rearrange(mel_spec, '1 d t -> d t')
+        
+        return dict(
+            mel_spec = mel_spec,
+            text = text,
+        )
+    
+
+# Dynamic Batch Sampler
+
+class DynamicBatchSampler(Sampler[list[int]]):
+    """ Extension of Sampler that will do the following:
+        1.  Change the batch size (essentially number of sequences)
+            in a batch to ensure that the total number of frames are less
+            than a certain threshold.
+        2.  Make sure the padding efficiency in the batch is high.
+    """
+
+    def __init__(self, sampler: Sampler[int], frames_threshold: int, max_samples=0, random_seed=None, drop_last: bool = False):
+        self.sampler = sampler
+        self.frames_threshold = frames_threshold
+        self.max_samples = max_samples
+
+        indices, batches = [], []
+        data_source = self.sampler.data_source
+        
+        for idx in tqdm(self.sampler, desc=f"Sorting with sampler... if slow, check whether dataset is provided with duration"):
+            indices.append((idx, data_source.get_frame_len(idx)))
+        indices.sort(key=lambda elem : elem[1])
+
+        batch = []
+        batch_frames = 0
+        for idx, frame_len in tqdm(indices, desc=f"Creating dynamic batches with {frames_threshold} audio frames per gpu"):
+            if batch_frames + frame_len <= self.frames_threshold and (max_samples == 0 or len(batch) < max_samples):
+                batch.append(idx)
+                batch_frames += frame_len
+            else:
+                if len(batch) > 0:
+                    batches.append(batch)
+                if frame_len <= self.frames_threshold:
+                    batch = [idx]
+                    batch_frames = frame_len
+                else:
+                    batch = []
+                    batch_frames = 0
+
+        if not drop_last and len(batch) > 0:
+            batches.append(batch)
+
+        del indices
+
+        # if want to have different batches between epochs, may just set a seed and log it in ckpt
+        # cuz during multi-gpu training, although the batch on per gpu not change between epochs, the formed general minibatch is different
+        # e.g. for epoch n, use (random_seed + n)
+        random.seed(random_seed)
+        random.shuffle(batches)
+
+        self.batches = batches
+
+    def __iter__(self):
+        return iter(self.batches)
+
+    def __len__(self):
+        return len(self.batches)
+
+
+# Load dataset
+
+def load_dataset(
+        dataset_name: str,
+        tokenizer: str = "pinyin",
+        dataset_type: str = "CustomDataset", 
+        audio_type: str = "raw", 
+        mel_spec_kwargs: dict = dict()
+        ) -> CustomDataset | HFDataset:
+    '''
+    dataset_type    - "CustomDataset" if you want to use tokenizer name and default data path to load for train_dataset
+                    - "CustomDatasetPath" if you just want to pass the full path to a preprocessed dataset without relying on tokenizer
+    '''
+    
+    print("Loading dataset ...")
+
+    if dataset_type == "CustomDataset":
+        if audio_type == "raw":
+            try:
+                train_dataset = load_from_disk(f"data/{dataset_name}_{tokenizer}/raw")
+            except:
+                train_dataset = Dataset_.from_file(f"data/{dataset_name}_{tokenizer}/raw.arrow")
+            preprocessed_mel = False
+        elif audio_type == "mel":
+            train_dataset = Dataset_.from_file(f"data/{dataset_name}_{tokenizer}/mel.arrow")
+            preprocessed_mel = True
+        with open(f"data/{dataset_name}_{tokenizer}/duration.json", 'r', encoding='utf-8') as f:
+            data_dict = json.load(f)
+        durations = data_dict["duration"]
+        train_dataset = CustomDataset(train_dataset, durations=durations, preprocessed_mel=preprocessed_mel, **mel_spec_kwargs)
+        
+    elif dataset_type == "CustomDatasetPath":
+        try:
+            train_dataset = load_from_disk(f"{dataset_name}/raw")
+        except:
+            train_dataset = Dataset_.from_file(f"{dataset_name}/raw.arrow")
+            
+        with open(f"{dataset_name}/duration.json", 'r', encoding='utf-8') as f:
+            data_dict = json.load(f)
+        durations = data_dict["duration"]
+        train_dataset = CustomDataset(train_dataset, durations=durations, preprocessed_mel=preprocessed_mel, **mel_spec_kwargs)
+            
+    elif dataset_type == "HFDataset":
+        print("Should manually modify the path of huggingface dataset to your need.\n" +
+              "May also the corresponding script cuz different dataset may have different format.")
+        pre, post = dataset_name.split("_")
+        train_dataset = HFDataset(load_dataset(f"{pre}/{pre}", split=f"train.{post}", cache_dir="./data"),)
+
+    return train_dataset
+
+
+# collation
+
+def collate_fn(batch):
+    mel_specs = [item['mel_spec'].squeeze(0) for item in batch]
+    mel_lengths = torch.LongTensor([spec.shape[-1] for spec in mel_specs])
+    max_mel_length = mel_lengths.amax()
+
+    padded_mel_specs = []
+    for spec in mel_specs:  # TODO. maybe records mask for attention here
+        padding = (0, max_mel_length - spec.size(-1))
+        padded_spec = F.pad(spec, padding, value = 0)
+        padded_mel_specs.append(padded_spec)
+    
+    mel_specs = torch.stack(padded_mel_specs)
+
+    text = [item['text'] for item in batch]
+    text_lengths = torch.LongTensor([len(item) for item in text])
+
+    return dict(
+        mel = mel_specs,
+        mel_lengths = mel_lengths,
+        text = text,
+        text_lengths = text_lengths,
+    )

tts/F5-TTS/model/ecapa_tdnn.py

@@ -0,0 +1,268 @@
+# just for speaker similarity evaluation, third-party code
+
+# From https://github.com/microsoft/UniSpeech/blob/main/downstreams/speaker_verification/models/
+# part of the code is borrowed from https://github.com/lawlict/ECAPA-TDNN
+
+import os
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+
+''' Res2Conv1d + BatchNorm1d + ReLU
+'''
+
+class Res2Conv1dReluBn(nn.Module):
+    '''
+    in_channels == out_channels == channels
+    '''
+
+    def __init__(self, channels, kernel_size=1, stride=1, padding=0, dilation=1, bias=True, scale=4):
+        super().__init__()
+        assert channels % scale == 0, "{} % {} != 0".format(channels, scale)
+        self.scale = scale
+        self.width = channels // scale
+        self.nums = scale if scale == 1 else scale - 1
+
+        self.convs = []
+        self.bns = []
+        for i in range(self.nums):
+            self.convs.append(nn.Conv1d(self.width, self.width, kernel_size, stride, padding, dilation, bias=bias))
+            self.bns.append(nn.BatchNorm1d(self.width))
+        self.convs = nn.ModuleList(self.convs)
+        self.bns = nn.ModuleList(self.bns)
+
+    def forward(self, x):
+        out = []
+        spx = torch.split(x, self.width, 1)
+        for i in range(self.nums):
+            if i == 0:
+                sp = spx[i]
+            else:
+                sp = sp + spx[i]
+            # Order: conv -> relu -> bn
+            sp = self.convs[i](sp)
+            sp = self.bns[i](F.relu(sp))
+            out.append(sp)
+        if self.scale != 1:
+            out.append(spx[self.nums])
+        out = torch.cat(out, dim=1)
+
+        return out
+
+
+''' Conv1d + BatchNorm1d + ReLU
+'''
+
+class Conv1dReluBn(nn.Module):
+    def __init__(self, in_channels, out_channels, kernel_size=1, stride=1, padding=0, dilation=1, bias=True):
+        super().__init__()
+        self.conv = nn.Conv1d(in_channels, out_channels, kernel_size, stride, padding, dilation, bias=bias)
+        self.bn = nn.BatchNorm1d(out_channels)
+
+    def forward(self, x):
+        return self.bn(F.relu(self.conv(x)))
+
+
+''' The SE connection of 1D case.
+'''
+
+class SE_Connect(nn.Module):
+    def __init__(self, channels, se_bottleneck_dim=128):
+        super().__init__()
+        self.linear1 = nn.Linear(channels, se_bottleneck_dim)
+        self.linear2 = nn.Linear(se_bottleneck_dim, channels)
+
+    def forward(self, x):
+        out = x.mean(dim=2)
+        out = F.relu(self.linear1(out))
+        out = torch.sigmoid(self.linear2(out))
+        out = x * out.unsqueeze(2)
+
+        return out
+
+
+''' SE-Res2Block of the ECAPA-TDNN architecture.
