storyteller-ml/tts/tacotron2-cac/infer_arpa.py

import os
import json
# import re
import argparse
# from string import punctuation

import torch
import numpy as np
from torch.utils.data import DataLoader
# from g2p_en import G2p
# from pypinyin import pinyin, Style

from utils.model import get_model, get_vocoder
from utils.tools import get_configs_of, to_device, infer_one_sample_s #, read_lexicon
from dataset import TextDataset
from text import text_to_sequence, sequence_to_text
from text.cleaners import to_arpa

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
from model import Tacotron2

def get_tac2(configs, device):
    (preprocess_config, model_config, train_config) = configs

    model = Tacotron2(preprocess_config, model_config, train_config).to(device)

    model.eval()
    model.requires_grad_ = False
    return model




def preprocess_english(text, preprocess_config):
    sequence = text_to_sequence(
        text, preprocess_config["preprocessing"]["text"]["text_cleaners"]
    )
    print("Raw Text Sequence: {}".format(text))
    print("Sequence: {}".format(" ".join([str(id) for id in sequence_to_text(sequence)])))
    print("Sequence Input: {}".format(" ".join([str(id) for id in sequence])))
    return np.array(sequence)


def synthesize(model, args, configs, mel_stats, vocoder, batchs):
    preprocess_config, model_config, train_config = configs

    for batch in batchs:
        batch = to_device(batch, device, mel_stats)
        with torch.no_grad():
            # Forward
            output = model.inference(*batch[2:4], *batch[5:7])
            infer_one_sample_s(
                batch,
                output,
                vocoder,
                mel_stats,
                model_config,
                preprocess_config,
                args.out_path,
                args,
            )


if __name__ == "__main__":

    parser = argparse.ArgumentParser()
    parser.add_argument(
        "--mode",
        type=str,
        choices=["batch", "single"],
        required=True,
        help="Synthesize a whole dataset or a single sentence",
    )
    parser.add_argument(
        "--source",
        type=str,
        default=None,
        help="path to a source file with format like train.txt and val.txt, for batch mode only",
    )
    parser.add_argument(
        "--text",
        type=str,
        default=None,
        help="raw text to synthesize, for single-sentence mode only",
    )
    parser.add_argument(
        "--speaker_id",
        type=str,
        default="p225",
        help="speaker ID for multi-speaker synthesis, for single-sentence mode only",
    )
    parser.add_argument(
        "--dataset",
        type=str,
        required=True,
        help="name of dataset",
    )
    parser.add_argument(
        "--checkpoint",
        type=str,
        required=True,
        help="Path to model file.",
    )
    parser.add_argument(
        "--out-path",
        type=str,
        required=True,
        help="Path to store files",
    )
    args = parser.parse_args()

    # Check source texts
    if args.mode == "batch":
        assert args.source is not None and args.text is None
    if args.mode == "single":
        assert args.source is None and args.text is not None

    # Read Config
    preprocess_config, model_config, train_config = get_configs_of(args.dataset)
    configs = (preprocess_config, model_config, train_config)
    stats = None
    mel_stats = None

    os.makedirs(
        args.out_path, exist_ok=True)

    # Get model
    model = get_tac2(configs, device)
    ckpt = torch.load(args.checkpoint)
    model.load_state_dict(ckpt["model"])


    # Load vocoder
    vocoder = get_vocoder(model_config, device)

    # Preprocess texts
    if args.mode == "batch":
        # Get dataset
        dataset = TextDataset(args.source, preprocess_config, model_config)
        batchs = DataLoader(
            dataset,
            batch_size=1, # currently only 1 is supported
            collate_fn=dataset.collate_fn,
        )
    if args.mode == "single":
        ids = raw_texts = [args.text[:100]]

        # Speaker Info
        is_multi_speaker = model_config["multi_speaker"]

        if is_multi_speaker:
            load_spker_embed = model_config["multi_speaker"] \
                and preprocess_config["preprocessing"]["speaker_embedder"] != 'none'
            with open(os.path.join(preprocess_config["path"]["preprocessed_path"], "speakers.json")) as f:
                speaker_map = json.load(f)
            speakers = np.array([speaker_map[args.speaker_id]]) if model_config["multi_speaker"] else np.array([0]) # single speaker is allocated 0
            spker_embed = np.load(os.path.join(
                preprocess_config["path"]["preprocessed_path"],
                "spker_embed",
                "{}-spker_embed.npy".format(args.speaker_id),
            )) if load_spker_embed else None
        else:
            speakers = np.array([0])
            spker_embed = None

        
        texts = np.array([preprocess_english(to_arpa(args.text), preprocess_config)])
        
        text_lens = np.array([len(texts[0])])
        batchs = [(ids, raw_texts, speakers, texts, text_lens, max(text_lens), spker_embed)]

    synthesize(model, args, configs, mel_stats, vocoder, batchs)