--- tags: - espnet - audio - codec language: multilingual datasets: - amuse license: cc-by-4.0 --- ## ESPnet2 Codec model ### `espnet/amuse_speech_soundstream_16k` This model was trained by ftshijt using amuse recipe in [espnet](https://github.com/espnet/espnet/). ### Demo: How to use in ESPnet2 Follow the [ESPnet installation instructions](https://espnet.github.io/espnet/installation.html) if you haven't done that already. ```bash cd espnet git checkout 5201685018b0e8fb9826bc51a710623140a06627 pip install -e . cd egs2/amuse/codec1 ./run.sh --skip_data_prep false --skip_train true --download_model espnet/amuse_speech_soundstream_16k ``` ## Codec config
expand ``` config: conf/train_soundstream4_fs16000.yaml print_config: false log_level: INFO drop_last_iter: false dry_run: false iterator_type: chunk valid_iterator_type: null output_dir: exp_16k/codec_soundstream4_16k_speech ngpu: 1 seed: 777 num_workers: 1 num_att_plot: 0 dist_backend: nccl dist_init_method: env:// dist_world_size: 4 dist_rank: 0 local_rank: 0 dist_master_addr: localhost dist_master_port: 48299 dist_launcher: null multiprocessing_distributed: true unused_parameters: true sharded_ddp: false cudnn_enabled: true cudnn_benchmark: false cudnn_deterministic: false use_tf32: true collect_stats: false write_collected_feats: false max_epoch: 360 patience: null val_scheduler_criterion: - valid - loss early_stopping_criterion: - valid - loss - min best_model_criterion: - - valid - mel_loss - min - - train - mel_loss - min - - train - total_count - max keep_nbest_models: 5 nbest_averaging_interval: 0 grad_clip: -1 grad_clip_type: 2.0 grad_noise: false accum_grad: 1 no_forward_run: false resume: true train_dtype: float32 use_amp: false log_interval: 50 use_matplotlib: true use_tensorboard: true create_graph_in_tensorboard: false use_wandb: false wandb_project: null wandb_id: null wandb_entity: null wandb_name: null wandb_model_log_interval: -1 detect_anomaly: false use_adapter: false adapter: lora save_strategy: all adapter_conf: {} pretrain_path: null init_param: [] ignore_init_mismatch: false freeze_param: [] num_iters_per_epoch: 5000 batch_size: 64 valid_batch_size: null batch_bins: 1000000 valid_batch_bins: null train_shape_file: - exp_16k/codec_stats_speech/train/audio_shape valid_shape_file: - exp_16k/codec_stats_speech/valid/audio_shape batch_type: unsorted valid_batch_type: null fold_length: - 256000 sort_in_batch: descending shuffle_within_batch: false sort_batch: descending multiple_iterator: false chunk_length: 32000 chunk_shift_ratio: 0.5 num_cache_chunks: 128 chunk_excluded_key_prefixes: [] chunk_default_fs: null train_data_path_and_name_and_type: - - dump_16k/raw/speech_train/wav.scp - audio - kaldi_ark valid_data_path_and_name_and_type: - - dump_16k/raw/dev-small/wav.scp - audio - kaldi_ark multi_task_dataset: false allow_variable_data_keys: false max_cache_size: 0.0 max_cache_fd: 32 allow_multi_rates: false valid_max_cache_size: null exclude_weight_decay: false exclude_weight_decay_conf: {} optim: adam optim_conf: lr: 0.0002 betas: - 0.5 - 0.9 eps: 1.0e-09 weight_decay: 0.0 scheduler: exponentiallr scheduler_conf: gamma: 0.999875 optim2: adam optim2_conf: lr: 0.0002 betas: - 0.5 - 0.9 eps: 1.0e-09 weight_decay: 0.0 scheduler2: exponentiallr scheduler2_conf: gamma: 0.999875 generator_first: true skip_discriminator_prob: 0.0 model_conf: {} use_preprocessor: true codec: soundstream codec_conf: sampling_rate: 16000 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 sample_rate: 16000 discriminator_params: scales: 3 scale_downsample_pooling: AvgPool1d scale_downsample_pooling_params: kernel_size: 4 stride: 2 padding: 2 scale_discriminator_params: in_channels: 1 out_channels: 1 kernel_sizes: - 15 - 41 - 5 - 3 channels: 128 max_downsample_channels: 1024 max_groups: 16 bias: true downsample_scales: - 2 - 2 - 4 - 4 - 1 nonlinear_activation: LeakyReLU nonlinear_activation_params: negative_slope: 0.1 scale_follow_official_norm: false complexstft_discriminator_params: in_channels: 1 channels: 32 strides: - - 1 - 2 - - 2 - 2 - - 1 - 2 - - 2 - 2 - - 1 - 2 - - 2 - 2 chan_mults: - 1 - 2 - 4 - 4 - 8 - 8 n_fft: 1024 hop_length: 256 win_length: 1024 stft_normalized: false 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 fs: 16000 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 required: - output_dir version: '202402' distributed: true ```
### Citing ESPnet ```BibTex @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}, url={http://dx.doi.org/10.21437/Interspeech.2018-1456} } ``` or arXiv: ```bibtex @misc{watanabe2018espnet, title={ESPnet: End-to-End Speech Processing Toolkit}, 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}, year={2018}, eprint={1804.00015}, archivePrefix={arXiv}, primaryClass={cs.CL} } ```