+'''
+
+# def SE_Res2Block(channels, kernel_size, stride, padding, dilation, scale):
+#     return nn.Sequential(
+#         Conv1dReluBn(channels, 512, kernel_size=1, stride=1, padding=0),
+#         Res2Conv1dReluBn(512, kernel_size, stride, padding, dilation, scale=scale),
+#         Conv1dReluBn(512, channels, kernel_size=1, stride=1, padding=0),
+#         SE_Connect(channels)
+#     )
+
+class SE_Res2Block(nn.Module):
+    def __init__(self, in_channels, out_channels, kernel_size, stride, padding, dilation, scale, se_bottleneck_dim):
+        super().__init__()
+        self.Conv1dReluBn1 = Conv1dReluBn(in_channels, out_channels, kernel_size=1, stride=1, padding=0)
+        self.Res2Conv1dReluBn = Res2Conv1dReluBn(out_channels, kernel_size, stride, padding, dilation, scale=scale)
+        self.Conv1dReluBn2 = Conv1dReluBn(out_channels, out_channels, kernel_size=1, stride=1, padding=0)
+        self.SE_Connect = SE_Connect(out_channels, se_bottleneck_dim)
+
+        self.shortcut = None
+        if in_channels != out_channels:
+            self.shortcut = nn.Conv1d(
+                in_channels=in_channels,
+                out_channels=out_channels,
+                kernel_size=1,
+            )
+
+    def forward(self, x):
+        residual = x
+        if self.shortcut:
+            residual = self.shortcut(x)
+
+        x = self.Conv1dReluBn1(x)
+        x = self.Res2Conv1dReluBn(x)
+        x = self.Conv1dReluBn2(x)
+        x = self.SE_Connect(x)
+
+        return x + residual
+
+
+''' Attentive weighted mean and standard deviation pooling.
+'''
+
+class AttentiveStatsPool(nn.Module):
+    def __init__(self, in_dim, attention_channels=128, global_context_att=False):
+        super().__init__()
+        self.global_context_att = global_context_att
+
+        # Use Conv1d with stride == 1 rather than Linear, then we don't need to transpose inputs.
+        if global_context_att:
+            self.linear1 = nn.Conv1d(in_dim * 3, attention_channels, kernel_size=1)  # equals W and b in the paper
+        else:
+            self.linear1 = nn.Conv1d(in_dim, attention_channels, kernel_size=1)  # equals W and b in the paper
+        self.linear2 = nn.Conv1d(attention_channels, in_dim, kernel_size=1)  # equals V and k in the paper
+
+    def forward(self, x):
+
+        if self.global_context_att:
+            context_mean = torch.mean(x, dim=-1, keepdim=True).expand_as(x)
+            context_std = torch.sqrt(torch.var(x, dim=-1, keepdim=True) + 1e-10).expand_as(x)
+            x_in = torch.cat((x, context_mean, context_std), dim=1)
+        else:
+            x_in = x
+
+        # DON'T use ReLU here! In experiments, I find ReLU hard to converge.
+        alpha = torch.tanh(self.linear1(x_in))
+        # alpha = F.relu(self.linear1(x_in))
+        alpha = torch.softmax(self.linear2(alpha), dim=2)
+        mean = torch.sum(alpha * x, dim=2)
+        residuals = torch.sum(alpha * (x ** 2), dim=2) - mean ** 2
+        std = torch.sqrt(residuals.clamp(min=1e-9))
+        return torch.cat([mean, std], dim=1)
+
+
+class ECAPA_TDNN(nn.Module):
+    def __init__(self, feat_dim=80, channels=512, emb_dim=192, global_context_att=False,
+                 feat_type='wavlm_large', sr=16000, feature_selection="hidden_states", update_extract=False, config_path=None):
+        super().__init__()
+
+        self.feat_type = feat_type
+        self.feature_selection = feature_selection
+        self.update_extract = update_extract
+        self.sr = sr
+        
+        torch.hub._validate_not_a_forked_repo=lambda a,b,c: True
+        try:
+            local_s3prl_path = os.path.expanduser("~/.cache/torch/hub/s3prl_s3prl_main")
+            self.feature_extract = torch.hub.load(local_s3prl_path, feat_type, source='local', config_path=config_path)
+        except:
+            self.feature_extract = torch.hub.load('s3prl/s3prl', feat_type)
+
+        if len(self.feature_extract.model.encoder.layers) == 24 and hasattr(self.feature_extract.model.encoder.layers[23].self_attn, "fp32_attention"):
+            self.feature_extract.model.encoder.layers[23].self_attn.fp32_attention = False
+        if len(self.feature_extract.model.encoder.layers) == 24 and hasattr(self.feature_extract.model.encoder.layers[11].self_attn, "fp32_attention"):
+            self.feature_extract.model.encoder.layers[11].self_attn.fp32_attention = False
+
+        self.feat_num = self.get_feat_num()
+        self.feature_weight = nn.Parameter(torch.zeros(self.feat_num))
+
+        if feat_type != 'fbank' and feat_type != 'mfcc':
+            freeze_list = ['final_proj', 'label_embs_concat', 'mask_emb', 'project_q', 'quantizer']
+            for name, param in self.feature_extract.named_parameters():
+                for freeze_val in freeze_list:
+                    if freeze_val in name:
+                        param.requires_grad = False
+                        break
+
+        if not self.update_extract:
+            for param in self.feature_extract.parameters():
+                param.requires_grad = False
+
+        self.instance_norm = nn.InstanceNorm1d(feat_dim)
+        # self.channels = [channels] * 4 + [channels * 3]
+        self.channels = [channels] * 4 + [1536]
+
+        self.layer1 = Conv1dReluBn(feat_dim, self.channels[0], kernel_size=5, padding=2)
+        self.layer2 = SE_Res2Block(self.channels[0], self.channels[1], kernel_size=3, stride=1, padding=2, dilation=2, scale=8, se_bottleneck_dim=128)
+        self.layer3 = SE_Res2Block(self.channels[1], self.channels[2], kernel_size=3, stride=1, padding=3, dilation=3, scale=8, se_bottleneck_dim=128)
+        self.layer4 = SE_Res2Block(self.channels[2], self.channels[3], kernel_size=3, stride=1, padding=4, dilation=4, scale=8, se_bottleneck_dim=128)
+
+        # self.conv = nn.Conv1d(self.channels[-1], self.channels[-1], kernel_size=1)
+        cat_channels = channels * 3
+        self.conv = nn.Conv1d(cat_channels, self.channels[-1], kernel_size=1)
+        self.pooling = AttentiveStatsPool(self.channels[-1], attention_channels=128, global_context_att=global_context_att)
+        self.bn = nn.BatchNorm1d(self.channels[-1] * 2)
+        self.linear = nn.Linear(self.channels[-1] * 2, emb_dim)
+
+
+    def get_feat_num(self):
+        self.feature_extract.eval()
+        wav = [torch.randn(self.sr).to(next(self.feature_extract.parameters()).device)]
+        with torch.no_grad():
+            features = self.feature_extract(wav)
+        select_feature = features[self.feature_selection]
+        if isinstance(select_feature, (list, tuple)):
+            return len(select_feature)
+        else:
+            return 1
+
+    def get_feat(self, x):
+        if self.update_extract:
+            x = self.feature_extract([sample for sample in x])
+        else:
+            with torch.no_grad():
+                if self.feat_type == 'fbank' or self.feat_type == 'mfcc':
+                    x = self.feature_extract(x) + 1e-6  # B x feat_dim x time_len
+                else:
+                    x = self.feature_extract([sample for sample in x])
+
+        if self.feat_type == 'fbank':
+            x = x.log()
+
+        if self.feat_type != "fbank" and self.feat_type != "mfcc":
+            x = x[self.feature_selection]
+            if isinstance(x, (list, tuple)):
+                x = torch.stack(x, dim=0)
+            else:
+                x = x.unsqueeze(0)
+            norm_weights = F.softmax(self.feature_weight, dim=-1).unsqueeze(-1).unsqueeze(-1).unsqueeze(-1)
+            x = (norm_weights * x).sum(dim=0)
+            x = torch.transpose(x, 1, 2) + 1e-6
+
+        x = self.instance_norm(x)
+        return x
+
+    def forward(self, x):
+        x = self.get_feat(x)
+
+        out1 = self.layer1(x)
+        out2 = self.layer2(out1)
+        out3 = self.layer3(out2)
+        out4 = self.layer4(out3)
+
+        out = torch.cat([out2, out3, out4], dim=1)
+        out = F.relu(self.conv(out))
+        out = self.bn(self.pooling(out))
+        out = self.linear(out)
+
+        return out
+
+
+def ECAPA_TDNN_SMALL(feat_dim, emb_dim=256, feat_type='wavlm_large', sr=16000, feature_selection="hidden_states", update_extract=False, config_path=None):
+    return ECAPA_TDNN(feat_dim=feat_dim, channels=512, emb_dim=emb_dim,
+                      feat_type=feat_type, sr=sr, feature_selection=feature_selection, update_extract=update_extract, config_path=config_path)

tts/F5-TTS/model/modules.py

@@ -0,0 +1,575 @@
+"""
+ein notation:
+b - batch
+n - sequence
+nt - text sequence
+nw - raw wave length
+d - dimension
+"""
+
+from __future__ import annotations
+from typing import Optional
+import math
+
+import torch
+from torch import nn
+import torch.nn.functional as F
+import torchaudio
+
+from einops import rearrange
+from x_transformers.x_transformers import apply_rotary_pos_emb
+
+
+# raw wav to mel spec
+
+class MelSpec(nn.Module):
+    def __init__(
+        self,
+        filter_length = 1024,
+        hop_length = 256,
+        win_length = 1024,
+        n_mel_channels = 100,
+        target_sample_rate = 24_000,
+        normalize = False,
+        power = 1,
+        norm = None,
+        center = True,
+    ):
+        super().__init__()
+        self.n_mel_channels = n_mel_channels
+
+        self.mel_stft = torchaudio.transforms.MelSpectrogram(
+            sample_rate = target_sample_rate,
+            n_fft = filter_length,
+            win_length = win_length,
+            hop_length = hop_length,
+            n_mels = n_mel_channels,
+            power = power,
+            center = center,
+            normalized = normalize,
+            norm = norm,
+        )
+
+        self.register_buffer('dummy', torch.tensor(0), persistent = False)
+
+    def forward(self, inp):
+        if len(inp.shape) == 3:
+            inp = rearrange(inp, 'b 1 nw -> b nw')
+
+        assert len(inp.shape) == 2
+
+        if self.dummy.device != inp.device:
+            self.to(inp.device)
+
+        mel = self.mel_stft(inp)
+        mel = mel.clamp(min = 1e-5).log()
+        return mel
+    
+
+# sinusoidal position embedding
+
+class SinusPositionEmbedding(nn.Module):
+    def __init__(self, dim):
+        super().__init__()
+        self.dim = dim
+
+    def forward(self, x, scale=1000):
+        device = x.device
+        half_dim = self.dim // 2
+        emb = math.log(10000) / (half_dim - 1)
+        emb = torch.exp(torch.arange(half_dim, device=device).float() * -emb)
+        emb = scale * x.unsqueeze(1) * emb.unsqueeze(0)
+        emb = torch.cat((emb.sin(), emb.cos()), dim=-1)
+        return emb
+
+
+# convolutional position embedding
+
+class ConvPositionEmbedding(nn.Module):
+    def __init__(self, dim, kernel_size = 31, groups = 16):
+        super().__init__()
+        assert kernel_size % 2 != 0
+        self.conv1d = nn.Sequential(
+            nn.Conv1d(dim, dim, kernel_size, groups = groups, padding = kernel_size // 2),
+            nn.Mish(),
+            nn.Conv1d(dim, dim, kernel_size, groups = groups, padding = kernel_size // 2),
+            nn.Mish(),
+        )
+
+    def forward(self, x: float['b n d'], mask: bool['b n'] | None  = None):
+        if mask is not None:
+            mask = mask[..., None]
+            x = x.masked_fill(~mask, 0.)
