NewGame/libritts_codec_train_encodec_libritts

ESPnet3 gancodec model

Packed model bundle generated from egs3/libritts/codec.

Model

• Repository: NewGame/libritts_codec_train_encodec_libritts
• Recipe: egs3/libritts/codec
• Corpus: libritts
• System: gancodec
• Creator: ttrachu
• Created: 2026-09-05T19:42:24
• Branch: espnet3/recipe/libritts_codec
• Git: ec0ec71 (dirty)
• Origin: https://github.com/NewGamezzz/espnet.git

Model summary

• Class: ESPnetGANCodecModel
• Total parameters: 20,175,980
• Learnable parameters: 20,175,980 (100.0%)
• Non-trainable parameters: 0
• Parameter size: 80.7 MB
• Buffers: 33,670,960 (134.68 MB)
• Modules: 1,052 total, 410 leaf
• DType composition: torch.float32(266.9%)

Usage

from espnet3.publication import InferenceModel

model = InferenceModel.from_pretrained("NewGame/libritts_codec_train_encodec_libritts", trust_user_code=True)
result = model(sample)

Packaging

• Bundle: model_pack
• Exp dir: ./exp/train_encodec_libritts
• Strategy: copy experiment outputs; include extra recipe assets; register named artifact files; apply exclude filters

Results

dataset	ci_sdr	dns_overall	dns_p808	plcmos	sar	sdr	si_snr	sir	utmos
test	5.992	3.115	3.771	4.3182	6.0536	6.0536	3.4668	inf	3.9874

