nvidia-nemo--speech/examples/tts/conf/aligner.yaml

# This config contains the default values for training Aligner model on LJSpeech dataset.
# If you want to train model on other dataset, you can change config values according to your dataset.
# Most dataset-specific arguments are in the head of the config file, see below.

name: Aligner

train_dataset: ???
validation_datasets: ???
sup_data_path: ???
sup_data_types: [ "align_prior_matrix" ]

# Default values for dataset with sample_rate=22050
sample_rate: 22050
n_mel_channels: 80
n_window_size: 1024
n_window_stride: 256
n_fft: 1024
lowfreq: 0
highfreq: 8000
window: hann

phoneme_dict_path: "scripts/tts_dataset_files/cmudict-0.7b_nv22.10"
heteronyms_path: "scripts/tts_dataset_files/heteronyms-052722"

model:
  symbols_embedding_dim: 384
  bin_loss_start_ratio: 0.2
  bin_loss_warmup_epochs: 100

  sample_rate: ${sample_rate}
  n_mel_channels: ${n_mel_channels}
  n_window_size: ${n_window_size}
  n_window_stride: ${n_window_stride}
  n_fft: ${n_fft}
  lowfreq: ${lowfreq}
  highfreq: ${highfreq}
  window: ${window}

  text_normalizer:
    _target_: nemo_text_processing.text_normalization.normalize.Normalizer
    lang: en
    input_case: cased

  text_normalizer_call_kwargs:
    verbose: false
    punct_pre_process: true
    punct_post_process: true

  text_tokenizer:
    _target_: nemo.collections.common.tokenizers.text_to_speech.tts_tokenizers.EnglishPhonemesTokenizer
    punct: true
    stresses: true
    chars: true
    apostrophe: true
    pad_with_space: true
    g2p:
      _target_: nemo.collections.tts.g2p.models.en_us_arpabet.EnglishG2p
      phoneme_dict: ${phoneme_dict_path}
      heteronyms: ${heteronyms_path}

  train_ds:
    dataset:
      _target_: nemo.collections.tts.data.dataset.TTSDataset
      manifest_filepath: ${train_dataset}
      sample_rate: ${model.sample_rate}
      sup_data_path: ${sup_data_path}
      sup_data_types: ${sup_data_types}
      n_fft: ${model.n_fft}
      win_length: ${model.n_window_size}
      hop_length: ${model.n_window_stride}
      window: ${model.window}
      n_mels: ${model.n_mel_channels}
      lowfreq: ${model.lowfreq}
      highfreq: ${model.highfreq}
      max_duration: null
      min_duration: 0.1
      ignore_file: null
      trim: false

    dataloader_params:
      drop_last: false
      shuffle: true
      batch_size: 64
      num_workers: 4
      pin_memory: true

  validation_ds:
    dataset:
      _target_: nemo.collections.tts.data.dataset.TTSDataset
      manifest_filepath: ${validation_datasets}
      sample_rate: ${model.sample_rate}
      sup_data_path: ${sup_data_path}
      sup_data_types: ${sup_data_types}
      n_fft: ${model.n_fft}
      win_length: ${model.n_window_size}
      hop_length: ${model.n_window_stride}
      window: ${model.window}
      n_mels: ${model.n_mel_channels}
      lowfreq: ${model.lowfreq}
      highfreq: ${model.highfreq}
      max_duration: null
      min_duration: 0.1
      ignore_file: null
      trim: false

    dataloader_params:
      drop_last: false
      shuffle: false
      batch_size: 64
      num_workers: 1
      pin_memory: true

  preprocessor:
    _target_: nemo.collections.asr.modules.AudioToMelSpectrogramPreprocessor
    features: ${model.n_mel_channels}
    lowfreq: ${model.lowfreq}
    highfreq: ${model.highfreq}
    n_fft: ${model.n_fft}
    n_window_size: ${model.n_window_size}
    window_size: false
    n_window_stride: ${model.n_window_stride}
    window_stride: false
    pad_to: 1
    pad_value: -11.52
    sample_rate: ${model.sample_rate}
    window: ${model.window}
    normalize: null
    preemph: null
    dither: 0.0
    frame_splicing: 1
    log: true
    log_zero_guard_type: clamp
    log_zero_guard_value: 1e-05
    mag_power: 1.0

  alignment_encoder:
    _target_: nemo.collections.tts.modules.aligner.AlignmentEncoder
    n_mel_channels: ${model.n_mel_channels}
    n_text_channels: ${model.symbols_embedding_dim}
    n_att_channels: ${model.n_mel_channels}

  optim:
    name: adam
    lr: 1e-3
    weight_decay: 1e-6

    sched:
      name: CosineAnnealing
      min_lr: 5e-5
      warmup_ratio: 0.35

trainer:
  devices: 1
  num_nodes: 1
  accelerator: gpu
  strategy: ddp
  precision: 32
  max_epochs: 1000
  accumulate_grad_batches: 1
  gradient_clip_val: 1000.0
  enable_checkpointing: false # Provided by exp_manager
  logger: false # Provided by exp_manager
  log_every_n_steps: 100
  check_val_every_n_epoch: 1
  benchmark: false

exp_manager:
  exp_dir: null
  name: ${name}
  create_tensorboard_logger: true
  create_checkpoint_callback: true
  checkpoint_callback_params:
    monitor: val_forward_sum_loss
    mode: min
  create_wandb_logger: false
  wandb_logger_kwargs:
    name: null
    project: null
    entity: null
  resume_if_exists: false
  resume_ignore_no_checkpoint: false