+
+        x = rearrange(x, 'b n d -> b d n')
+        x = self.conv1d(x)
+        out = rearrange(x, 'b d n -> b n d')
+
+        if mask is not None:
+            out = out.masked_fill(~mask, 0.)
+
+        return out
+
+
+# rotary positional embedding related
+
+def precompute_freqs_cis(dim: int, end: int, theta: float = 10000.0, theta_rescale_factor=1.):
+    # proposed by reddit user bloc97, to rescale rotary embeddings to longer sequence length without fine-tuning
+    # has some connection to NTK literature
+    # https://www.reddit.com/r/LocalLLaMA/comments/14lz7j5/ntkaware_scaled_rope_allows_llama_models_to_have/
+    # https://github.com/lucidrains/rotary-embedding-torch/blob/main/rotary_embedding_torch/rotary_embedding_torch.py
+    theta *= theta_rescale_factor ** (dim / (dim - 2))
+    freqs = 1.0 / (theta ** (torch.arange(0, dim, 2)[: (dim // 2)].float() / dim))
+    t = torch.arange(end, device=freqs.device)  # type: ignore
+    freqs = torch.outer(t, freqs).float()  # type: ignore
+    freqs_cos = torch.cos(freqs)  # real part
+    freqs_sin = torch.sin(freqs)  # imaginary part
+    return torch.cat([freqs_cos, freqs_sin], dim=-1)
+
+def get_pos_embed_indices(start, length, max_pos, scale=1.):
+    # length = length if isinstance(length, int) else length.max()
+    scale = scale * torch.ones_like(start, dtype=torch.float32)  # in case scale is a scalar
+    pos = start.unsqueeze(1) + (
+            torch.arange(length, device=start.device, dtype=torch.float32).unsqueeze(0) *
+            scale.unsqueeze(1)).long()
+    # avoid extra long error.
+    pos = torch.where(pos < max_pos, pos, max_pos - 1)
+    return pos
+
+
+# Global Response Normalization layer (Instance Normalization ?)
+
+class GRN(nn.Module):
+    def __init__(self, dim):
+        super().__init__()
+        self.gamma = nn.Parameter(torch.zeros(1, 1, dim))
+        self.beta = nn.Parameter(torch.zeros(1, 1, dim))
+
+    def forward(self, x):
+        Gx = torch.norm(x, p=2, dim=1, keepdim=True)
+        Nx = Gx / (Gx.mean(dim=-1, keepdim=True) + 1e-6)
+        return self.gamma * (x * Nx) + self.beta + x
+
+
+# ConvNeXt-V2 Block https://github.com/facebookresearch/ConvNeXt-V2/blob/main/models/convnextv2.py
+# ref: https://github.com/bfs18/e2_tts/blob/main/rfwave/modules.py#L108
+
+class ConvNeXtV2Block(nn.Module):
+    def __init__(
+        self,
+        dim: int,
+        intermediate_dim: int,
+        dilation: int = 1,
+    ):
+        super().__init__()
+        padding = (dilation * (7 - 1)) // 2
+        self.dwconv = nn.Conv1d(dim, dim, kernel_size=7, padding=padding, groups=dim, dilation=dilation)  # depthwise conv
+        self.norm = nn.LayerNorm(dim, eps=1e-6)
+        self.pwconv1 = nn.Linear(dim, intermediate_dim)  # pointwise/1x1 convs, implemented with linear layers
+        self.act = nn.GELU()
+        self.grn = GRN(intermediate_dim)
+        self.pwconv2 = nn.Linear(intermediate_dim, dim)
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        residual = x
+        x = x.transpose(1, 2)  # b n d -> b d n
+        x = self.dwconv(x)
+        x = x.transpose(1, 2)  # b d n -> b n d
+        x = self.norm(x)
+        x = self.pwconv1(x)
+        x = self.act(x)
+        x = self.grn(x)
+        x = self.pwconv2(x)
+        return residual + x
+
+
+# AdaLayerNormZero
+# return with modulated x for attn input, and params for later mlp modulation
+
+class AdaLayerNormZero(nn.Module):
+    def __init__(self, dim):
+        super().__init__()
+
+        self.silu = nn.SiLU()
+        self.linear = nn.Linear(dim, dim * 6)
+
+        self.norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)
+
+    def forward(self, x, emb = None):
+        emb = self.linear(self.silu(emb))
+        shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = torch.chunk(emb, 6, dim=1)
+
+        x = self.norm(x) * (1 + scale_msa[:, None]) + shift_msa[:, None]
+        return x, gate_msa, shift_mlp, scale_mlp, gate_mlp
+
+
+# AdaLayerNormZero for final layer
+# return only with modulated x for attn input, cuz no more mlp modulation
+
+class AdaLayerNormZero_Final(nn.Module):
+    def __init__(self, dim):
+        super().__init__()
+
+        self.silu = nn.SiLU()
+        self.linear = nn.Linear(dim, dim * 2)
+
+        self.norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)
+
+    def forward(self, x, emb):
+        emb = self.linear(self.silu(emb))
+        scale, shift = torch.chunk(emb, 2, dim=1)
+
+        x = self.norm(x) * (1 + scale)[:, None, :] + shift[:, None, :]
+        return x
+
+
+# FeedForward
+
+class FeedForward(nn.Module):
+    def __init__(self, dim, dim_out = None, mult = 4, dropout = 0., approximate: str = 'none'):
+        super().__init__()
+        inner_dim = int(dim * mult)
+        dim_out = dim_out if dim_out is not None else dim
+
+        activation = nn.GELU(approximate=approximate)
+        project_in = nn.Sequential(
+            nn.Linear(dim, inner_dim),
+            activation
+        )
+        self.ff = nn.Sequential(
+            project_in,
+            nn.Dropout(dropout),
+            nn.Linear(inner_dim, dim_out)
+        )
+
+    def forward(self, x):
+        return self.ff(x)
+
+
+# Attention with possible joint part
+# modified from diffusers/src/diffusers/models/attention_processor.py
+
+class Attention(nn.Module):
+    def __init__(
+        self,
+        processor: JointAttnProcessor | AttnProcessor,
+        dim: int,
+        heads: int = 8,
+        dim_head: int = 64,
+        dropout: float = 0.0,
+        context_dim: Optional[int] = None, # if not None -> joint attention
+        context_pre_only = None,
+    ):
+        super().__init__()
+
+        if not hasattr(F, "scaled_dot_product_attention"):
+            raise ImportError("Attention equires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.")
