Introduce code from Rust monorepo. This is a mess

2026-10-18 07:41:03 +00:00 · 2022-02-28 05:43:54 -05:00
parent 921fb49f9a
commit ba326bfcfd
1058 changed files with 192192 additions and 0 deletions
@@ -0,0 +1,157 @@
+name: &name "SpeakerNet"
+sample_rate: &sample_rate 16000
+repeat: &rep 2
+dropout: &drop 0.5
+separable: &separable True
+n_filters: &n_filters 512
+
+model:
+  train_ds:
+    manifest_filepath: ???
+    sample_rate: 16000
+    labels: null
+    batch_size: 64
+    shuffle: True
+    time_length: 8
+    is_tarred: False
+    tarred_audio_filepaths: null
+    tarred_shard_strategy: "scatter"
+
+  validation_ds:
+    manifest_filepath: ???
+    sample_rate: 16000
+    labels: null
+    batch_size: 128
+    shuffle: False
+    time_length: 8
+
+  test_ds:
+    manifest_filepath: ???
+    sample_rate: 16000
+    labels: null
+    batch_size: 1
+    shuffle: False
+    time_length: 8
+    embedding_dir: '.'
+
+  preprocessor:
+    _target_: nemo.collections.asr.modules.AudioToMelSpectrogramPreprocessor
+    normalize: "per_feature"
+    window_size: 0.02
+    sample_rate: *sample_rate
+    window_stride: 0.01
+    window: "hann"
+    features: &n_mels 64
+    n_fft: 512
+    frame_splicing: 1
+    dither: 0.00001
+
+  encoder:
+    _target_: nemo.collections.asr.modules.ConvASREncoder
+    feat_in: *n_mels
+    activation: relu
+    conv_mask: true
+
+    jasper:
+      -   filters: *n_filters
+          repeat: 1
+          kernel: [3]
+          stride: [1]
+          dilation: [1]
+          dropout: *drop
+          residual: true
+          separable: *separable
+
+      -   filters: *n_filters
+          repeat: *rep
+          kernel: [7]
+          stride: [1]
+          dilation: [1]
+          dropout: *drop
+          residual: true
+          separable: *separable
+
+      -   filters: *n_filters
+          repeat: *rep
+          kernel: [11]
+          stride: [1]
+          dilation: [1]
+          dropout: *drop
+          residual: true
+          separable: *separable
+
+      -   filters: *n_filters
+          repeat: *rep
+          kernel: [15]
+          stride: [1]
+          dilation: [1]
+          dropout: *drop
+          residual: true
+          separable: *separable
+
+      -   filters: &enc_feat_out 1500
+          repeat: 1
+          kernel: [1]
+          stride: [1]
+          dilation: [1]
+          dropout: 0.0
+          residual: false
+          separable: *separable
+
+  decoder:
+    _target_: nemo.collections.asr.modules.SpeakerDecoder
+    feat_in: *enc_feat_out
+    num_classes: 7205
+    pool_mode: 'xvector'
+    emb_sizes: 512,512
+    angular: False
+
+  loss:
+    scale: 30
+    margin: 0.2
+
+  optim:
+    name: novograd
+    # _target_: nemo.core.optim.optimizers.Novograd
+    lr: .008
+    # optimizer arguments
+    args:
+      name: auto
+      # _target_: nemo.core.config.optimizers.NovogradParams
+      betas: [0.95, 0.5]
+      weight_decay: 0.001
+
+    # scheduler setup
+    sched:
+      name: CosineAnnealing
+      iters_per_batch: 1 # computed at runtime
+      max_steps: null # computed at runtime or explicitly set here
+
+      # scheduler config override
+      args:
+        name: auto
+        # _target_: nemo.core.config.schedulers.CosineAnnealingParams
+        warmup_steps: null
+        warmup_ratio: 0.1
+        min_lr: 0.0
+        last_epoch: -1
+
+trainer:
+  gpus: 1 # number of gpus
+  max_epochs: 200
+  max_steps: null # computed at runtime if not set
+  num_nodes: 1
+  accelerator: ddp
+  accumulate_grad_batches: 1
+  amp_level: O0
+  deterministic: True
+  checkpoint_callback: False
+  logger: False
+  log_every_n_steps: 1  # Interval of logging.
+  val_check_interval: 1.0  # Set to 0.25 to check 4 times per epoch, or an int for number of iterations
+
+exp_manager:
+  exp_dir: null
+  name: *name
+  create_tensorboard_logger: True
+  create_checkpoint_callback: True
@@ -0,0 +1,145 @@
+name: &name "SpeakerNet"
+sample_rate: &sample_rate 16000
+repeat: &rep 2
+dropout: &drop 0.5
+separable: &separable True
+n_filters: &n_filters 512
+
+model:
+  train_ds:
+    manifest_filepath: ???