Training config

expand

num_device: 1
num_nodes: 1
task: espnet2.tasks.gan_codec.GANCodecTask
recipe_dir: .
data_dir: ./data
exp_tag: train_encodec_libritts
exp_dir: ./exp/train_encodec_libritts
stats_dir: ./exp/train_encodec_libritts/stats
inference_dir: ./exp/train_encodec_libritts/inference
create_dataset:
  recipe_dir: .
dataset:
  _target_: espnet3.components.data.data_organizer.DataOrganizer
  recipe_dir: .
  train:
  - data_src_args:
      split: train
  valid:
  - data_src_args:
      split: valid
  preprocessor:
    _target_: espnet2.train.preprocessor.CommonPreprocessor
    train: true
    token_type: null
    speech_name: audio
    force_single_channel: true
    audio_pad_value: 0.0
    _convert_: all
  _convert_: all
model:
  codec: encodec
  codec_conf:
    sampling_rate: 24000
    generator_params:
      hidden_dim: 512
      encdec_channels: 1
      encdec_n_filters: 32
      encdec_n_residual_layers: 3
      encdec_ratios:
      - 8
      - 5
      - 4
      - 2
      encdec_activation: ELU
      encdec_activation_params:
        alpha: 1.0
      encdec_norm: weight_norm
      encdec_kernel_size: 7
      encdec_residual_kernel_size: 7
      encdec_last_kernel_size: 7
      encdec_dilation_base: 2
      encdec_causal: false
      encdec_pad_mode: reflect
      encdec_true_skip: false
      encdec_compress: 2
      encdec_lstm: 2
      decoder_trim_right_ratio: 1.0
      decoder_final_activation: null
      decoder_final_activation_params: null
      quantizer_n_q: 32
      quantizer_bins: 1024
      quantizer_decay: 0.99
      quantizer_kmeans_init: true
      quantizer_kmeans_iters: 50
      quantizer_threshold_ema_dead_code: 2
      quantizer_target_bandwidth:
      - 2
      - 4
      - 8
      - 16
      - 32
    discriminator_params:
      msstft_discriminator_params:
        filters: 32
        in_channels: 1
        out_channels: 1
        norm: weight_norm
        n_ffts:
        - 1024
        - 2048
        - 512
        - 256
        - 128
        hop_lengths:
        - 256
        - 512
        - 128
        - 64
        - 32
        win_lengths:
        - 1024
        - 2048
        - 512
        - 256
        - 128
        activation: LeakyReLU
        activation_params:
          negative_slope: 0.3
    generator_adv_loss_params:
      average_by_discriminators: false
      loss_type: mse
    discriminator_adv_loss_params:
      average_by_discriminators: false
      loss_type: mse
    use_feat_match_loss: true
    feat_match_loss_params:
      average_by_discriminators: false
      average_by_layers: false
      include_final_outputs: true
    use_mel_loss: true
    mel_loss_params:
      range_start: 6
      range_end: 11
      window: hann
      n_mels: 80
      fmin: 0
      fmax: null
      log_base: null
    lambda_quantization: 0.0
    lambda_commit: 1.0
    lambda_reconstruct: 1.0
    lambda_adv: 1.0
    lambda_mel: 45.0
    lambda_feat_match: 2.0
    cache_generator_outputs: true
    use_loss_balancer: false
optimizers:
  generator:
    optimizer:
      _target_: torch.optim.AdamW
      lr: 0.0003
      betas:
      - 0.5
      - 0.9
      eps: 1.0e-09
      weight_decay: 0.0
      _convert_: all
    params: generator
    accum_grad_steps: 1
    step_every_n_iters: 1
    gradient_clip_val: null
  discriminator:
    optimizer:
      _target_: torch.optim.AdamW
      lr: 0.0003
      betas:
      - 0.5
      - 0.9
      eps: 1.0e-09
      weight_decay: 0.0
      _convert_: all
    params: discriminator
    accum_grad_steps: 1
    step_every_n_iters: 1
    gradient_clip_val: null
schedulers:
  generator:
    scheduler:
      _target_: torch.optim.lr_scheduler.ExponentialLR
      gamma: 0.999875
      _convert_: all
    interval: epoch
  discriminator:
    scheduler:
      _target_: torch.optim.lr_scheduler.ExponentialLR
      gamma: 0.999875
      _convert_: all
    interval: epoch
optimizer: null
scheduler: null
scheduler_interval: null
scheduler_monitor: null
best_model_criterion:
- - valid/mel_loss
  - 5
  - min
- - train/mel_loss
  - 5
  - min
- - step
  - 5
  - max
seed: null
init: null
parallel:
  env: local
  n_workers: 1
dataloader:
  collate_fn:
    _target_: espnet2.train.collate_fn.CommonCollateFn
    float_pad_value: 0.0
    int_pad_value: 0
    _convert_: all
  train:
    iter_factory:
      _target_: espnet2.iterators.chunk_iter_factory.ChunkIterFactory
      batch_size: 4
      chunk_length: 61440
      num_cache_chunks: 64
      num_samples_per_epoch: 10000
      shuffle: true
      num_workers: 4
      collate_fn:
        _target_: espnet2.train.collate_fn.CommonCollateFn
        float_pad_value: 0.0
        int_pad_value: 0
        _convert_: all
      batches:
        type: unsorted
        batch_size: 1
        batch_bins: 0
        shape_files:
        - ./data/manifest/train.tsv
      _convert_: all
  valid:
    iter_factory:
      _target_: espnet2.iterators.chunk_iter_factory.ChunkIterFactory
      batch_size: 4
      chunk_length: 61440
      num_cache_chunks: 64
      shuffle: false
      num_workers: 4
      collate_fn:
        _target_: espnet2.train.collate_fn.CommonCollateFn
        float_pad_value: 0.0
        int_pad_value: 0
        _convert_: all
      batches:
        type: unsorted
        batch_size: 1
        batch_bins: 0
        shape_files:
        - ./data/manifest/valid.tsv
      _convert_: all
trainer:
  accelerator: auto
  devices: 1
  num_nodes: 1
  strategy: auto
  accumulate_grad_batches: 1
  check_val_every_n_epoch: 1
  log_every_n_steps: 500
  max_epochs: 60
  gan:
    generator_first: true
    skip_discriminator_prob: 0.3
  logger:
  - _target_: lightning.pytorch.loggers.TensorBoardLogger
    save_dir: ./exp/train_encodec_libritts/tensorboard
    name: tb_logger
    _convert_: all
fit: {}

Citing ESPnet

@inproceedings{watanabe2018espnet,
  author={Shinji Watanabe and Takaaki Hori and Shigeki Karita and Tomoki Hayashi and
    Jiro Nishitoba and Yuya Unno and Nelson Yalta and Jahn Heymann and Matthew Wiesner
    and Nanxin Chen and Adithya Renduchintala and Tsubasa Ochiai},
  title={{ESPnet}: End-to-End Speech Processing Toolkit},
  year={2018},
  booktitle={Proceedings of Interspeech},
  pages={2207--2211},
  doi={10.21437/Interspeech.2018-1456}
}