+
+        self.processor = processor
+
+        self.dim = dim
+        self.heads = heads
+        self.inner_dim = dim_head * heads
+        self.dropout = dropout
+
+        self.context_dim = context_dim
+        self.context_pre_only = context_pre_only
+
+        self.to_q = nn.Linear(dim, self.inner_dim)
+        self.to_k = nn.Linear(dim, self.inner_dim)
+        self.to_v = nn.Linear(dim, self.inner_dim)
+
+        if self.context_dim is not None:
+            self.to_k_c = nn.Linear(context_dim, self.inner_dim)
+            self.to_v_c = nn.Linear(context_dim, self.inner_dim)
+            if self.context_pre_only is not None:
+                self.to_q_c = nn.Linear(context_dim, self.inner_dim)
+
+        self.to_out = nn.ModuleList([])
+        self.to_out.append(nn.Linear(self.inner_dim, dim))
+        self.to_out.append(nn.Dropout(dropout))
+
+        if self.context_pre_only is not None and not self.context_pre_only:
+            self.to_out_c = nn.Linear(self.inner_dim, dim)
+
+    def forward(
+        self,
+        x: float['b n d'], # noised input x
+        c: float['b n d'] = None,  # context c
+        mask: bool['b n'] | None = None,
+        rope = None,  # rotary position embedding for x
+        c_rope = None,  # rotary position embedding for c
+    ) -> torch.Tensor:
+        if c is not None:
+            return self.processor(self, x, c = c, mask = mask, rope = rope, c_rope = c_rope)
+        else:
+            return self.processor(self, x, mask = mask, rope = rope)
+
+
+# Attention processor
+
+class AttnProcessor:
+    def __init__(self):
+        pass
+
+    def __call__(
+        self,
+        attn: Attention,
+        x: float['b n d'], # noised input x
+        mask: bool['b n'] | None = None,
+        rope = None,  # rotary position embedding
+    ) -> torch.FloatTensor:
+
+        batch_size = x.shape[0]
+
+        # `sample` projections.
+        query = attn.to_q(x)
+        key = attn.to_k(x)
+        value = attn.to_v(x)
+
+        # apply rotary position embedding
+        if rope is not None:
+            freqs, xpos_scale = rope
+            q_xpos_scale, k_xpos_scale = (xpos_scale, xpos_scale ** -1.) if xpos_scale is not None else (1., 1.)
+
+            query = apply_rotary_pos_emb(query, freqs, q_xpos_scale)
+            key = apply_rotary_pos_emb(key, freqs, k_xpos_scale)
+
+        # attention
+        inner_dim = key.shape[-1]
+        head_dim = inner_dim // attn.heads
+        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+        # mask. e.g. inference got a batch with different target durations, mask out the padding
+        if mask is not None:
+            attn_mask = mask
+            attn_mask = rearrange(attn_mask, 'b n -> b 1 1 n')
+            attn_mask = attn_mask.expand(batch_size, attn.heads, query.shape[-2], key.shape[-2])
+        else:
+            attn_mask = None
+
+        x = F.scaled_dot_product_attention(query, key, value, attn_mask=attn_mask, dropout_p=0.0, is_causal=False)
+        x = x.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
+        x = x.to(query.dtype)
+
+        # linear proj
+        x = attn.to_out[0](x)
+        # dropout
+        x = attn.to_out[1](x)
+
+        if mask is not None:
+            mask = rearrange(mask, 'b n -> b n 1')
+            x = x.masked_fill(~mask, 0.)
+
+        return x
+    
+
+# Joint Attention processor for MM-DiT
+# modified from diffusers/src/diffusers/models/attention_processor.py
+
+class JointAttnProcessor:
+    def __init__(self):
+        pass
+
+    def __call__(
+        self,
+        attn: Attention,
+        x: float['b n d'], # noised input x
+        c: float['b nt d'] = None,  # context c, here text
+        mask: bool['b n'] | None = None,
+        rope = None,  # rotary position embedding for x
+        c_rope = None,  # rotary position embedding for c
+    ) -> torch.FloatTensor:
+        residual = x
+
+        batch_size = c.shape[0]
+
+        # `sample` projections.
+        query = attn.to_q(x)
+        key = attn.to_k(x)
+        value = attn.to_v(x)
+
+        # `context` projections.
+        c_query = attn.to_q_c(c)
+        c_key = attn.to_k_c(c)
+        c_value = attn.to_v_c(c)
+
+        # apply rope for context and noised input independently
+        if rope is not None:
+            freqs, xpos_scale = rope
+            q_xpos_scale, k_xpos_scale = (xpos_scale, xpos_scale ** -1.) if xpos_scale is not None else (1., 1.)
+            query = apply_rotary_pos_emb(query, freqs, q_xpos_scale)
+            key = apply_rotary_pos_emb(key, freqs, k_xpos_scale)
+        if c_rope is not None:
+            freqs, xpos_scale = c_rope
+            q_xpos_scale, k_xpos_scale = (xpos_scale, xpos_scale ** -1.) if xpos_scale is not None else (1., 1.)
+            c_query = apply_rotary_pos_emb(c_query, freqs, q_xpos_scale)
+            c_key = apply_rotary_pos_emb(c_key, freqs, k_xpos_scale)
+
+        # attention
+        query = torch.cat([query, c_query], dim=1)
+        key = torch.cat([key, c_key], dim=1)
+        value = torch.cat([value, c_value], dim=1)
+
+        inner_dim = key.shape[-1]
+        head_dim = inner_dim // attn.heads
+        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+        # mask. e.g. inference got a batch with different target durations, mask out the padding
+        if mask is not None:
+            attn_mask = F.pad(mask, (0, c.shape[1]), value = True)  # no mask for c (text)
+            attn_mask = rearrange(attn_mask, 'b n -> b 1 1 n')
+            attn_mask = attn_mask.expand(batch_size, attn.heads, query.shape[-2], key.shape[-2])
+        else:
+            attn_mask = None
+
+        x = F.scaled_dot_product_attention(query, key, value, attn_mask=attn_mask, dropout_p=0.0, is_causal=False)
+        x = x.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
+        x = x.to(query.dtype)
+
+        # Split the attention outputs.
+        x, c = (
+            x[:, :residual.shape[1]],
+            x[:, residual.shape[1]:],
+        )
+
+        # linear proj
+        x = attn.to_out[0](x)
+        # dropout
+        x = attn.to_out[1](x)
+        if not attn.context_pre_only:
+            c = attn.to_out_c(c)
+
+        if mask is not None:
+            mask = rearrange(mask, 'b n -> b n 1')
+            x = x.masked_fill(~mask, 0.)
+            # c = c.masked_fill(~mask, 0.)  # no mask for c (text)
+
+        return x, c
+
+
+# DiT Block
+
+class DiTBlock(nn.Module):
+
+    def __init__(self, dim, heads, dim_head, ff_mult = 4, dropout = 0.1):
+        super().__init__()
+        
+        self.attn_norm = AdaLayerNormZero(dim)
+        self.attn = Attention(
+            processor = AttnProcessor(),
+            dim = dim,
+            heads = heads,
+            dim_head = dim_head,
+            dropout = dropout,
+            )
+        
+        self.ff_norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)
+        self.ff = FeedForward(dim = dim, mult = ff_mult, dropout = dropout, approximate = "tanh")
+
+    def forward(self, x, t, mask = None, rope = None): # x: noised input, t: time embedding
+        # pre-norm & modulation for attention input
+        norm, gate_msa, shift_mlp, scale_mlp, gate_mlp = self.attn_norm(x, emb=t)
+
+        # attention
+        attn_output = self.attn(x=norm, mask=mask, rope=rope)
+
+        # process attention output for input x
+        x = x + gate_msa.unsqueeze(1) * attn_output
+        
+        norm = self.ff_norm(x) * (1 + scale_mlp[:, None]) + shift_mlp[:, None]
+        ff_output = self.ff(norm)
+        x = x + gate_mlp.unsqueeze(1) * ff_output
+
+        return x
+
+
+# MMDiT Block https://arxiv.org/abs/2403.03206
+
+class MMDiTBlock(nn.Module):
+    r""" 
+    modified from diffusers/src/diffusers/models/attention.py
+
+    notes.