+    sample_rate: 16000
+    labels: null
+    batch_size: 64
+    shuffle: True
+    time_length: 8
+    is_tarred: False
+    tarred_audio_filepaths: null
+    tarred_shard_strategy: "scatter"
+
+  validation_ds:
+    manifest_filepath: ???
+    sample_rate: 16000
+    labels: null
+    batch_size: 128
+    shuffle: False
+    time_length: 8
+
+  preprocessor:
+    _target_: nemo.collections.asr.modules.AudioToMelSpectrogramPreprocessor
+    normalize: "per_feature"
+    window_size: 0.02
+    sample_rate: *sample_rate
+    window_stride: 0.01
+    window: "hann"
+    features: &n_mels 64
+    n_fft: 512
+    frame_splicing: 1
+    dither: 0.00001
+
+  encoder:
+    _target_: nemo.collections.asr.modules.ConvASREncoder
+    feat_in: *n_mels
+    activation: relu
+    conv_mask: true
+
+    jasper:
+      -   filters: *n_filters
+          repeat: 1
+          kernel: [3]
+          stride: [1]
+          dilation: [1]
+          dropout: *drop
+          residual: true
+          separable: *separable
+
+      -   filters: *n_filters
+          repeat: *rep
+          kernel: [7]
+          stride: [1]
+          dilation: [1]
+          dropout: *drop
+          residual: true
+          separable: *separable
+
+      -   filters: *n_filters
+          repeat: *rep
+          kernel: [11]
+          stride: [1]
+          dilation: [1]
+          dropout: *drop
+          residual: true
+          separable: *separable
+
+      -   filters: *n_filters
+          repeat: *rep
+          kernel: [15]
+          stride: [1]
+          dilation: [1]
+          dropout: *drop
+          residual: true
+          separable: *separable
+
+      -   filters: &enc_feat_out 1500
+          repeat: 1
+          kernel: [1]
+          stride: [1]
+          dilation: [1]
+          dropout: 0.0
+          residual: false
+          separable: *separable
+
+  decoder:
+    _target_: nemo.collections.asr.modules.SpeakerDecoder
+    feat_in: *enc_feat_out
+    num_classes: 7205
+    pool_mode: 'xvector'
+    emb_sizes: 256
+    angular: True
+
+  loss:
+    scale: 30
+    margin: 0.2
+
+  optim:
+    name: sgd
+    lr: .006
+    # optimizer arguments
+    args:
+      name: auto
+      weight_decay: 0.001
+      momentum: 0.9
+
+    # scheduler setup
+    sched:
+      name: CosineAnnealing
+      iters_per_batch: 1 # computed at runtime
+      max_steps: null # computed at runtime or explicitly set here
+
+      # scheduler config override
+      args:
+        name: auto
+        warmup_steps: null
+        warmup_ratio: 0.1
+        min_lr: 0.0
+        last_epoch: -1
+
+trainer:
+  gpus: 1 # number of gpus
+  max_epochs: 200
+  max_steps: null # computed at runtime if not set
+  num_nodes: 1
+  accelerator: ddp
+  accumulate_grad_batches: 1
+  amp_level: O0
+  deterministic: True
+  checkpoint_callback: False
+  logger: False
+  log_every_n_steps: 1  # Interval of logging.
+  val_check_interval: 1.0  # Set to 0.25 to check 4 times per epoch, or an int for number of iterations
+
+exp_manager:
+  exp_dir: null
+  name: *name
+  create_tensorboard_logger: True
+  create_checkpoint_callback: True
@@ -0,0 +1,27 @@
+name: &name "ClusterDiarizer"
+
+num_workers: 4
+sample_rate: &sample_rate 16000
+
+diarizer:
+  oracle_num_speakers: ??? # oracle number of speakers or pass null if not known. [int or path to file containing uniq-id with num of speakers of that session]
+  max_num_speakers: 20 # max number of speakers for each recording. If oracle num speakers is passed this value is ignored.
+  out_dir: ???
+  paths2audio_files: ??? #file containing paths to audio files for inference
+  path2groundtruth_rttm_files: null #file containing paths to ground truth rttm files for score computation or null if no reference rttms are present
+
+  vad:
+    model_path: null #.nemo local model path or pretrained model name or none
+    window_length_in_sec: 0.15
+    shift_length_in_sec: 0.01
+    threshold: 0.7 # tune threshold on dev set. Check <NeMo_git_root>/scripts/voice_activity_detection/vad_tune_threshold.py
+    vad_decision_smoothing: True
+    smoothing_params:
+      method: "median" 
+      overlap: 0.875
+
+  speaker_embeddings:
+    oracle_vad_manifest: null
+    model_path: ??? #.nemo local model path or pretrained model name (speakerverification_speakernet or speakerdiarization_speakernet)
+    window_length_in_sec: 1.5 # window length in sec for speaker embedding extraction
+    shift_length_in_sec: 0.75 # shift length in sec for speaker embedding extraction