+    _c: context related. text, cond, etc. (left part in sd3 fig2.b)
+    _x: noised input related. (right part)
+    context_pre_only: last layer only do prenorm + modulation cuz no more ffn
+    """
+
+    def __init__(self, dim, heads, dim_head, ff_mult = 4, dropout = 0.1, context_pre_only = False):
+        super().__init__()
+
+        self.context_pre_only = context_pre_only
+        
+        self.attn_norm_c = AdaLayerNormZero_Final(dim) if context_pre_only else AdaLayerNormZero(dim)
+        self.attn_norm_x = AdaLayerNormZero(dim)
+        self.attn = Attention(
+            processor = JointAttnProcessor(),
+            dim = dim,
+            heads = heads,
+            dim_head = dim_head,
+            dropout = dropout,
+            context_dim = dim,
+            context_pre_only = context_pre_only,
+            )
+
+        if not context_pre_only:
+            self.ff_norm_c = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)
+            self.ff_c = FeedForward(dim = dim, mult = ff_mult, dropout = dropout, approximate = "tanh")
+        else:
+            self.ff_norm_c = None
+            self.ff_c = None
+        self.ff_norm_x = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)
+        self.ff_x = FeedForward(dim = dim, mult = ff_mult, dropout = dropout, approximate = "tanh")
+
+    def forward(self, x, c, t, mask = None, rope = None, c_rope = None): # x: noised input, c: context, t: time embedding
+        # pre-norm & modulation for attention input
+        if self.context_pre_only:
+            norm_c = self.attn_norm_c(c, t)
+        else:
+            norm_c, c_gate_msa, c_shift_mlp, c_scale_mlp, c_gate_mlp = self.attn_norm_c(c, emb=t)
+        norm_x, x_gate_msa, x_shift_mlp, x_scale_mlp, x_gate_mlp = self.attn_norm_x(x, emb=t)
+
+        # attention
+        x_attn_output, c_attn_output = self.attn(x=norm_x, c=norm_c, mask=mask, rope=rope, c_rope=c_rope)
+
+        # process attention output for context c
+        if self.context_pre_only:
+            c = None
+        else: # if not last layer
+            c = c + c_gate_msa.unsqueeze(1) * c_attn_output
+
+            norm_c = self.ff_norm_c(c) * (1 + c_scale_mlp[:, None]) + c_shift_mlp[:, None]
+            c_ff_output = self.ff_c(norm_c)
+            c = c + c_gate_mlp.unsqueeze(1) * c_ff_output
+
+        # process attention output for input x
+        x = x + x_gate_msa.unsqueeze(1) * x_attn_output
+        
+        norm_x = self.ff_norm_x(x) * (1 + x_scale_mlp[:, None]) + x_shift_mlp[:, None]
+        x_ff_output = self.ff_x(norm_x)
+        x = x + x_gate_mlp.unsqueeze(1) * x_ff_output
+
+        return c, x
+
+
+# time step conditioning embedding
+
+class TimestepEmbedding(nn.Module):
+    def __init__(self, dim, freq_embed_dim=256):
+        super().__init__()
+        self.time_embed = SinusPositionEmbedding(freq_embed_dim)
+        self.time_mlp = nn.Sequential(
+            nn.Linear(freq_embed_dim, dim),
+            nn.SiLU(),
+            nn.Linear(dim, dim)
+        )
+
+    def forward(self, timestep: float['b']):
+        time_hidden = self.time_embed(timestep)
+        time = self.time_mlp(time_hidden)  # b d
+        return time

tts/F5-TTS/model/trainer.py

@@ -0,0 +1,250 @@
+from __future__ import annotations
+
+import os
+import gc
+from tqdm import tqdm
+import wandb
+
+import torch
+from torch.optim import AdamW
+from torch.utils.data import DataLoader, Dataset, SequentialSampler
+from torch.optim.lr_scheduler import LinearLR, SequentialLR
+
+from einops import rearrange
+
+from accelerate import Accelerator
+from accelerate.utils import DistributedDataParallelKwargs
+
+from ema_pytorch import EMA
+
+from model import CFM
+from model.utils import exists, default
+from model.dataset import DynamicBatchSampler, collate_fn
+
+
+# trainer
+
+class Trainer:
+    def __init__(
+        self,
+        model: CFM,
+        epochs,
+        learning_rate,
+        num_warmup_updates = 20000,
+        save_per_updates = 1000, 
+        checkpoint_path = None,
+        batch_size = 32, 
+        batch_size_type: str = "sample",
+        max_samples = 32,
+        grad_accumulation_steps = 1,
+        max_grad_norm = 1.0,
+        noise_scheduler: str | None = None,
+        duration_predictor: torch.nn.Module | None = None,
+        wandb_project = "test_e2-tts",
+        wandb_run_name = "test_run",
+        wandb_resume_id: str = None,
+        last_per_steps = None,
+        accelerate_kwargs: dict = dict(),
+        ema_kwargs: dict = dict()
+    ):
+        
+        ddp_kwargs = DistributedDataParallelKwargs(find_unused_parameters = True)
+
+        self.accelerator = Accelerator(
+            log_with = "wandb",
+            kwargs_handlers = [ddp_kwargs],
+            gradient_accumulation_steps = grad_accumulation_steps,
+            **accelerate_kwargs
+        )
+        
+        if exists(wandb_resume_id):
+            init_kwargs={"wandb": {"resume": "allow", "name": wandb_run_name, 'id': wandb_resume_id}}
+        else:
+            init_kwargs={"wandb": {"resume": "allow", "name": wandb_run_name}}
+        self.accelerator.init_trackers(
+            project_name = wandb_project, 
+            init_kwargs=init_kwargs,
+            config={"epochs": epochs,
+                    "learning_rate": learning_rate,
+                    "num_warmup_updates": num_warmup_updates, 
+                    "batch_size": batch_size,
+                    "batch_size_type": batch_size_type,
+                    "max_samples": max_samples,
+                    "grad_accumulation_steps": grad_accumulation_steps,
+                    "max_grad_norm": max_grad_norm,
+                    "gpus": self.accelerator.num_processes,
+                    "noise_scheduler": noise_scheduler}
+            )
+
+        self.model = model
+
+        if self.is_main:
+            self.ema_model = EMA(
+                model,
+                include_online_model = False,
+                **ema_kwargs
+            )
+
+            self.ema_model.to(self.accelerator.device)
+
+        self.epochs = epochs
+        self.num_warmup_updates = num_warmup_updates
+        self.save_per_updates = save_per_updates
+        self.last_per_steps = default(last_per_steps, save_per_updates * grad_accumulation_steps)
+        self.checkpoint_path = default(checkpoint_path, 'ckpts/test_e2-tts')
+
+        self.batch_size = batch_size
+        self.batch_size_type = batch_size_type
+        self.max_samples = max_samples
+        self.grad_accumulation_steps = grad_accumulation_steps
+        self.max_grad_norm = max_grad_norm
+
+        self.noise_scheduler = noise_scheduler
+
+        self.duration_predictor = duration_predictor
+
+        self.optimizer = AdamW(model.parameters(), lr=learning_rate)
+        self.model, self.optimizer = self.accelerator.prepare(
+            self.model, self.optimizer
+        )
+
+    @property
+    def is_main(self):
+        return self.accelerator.is_main_process
+
+    def save_checkpoint(self, step, last=False):
+        self.accelerator.wait_for_everyone()
+        if self.is_main:
+            checkpoint = dict(
+                model_state_dict = self.accelerator.unwrap_model(self.model).state_dict(),
+                optimizer_state_dict = self.accelerator.unwrap_model(self.optimizer).state_dict(),
+                ema_model_state_dict = self.ema_model.state_dict(),
+                scheduler_state_dict = self.scheduler.state_dict(),
+                step = step
+            )
+            if not os.path.exists(self.checkpoint_path):
+                os.makedirs(self.checkpoint_path)
+            if last == True:
+                self.accelerator.save(checkpoint, f"{self.checkpoint_path}/model_last.pt")
+                print(f"Saved last checkpoint at step {step}")
+            else:
+                self.accelerator.save(checkpoint, f"{self.checkpoint_path}/model_{step}.pt")
+
+    def load_checkpoint(self):
+        if not exists(self.checkpoint_path) or not os.path.exists(self.checkpoint_path) or not os.listdir(self.checkpoint_path):
+            return 0
+        
+        self.accelerator.wait_for_everyone()
+        if "model_last.pt" in os.listdir(self.checkpoint_path):
+            latest_checkpoint = "model_last.pt"
+        else:
+            latest_checkpoint = sorted([f for f in os.listdir(self.checkpoint_path) if f.endswith('.pt')], key=lambda x: int(''.join(filter(str.isdigit, x))))[-1]
+        # checkpoint = torch.load(f"{self.checkpoint_path}/{latest_checkpoint}", map_location=self.accelerator.device)  # rather use accelerator.load_state ಥ_ಥ
+        checkpoint = torch.load(f"{self.checkpoint_path}/{latest_checkpoint}", weights_only=True, map_location="cpu")
+
+        if self.is_main:
+            self.ema_model.load_state_dict(checkpoint['ema_model_state_dict'])
+
+        if 'step' in checkpoint:
+            self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint['model_state_dict'])
+            self.accelerator.unwrap_model(self.optimizer).load_state_dict(checkpoint['optimizer_state_dict'])
+            if self.scheduler:
+                self.scheduler.load_state_dict(checkpoint['scheduler_state_dict'])
+            step = checkpoint['step']
+        else:
+            checkpoint['model_state_dict'] = {k.replace("ema_model.", ""): v for k, v in checkpoint['ema_model_state_dict'].items() if k not in ["initted", "step"]}
+            self.accelerator.unwrap_model(self.model).load_state_dict(checkpoint['model_state_dict'])
+            step = 0
+
+        del checkpoint; gc.collect()
+        return step
+
+    def train(self, train_dataset: Dataset, num_workers=16, resumable_with_seed: int = None):
+        
+        if exists(resumable_with_seed):
+            generator = torch.Generator()
+            generator.manual_seed(resumable_with_seed)
+        else: 
+            generator = None
+
+        if self.batch_size_type == "sample":
+            train_dataloader = DataLoader(train_dataset, collate_fn=collate_fn, num_workers=num_workers, pin_memory=True, persistent_workers=True,
+                                          batch_size=self.batch_size, shuffle=True, generator=generator)
+        elif self.batch_size_type == "frame":
+            self.accelerator.even_batches = False
+            sampler = SequentialSampler(train_dataset)
+            batch_sampler = DynamicBatchSampler(sampler, self.batch_size, max_samples=self.max_samples, random_seed=resumable_with_seed, drop_last=False)
+            train_dataloader = DataLoader(train_dataset, collate_fn=collate_fn, num_workers=num_workers, pin_memory=True, persistent_workers=True,
+                                          batch_sampler=batch_sampler)
+        else:
+            raise ValueError(f"batch_size_type must be either 'sample' or 'frame', but received {self.batch_size_type}")
+        
+        #  accelerator.prepare() dispatches batches to devices;
+        #  which means the length of dataloader calculated before, should consider the number of devices
+        warmup_steps = self.num_warmup_updates * self.accelerator.num_processes  # consider a fixed warmup steps while using accelerate multi-gpu ddp
+                                                                                 # otherwise by default with split_batches=False, warmup steps change with num_processes
+        total_steps = len(train_dataloader) * self.epochs / self.grad_accumulation_steps
+        decay_steps = total_steps - warmup_steps
+        warmup_scheduler = LinearLR(self.optimizer, start_factor=1e-8, end_factor=1.0, total_iters=warmup_steps)
+        decay_scheduler = LinearLR(self.optimizer, start_factor=1.0, end_factor=1e-8, total_iters=decay_steps)
+        self.scheduler = SequentialLR(self.optimizer, 
+                                      schedulers=[warmup_scheduler, decay_scheduler],
+                                      milestones=[warmup_steps])
+        train_dataloader, self.scheduler = self.accelerator.prepare(train_dataloader, self.scheduler)  # actual steps = 1 gpu steps / gpus
+        start_step = self.load_checkpoint()
+        global_step = start_step
+
+        if exists(resumable_with_seed):
+            orig_epoch_step = len(train_dataloader)
+            skipped_epoch = int(start_step // orig_epoch_step)
+            skipped_batch = start_step % orig_epoch_step
+            skipped_dataloader = self.accelerator.skip_first_batches(train_dataloader, num_batches=skipped_batch)
+        else:
+            skipped_epoch = 0
+
+        for epoch in range(skipped_epoch, self.epochs):
+            self.model.train()
+            if exists(resumable_with_seed) and epoch == skipped_epoch:
+                progress_bar = tqdm(skipped_dataloader, desc=f"Epoch {epoch+1}/{self.epochs}", unit="step", disable=not self.accelerator.is_local_main_process, 
+                                    initial=skipped_batch, total=orig_epoch_step)
+            else:
+                progress_bar = tqdm(train_dataloader, desc=f"Epoch {epoch+1}/{self.epochs}", unit="step", disable=not self.accelerator.is_local_main_process)
+
+            for batch in progress_bar:
+                with self.accelerator.accumulate(self.model):
+                    text_inputs = batch['text']
+                    mel_spec = rearrange(batch['mel'], 'b d n -> b n d')
+                    mel_lengths = batch["mel_lengths"]
+
+                    # TODO. add duration predictor training
+                    if self.duration_predictor is not None and self.accelerator.is_local_main_process:
+                        dur_loss = self.duration_predictor(mel_spec, lens=batch.get('durations'))
+                        self.accelerator.log({"duration loss": dur_loss.item()}, step=global_step)
+
+                    loss, cond, pred = self.model(mel_spec, text=text_inputs, lens=mel_lengths, noise_scheduler=self.noise_scheduler)
+                    self.accelerator.backward(loss)
+
+                    if self.max_grad_norm > 0 and self.accelerator.sync_gradients:
+                        self.accelerator.clip_grad_norm_(self.model.parameters(), self.max_grad_norm)
+
+                    self.optimizer.step()
+                    self.scheduler.step()
+                    self.optimizer.zero_grad()
+
+                if self.is_main:
+                    self.ema_model.update()
+
+                global_step += 1
+
+                if self.accelerator.is_local_main_process:
+                    self.accelerator.log({"loss": loss.item(), "lr": self.scheduler.get_last_lr()[0]}, step=global_step)
+                
+                progress_bar.set_postfix(step=str(global_step), loss=loss.item())
+                
+                if global_step % (self.save_per_updates * self.grad_accumulation_steps) == 0:
+                    self.save_checkpoint(global_step)
+                
+                if global_step % self.last_per_steps == 0:
+                    self.save_checkpoint(global_step, last=True)
+        
+        self.accelerator.end_training()

tts/F5-TTS/model/utils.py

@@ -0,0 +1,580 @@
+from __future__ import annotations
+
+import os
+import re
+import math
+import random
+import string
+from tqdm import tqdm
+from collections import defaultdict
+
+import matplotlib
+matplotlib.use("Agg")
+import matplotlib.pylab as plt
+
+import torch
+import torch.nn.functional as F
+from torch.nn.utils.rnn import pad_sequence
+import torchaudio
+
+import einx
+from einops import rearrange, reduce
+
+import jieba
+from pypinyin import lazy_pinyin, Style
+
+from model.ecapa_tdnn import ECAPA_TDNN_SMALL
+from model.modules import MelSpec
+
+
+# seed everything
+
+def seed_everything(seed = 0):
+    random.seed(seed)
+    os.environ['PYTHONHASHSEED'] = str(seed)
+    torch.manual_seed(seed)
+    torch.cuda.manual_seed(seed)
+    torch.cuda.manual_seed_all(seed)
+    torch.backends.cudnn.deterministic = True
+    torch.backends.cudnn.benchmark = False
+
+# helpers
+
+def exists(v):
+    return v is not None
+
+def default(v, d):
+    return v if exists(v) else d
+
+# tensor helpers
+
+def lens_to_mask(
+    t: int['b'],
+    length: int | None = None
+) -> bool['b n']:
+
+    if not exists(length):
+        length = t.amax()
+
+    seq = torch.arange(length, device = t.device)
+    return einx.less('n, b -> b n', seq, t)
+
+def mask_from_start_end_indices(
+    seq_len: int['b'],
+    start: int['b'],
+    end: int['b']
+):
+    max_seq_len = seq_len.max().item()  
+    seq = torch.arange(max_seq_len, device = start.device).long()
+    return einx.greater_equal('n, b -> b n', seq, start) & einx.less('n, b -> b n', seq, end)
+
+def mask_from_frac_lengths(
+    seq_len: int['b'],
+    frac_lengths: float['b']
+):
+    lengths = (frac_lengths * seq_len).long()
+    max_start = seq_len - lengths
+
+    rand = torch.rand_like(frac_lengths)
+    start = (max_start * rand).long().clamp(min = 0)
+    end = start + lengths
+
+    return mask_from_start_end_indices(seq_len, start, end)
+
+def maybe_masked_mean(
+    t: float['b n d'],
+    mask: bool['b n'] = None
+) -> float['b d']:
+
+    if not exists(mask):
+        return t.mean(dim = 1)
+
+    t = einx.where('b n, b n d, -> b n d', mask, t, 0.)
+    num = reduce(t, 'b n d -> b d', 'sum')
+    den = reduce(mask.float(), 'b n -> b', 'sum')
+
+    return einx.divide('b d, b -> b d', num, den.clamp(min = 1.))
+
+
+# simple utf-8 tokenizer, since paper went character based
+def list_str_to_tensor(
+    text: list[str],
+    padding_value = -1
+) -> int['b nt']:
+    list_tensors = [torch.tensor([*bytes(t, 'UTF-8')]) for t in text]  # ByT5 style
+    text = pad_sequence(list_tensors, padding_value = padding_value, batch_first = True)
+    return text
+
+# char tokenizer, based on custom dataset's extracted .txt file
+def list_str_to_idx(
+    text: list[str] | list[list[str]],
+    vocab_char_map: dict[str, int],  # {char: idx}
+    padding_value = -1
+) -> int['b nt']:
+    list_idx_tensors = [torch.tensor([vocab_char_map.get(c, 0) for c in t]) for t in text]  # pinyin or char style
+    text = pad_sequence(list_idx_tensors, padding_value = padding_value, batch_first = True)
+    return text
+
+
+# Get tokenizer
+
+def get_tokenizer(dataset_name, tokenizer: str = "pinyin"):
+    ''' 
+    tokenizer   - "pinyin" do g2p for only chinese characters, need .txt vocab_file
+                - "char" for char-wise tokenizer, need .txt vocab_file
+                - "byte" for utf-8 tokenizer
+                - "custom" if you're directly passing in a path to the vocab.txt you want to use
+    vocab_size  - if use "pinyin", all available pinyin types, common alphabets (also those with accent) and symbols
+                - if use "char", derived from unfiltered character & symbol counts of custom dataset
+                - if use "byte", set to 256 (unicode byte range) 
+    ''' 
+    if tokenizer in ["pinyin", "char"]:
+        with open (f"data/{dataset_name}_{tokenizer}/vocab.txt", "r", encoding="utf-8") as f:
+            vocab_char_map = {}
+            for i, char in enumerate(f):
+                vocab_char_map[char[:-1]] = i
+        vocab_size = len(vocab_char_map)
+        assert vocab_char_map[" "] == 0, "make sure space is of idx 0 in vocab.txt, cuz 0 is used for unknown char"
+
+    elif tokenizer == "byte":
+        vocab_char_map = None
+        vocab_size = 256
+    elif tokenizer == "custom":
+        with open (dataset_name, "r", encoding="utf-8") as f:
+            vocab_char_map = {}
+            for i, char in enumerate(f):
+                vocab_char_map[char[:-1]] = i
+        vocab_size = len(vocab_char_map)
+
+    return vocab_char_map, vocab_size
+
+
+# convert char to pinyin
+
+def convert_char_to_pinyin(text_list, polyphone = True):
+    final_text_list = []
+    god_knows_why_en_testset_contains_zh_quote = str.maketrans({'“': '"', '”': '"', '‘': "'", '’': "'"})  # in case librispeech (orig no-pc) test-clean
+    custom_trans = str.maketrans({';': ','})  # add custom trans here, to address oov
+    for text in text_list:
+        char_list = []
+        text = text.translate(god_knows_why_en_testset_contains_zh_quote)
+        text = text.translate(custom_trans)
+        for seg in jieba.cut(text):
+            seg_byte_len = len(bytes(seg, 'UTF-8'))
+            if seg_byte_len == len(seg):  # if pure alphabets and symbols
+                if char_list and seg_byte_len > 1 and char_list[-1] not in " :'\"":
+                    char_list.append(" ")
+                char_list.extend(seg)
+            elif polyphone and seg_byte_len == 3 * len(seg):  # if pure chinese characters
+                seg = lazy_pinyin(seg, style=Style.TONE3, tone_sandhi=True)
+                for c in seg:
+                    if c not in "。，、；：？！《》【】—…":
+                        char_list.append(" ")
+                    char_list.append(c)
+            else:  # if mixed chinese characters, alphabets and symbols
+                for c in seg:
+                    if ord(c) < 256:
+                        char_list.extend(c)
+                    else:
+                        if c not in "。，、；：？！《》【】—…":
+                            char_list.append(" ")
+                            char_list.extend(lazy_pinyin(c, style=Style.TONE3, tone_sandhi=True))
+                        else:  # if is zh punc
+                            char_list.append(c)
+        final_text_list.append(char_list)
+
+    return final_text_list
+
+
+# save spectrogram
+def save_spectrogram(spectrogram, path):
+    plt.figure(figsize=(12, 4))
+    plt.imshow(spectrogram, origin='lower', aspect='auto')
+    plt.colorbar()
+    plt.savefig(path)
+    plt.close()
+
+
+# seedtts testset metainfo: utt, prompt_text, prompt_wav, gt_text, gt_wav
+def get_seedtts_testset_metainfo(metalst):
+    f = open(metalst); lines = f.readlines(); f.close()
+    metainfo = []
+    for line in lines:
+        if len(line.strip().split('|')) == 5:
+            utt, prompt_text, prompt_wav, gt_text, gt_wav = line.strip().split('|')
+        elif len(line.strip().split('|')) == 4:
+            utt, prompt_text, prompt_wav, gt_text = line.strip().split('|')
+            gt_wav = os.path.join(os.path.dirname(metalst), "wavs", utt + ".wav")
+        if not os.path.isabs(prompt_wav):
+            prompt_wav = os.path.join(os.path.dirname(metalst), prompt_wav)
+        metainfo.append((utt, prompt_text, prompt_wav, gt_text, gt_wav))
+    return metainfo
+
+
+# librispeech test-clean metainfo: gen_utt, ref_txt, ref_wav, gen_txt, gen_wav
+def get_librispeech_test_clean_metainfo(metalst, librispeech_test_clean_path):
+    f = open(metalst); lines = f.readlines(); f.close()
+    metainfo = []
+    for line in lines:
+        ref_utt, ref_dur, ref_txt, gen_utt, gen_dur, gen_txt = line.strip().split('\t')
+
+        # ref_txt = ref_txt[0] + ref_txt[1:].lower() + '.'  # if use librispeech test-clean (no-pc)
+        ref_spk_id, ref_chaptr_id, _ =  ref_utt.split('-')
+        ref_wav = os.path.join(librispeech_test_clean_path, ref_spk_id, ref_chaptr_id, ref_utt + '.flac')
+
+        # gen_txt = gen_txt[0] + gen_txt[1:].lower() + '.'  # if use librispeech test-clean (no-pc)
+        gen_spk_id, gen_chaptr_id, _ =  gen_utt.split('-')
+        gen_wav = os.path.join(librispeech_test_clean_path, gen_spk_id, gen_chaptr_id, gen_utt + '.flac')
+
+        metainfo.append((gen_utt, ref_txt, ref_wav, " " + gen_txt, gen_wav))
+
+    return metainfo
+
+
+# padded to max length mel batch
+def padded_mel_batch(ref_mels):
+    max_mel_length = torch.LongTensor([mel.shape[-1] for mel in ref_mels]).amax()
+    padded_ref_mels = []
+    for mel in ref_mels:
+        padded_ref_mel = F.pad(mel, (0, max_mel_length - mel.shape[-1]), value = 0)
+        padded_ref_mels.append(padded_ref_mel)
+    padded_ref_mels = torch.stack(padded_ref_mels)
+    padded_ref_mels = rearrange(padded_ref_mels, 'b d n -> b n d')
+    return padded_ref_mels
+
+
+# get prompts from metainfo containing: utt, prompt_text, prompt_wav, gt_text, gt_wav
+
+def get_inference_prompt(
+    metainfo, 
+    speed = 1., tokenizer = "pinyin", polyphone = True, 
+    target_sample_rate = 24000, n_mel_channels = 100, hop_length = 256, target_rms = 0.1,
+    use_truth_duration = False,
+    infer_batch_size = 1, num_buckets = 200, min_secs = 3, max_secs = 40,
+):
+    prompts_all = []
+
+    min_tokens = min_secs * target_sample_rate // hop_length
+    max_tokens = max_secs * target_sample_rate // hop_length
+
+    batch_accum = [0] * num_buckets
+    utts, ref_rms_list, ref_mels, ref_mel_lens, total_mel_lens, final_text_list = \
+        ([[] for _ in range(num_buckets)] for _ in range(6))
+
+    mel_spectrogram = MelSpec(target_sample_rate=target_sample_rate, n_mel_channels=n_mel_channels, hop_length=hop_length)
+
+    for utt, prompt_text, prompt_wav, gt_text, gt_wav in tqdm(metainfo, desc="Processing prompts..."):
+
+        # Audio
+        ref_audio, ref_sr = torchaudio.load(prompt_wav)
+        ref_rms = torch.sqrt(torch.mean(torch.square(ref_audio)))
+        if ref_rms < target_rms:
+            ref_audio = ref_audio * target_rms / ref_rms
+        assert ref_audio.shape[-1] > 5000, f"Empty prompt wav: {prompt_wav}, or torchaudio backend issue."
+        if ref_sr != target_sample_rate:
+            resampler = torchaudio.transforms.Resample(ref_sr, target_sample_rate)
+            ref_audio = resampler(ref_audio)
+
+        # Text
+        if len(prompt_text[-1].encode('utf-8')) == 1:
+            prompt_text = prompt_text + " "
+        text = [prompt_text + gt_text]
+        if tokenizer == "pinyin":
+            text_list = convert_char_to_pinyin(text, polyphone = polyphone)
+        else:
+            text_list = text
+
+        # Duration, mel frame length
+        ref_mel_len = ref_audio.shape[-1] // hop_length
+        if use_truth_duration:
+            gt_audio, gt_sr = torchaudio.load(gt_wav)
+            if gt_sr != target_sample_rate:
+                resampler = torchaudio.transforms.Resample(gt_sr, target_sample_rate)
+                gt_audio = resampler(gt_audio)
+            total_mel_len = ref_mel_len + int(gt_audio.shape[-1] / hop_length / speed)
+
+            # # test vocoder resynthesis
+            # ref_audio = gt_audio
+        else:
+            zh_pause_punc = r"。，、；：？！"
+            ref_text_len = len(prompt_text.encode('utf-8')) + 3 * len(re.findall(zh_pause_punc, prompt_text))
+            gen_text_len = len(gt_text.encode('utf-8')) + 3 * len(re.findall(zh_pause_punc, gt_text))
+            total_mel_len = ref_mel_len + int(ref_mel_len / ref_text_len * gen_text_len / speed)
+
+        # to mel spectrogram
+        ref_mel = mel_spectrogram(ref_audio)
+        ref_mel = rearrange(ref_mel, '1 d n -> d n')
+
+        # deal with batch
+        assert infer_batch_size > 0, "infer_batch_size should be greater than 0."
+        assert min_tokens <= total_mel_len <= max_tokens, \
+            f"Audio {utt} has duration {total_mel_len*hop_length//target_sample_rate}s out of range [{min_secs}, {max_secs}]."
+        bucket_i = math.floor((total_mel_len - min_tokens) / (max_tokens - min_tokens + 1) * num_buckets)
+
+        utts[bucket_i].append(utt)
+        ref_rms_list[bucket_i].append(ref_rms)
+        ref_mels[bucket_i].append(ref_mel)
+        ref_mel_lens[bucket_i].append(ref_mel_len)
+        total_mel_lens[bucket_i].append(total_mel_len)
+        final_text_list[bucket_i].extend(text_list)
+
+        batch_accum[bucket_i] += total_mel_len
+
+        if batch_accum[bucket_i] >= infer_batch_size:
+            # print(f"\n{len(ref_mels[bucket_i][0][0])}\n{ref_mel_lens[bucket_i]}\n{total_mel_lens[bucket_i]}")
+            prompts_all.append((
+                utts[bucket_i], 
+                ref_rms_list[bucket_i], 
+                padded_mel_batch(ref_mels[bucket_i]), 
+                ref_mel_lens[bucket_i], 
+                total_mel_lens[bucket_i], 
+                final_text_list[bucket_i]
+            ))
+            batch_accum[bucket_i] = 0
+            utts[bucket_i], ref_rms_list[bucket_i], ref_mels[bucket_i], ref_mel_lens[bucket_i], total_mel_lens[bucket_i], final_text_list[bucket_i] = [], [], [], [], [], []
+
+    # add residual
+    for bucket_i, bucket_frames in enumerate(batch_accum):
+        if bucket_frames > 0:
+            prompts_all.append((
+                utts[bucket_i], 
+                ref_rms_list[bucket_i], 
+                padded_mel_batch(ref_mels[bucket_i]), 
+                ref_mel_lens[bucket_i], 
+                total_mel_lens[bucket_i], 
+                final_text_list[bucket_i]
+            ))
+    # not only leave easy work for last workers
+    random.seed(666)
+    random.shuffle(prompts_all)
+
+    return prompts_all
+
+
+# get wav_res_ref_text of seed-tts test metalst
+# https://github.com/BytedanceSpeech/seed-tts-eval
+
+def get_seed_tts_test(metalst, gen_wav_dir, gpus):
+    f = open(metalst)
+    lines = f.readlines()
+    f.close()
+
+    test_set_ = []
+    for line in tqdm(lines):
+        if len(line.strip().split('|')) == 5:
+            utt, prompt_text, prompt_wav, gt_text, gt_wav = line.strip().split('|')
+        elif len(line.strip().split('|')) == 4:
+            utt, prompt_text, prompt_wav, gt_text = line.strip().split('|')
+
+        if not os.path.exists(os.path.join(gen_wav_dir, utt + '.wav')):
+            continue
+        gen_wav = os.path.join(gen_wav_dir, utt + '.wav')
+        if not os.path.isabs(prompt_wav):
+            prompt_wav = os.path.join(os.path.dirname(metalst), prompt_wav)
+
+        test_set_.append((gen_wav, prompt_wav, gt_text))
+
+    num_jobs = len(gpus)
+    if num_jobs == 1:
+        return [(gpus[0], test_set_)]
+    
+    wav_per_job = len(test_set_) // num_jobs + 1
+    test_set = []
+    for i in range(num_jobs):
+        test_set.append((gpus[i], test_set_[i*wav_per_job:(i+1)*wav_per_job]))
+
+    return test_set
+
+
+# get librispeech test-clean cross sentence test
+
+def get_librispeech_test(metalst, gen_wav_dir, gpus, librispeech_test_clean_path, eval_ground_truth = False):
+    f = open(metalst)
+    lines = f.readlines()
+    f.close()
+
+    test_set_ = []
+    for line in tqdm(lines):
+        ref_utt, ref_dur, ref_txt, gen_utt, gen_dur, gen_txt = line.strip().split('\t')
+
+        if eval_ground_truth:
+            gen_spk_id, gen_chaptr_id, _ =  gen_utt.split('-')
+            gen_wav = os.path.join(librispeech_test_clean_path, gen_spk_id, gen_chaptr_id, gen_utt + '.flac')
+        else:
+            if not os.path.exists(os.path.join(gen_wav_dir, gen_utt + '.wav')):
+                raise FileNotFoundError(f"Generated wav not found: {gen_utt}")
+            gen_wav = os.path.join(gen_wav_dir, gen_utt + '.wav')
+
+        ref_spk_id, ref_chaptr_id, _ =  ref_utt.split('-')
+        ref_wav = os.path.join(librispeech_test_clean_path, ref_spk_id, ref_chaptr_id, ref_utt + '.flac')
+
+        test_set_.append((gen_wav, ref_wav, gen_txt))
+
+    num_jobs = len(gpus)
+    if num_jobs == 1:
+        return [(gpus[0], test_set_)]
+    
+    wav_per_job = len(test_set_) // num_jobs + 1
+    test_set = []
+    for i in range(num_jobs):
+        test_set.append((gpus[i], test_set_[i*wav_per_job:(i+1)*wav_per_job]))
+
+    return test_set
+
+
+# load asr model
+
+def load_asr_model(lang, ckpt_dir = ""):
+    if lang == "zh":
+        from funasr import AutoModel
+        model = AutoModel(
+            model = os.path.join(ckpt_dir, "paraformer-zh"), 
+            # vad_model = os.path.join(ckpt_dir, "fsmn-vad"), 
+            # punc_model = os.path.join(ckpt_dir, "ct-punc"),
+            # spk_model = os.path.join(ckpt_dir, "cam++"), 
+            disable_update=True,
+            )  # following seed-tts setting
+    elif lang == "en":
+        from faster_whisper import WhisperModel
+        model_size = "large-v3" if ckpt_dir == "" else ckpt_dir
+        model = WhisperModel(model_size, device="cuda", compute_type="float16")
+    return model
+
+
+# WER Evaluation, the way Seed-TTS does
+
+def run_asr_wer(args):
+    rank, lang, test_set, ckpt_dir = args
+
+    if lang == "zh":
+        import zhconv
+        torch.cuda.set_device(rank)
+    elif lang == "en":
+        os.environ["CUDA_VISIBLE_DEVICES"] = str(rank)
+    else:
+        raise NotImplementedError("lang support only 'zh' (funasr paraformer-zh), 'en' (faster-whisper-large-v3), for now.")
+
+    asr_model = load_asr_model(lang, ckpt_dir = ckpt_dir)
+    
+    from zhon.hanzi import punctuation
+    punctuation_all = punctuation + string.punctuation
+    wers = []
+
+    from jiwer import compute_measures
+    for gen_wav, prompt_wav, truth in tqdm(test_set):
+        if lang == "zh":
+            res = asr_model.generate(input=gen_wav, batch_size_s=300, disable_pbar=True)
+            hypo = res[0]["text"]
+            hypo = zhconv.convert(hypo, 'zh-cn')
+        elif lang == "en":
+            segments, _ = asr_model.transcribe(gen_wav, beam_size=5, language="en")
+            hypo = ''
+            for segment in segments:
+                hypo = hypo + ' ' + segment.text
+
+        # raw_truth = truth
+        # raw_hypo = hypo
+
+        for x in punctuation_all:
+            truth = truth.replace(x, '')
+            hypo = hypo.replace(x, '')
+
+        truth = truth.replace('  ', ' ')
+        hypo = hypo.replace('  ', ' ')
+
+        if lang == "zh":
+            truth = " ".join([x for x in truth])
+            hypo = " ".join([x for x in hypo])
+        elif lang == "en":
+            truth = truth.lower()
+            hypo = hypo.lower()
+
+        measures = compute_measures(truth, hypo)
+        wer = measures["wer"]
+
+        # ref_list = truth.split(" ")
+        # subs = measures["substitutions"] / len(ref_list)
+        # dele = measures["deletions"] / len(ref_list)
+        # inse = measures["insertions"] / len(ref_list)
+
+        wers.append(wer)
+
+    return wers
+
+
+# SIM Evaluation
+
+def run_sim(args):
+    rank, test_set, ckpt_dir = args
+    device = f"cuda:{rank}"
+
+    model = ECAPA_TDNN_SMALL(feat_dim=1024, feat_type='wavlm_large', config_path=None)
+    state_dict = torch.load(ckpt_dir, weights_only=True, map_location=lambda storage, loc: storage)
+    model.load_state_dict(state_dict['model'], strict=False)
+
+    use_gpu=True if torch.cuda.is_available() else False
+    if use_gpu:
+        model = model.cuda(device)
+    model.eval()
+
+    sim_list = []
+    for wav1, wav2, truth in tqdm(test_set):
+
+        wav1, sr1 = torchaudio.load(wav1)
+        wav2, sr2 = torchaudio.load(wav2)
+
+        resample1 = torchaudio.transforms.Resample(orig_freq=sr1, new_freq=16000)
+        resample2 = torchaudio.transforms.Resample(orig_freq=sr2, new_freq=16000)
+        wav1 = resample1(wav1)
+        wav2 = resample2(wav2)
+
+        if use_gpu:
+            wav1 = wav1.cuda(device)
+            wav2 = wav2.cuda(device)
+        with torch.no_grad():
+            emb1 = model(wav1)
+            emb2 = model(wav2)
+        
+        sim = F.cosine_similarity(emb1, emb2)[0].item()
+        # print(f"VSim score between two audios: {sim:.4f} (-1.0, 1.0).")
+        sim_list.append(sim)
+    
+    return sim_list
+
+
+# filter func for dirty data with many repetitions
+
+def repetition_found(text, length = 2, tolerance = 10):
+    pattern_count = defaultdict(int)
+    for i in range(len(text) - length + 1):
+        pattern = text[i:i + length]
+        pattern_count[pattern] += 1
+    for pattern, count in pattern_count.items():
+        if count > tolerance:
+            return True
+    return False
+
+
+# load model checkpoint for inference
+
+def load_checkpoint(model, ckpt_path, device, use_ema = True):
+    from ema_pytorch import EMA
+
+    ckpt_type = ckpt_path.split(".")[-1]
+    if ckpt_type == "safetensors":
+        from safetensors.torch import load_file
+        checkpoint = load_file(ckpt_path, device=device)
+    else:
+        checkpoint = torch.load(ckpt_path, weights_only=True, map_location=device)
+
+    if use_ema == True:
+        ema_model = EMA(model, include_online_model = False).to(device)
+        if ckpt_type == "safetensors":
+            ema_model.load_state_dict(checkpoint)
+        else:
+            ema_model.load_state_dict(checkpoint['ema_model_state_dict'])
+        ema_model.copy_params_from_ema_to_model()
+    else:
+        model.load_state_dict(checkpoint['model_state_dict'])
+        
+    return model