ESPnet2 pretrained model, kan-bayashi/csmsc_tts_train_transformer_raw_phn_pypinyin_g2p_phone_train.loss.ave, fs=24000, lang=zh

This model was trained by kan-bayashi using csmsc/tts1 recipe in espnet.

• Python API

  See https://github.com/espnet/espnet_model_zoo

• Evaluate in the recipe

  git clone https://github.com/espnet/espnet
  cd espnet
  git checkout 67ca53d61932e4ff3325f1ea7215eead03f300bb
  pip install -e .
  cd egs2/csmsc/tts1
  # Download the model file here
  ./run.sh --skip_data_prep false --skip_train true --download_model kan-bayashi/csmsc_tts_train_transformer_raw_phn_pypinyin_g2p_phone_train.loss.ave
  

• Config

  config: conf/tuning/train_transformer.yaml
  print_config: false
  log_level: INFO
  dry_run: false
  iterator_type: sequence
  output_dir: exp/tts_train_transformer_raw_phn_pypinyin_g2p_phone
  ngpu: 1
  seed: 0
  num_workers: 1
  num_att_plot: 3
  dist_backend: nccl
  dist_init_method: env://
  dist_world_size: 4
  dist_rank: 0
  local_rank: 0
  dist_master_addr: localhost
  dist_master_port: 35399
  dist_launcher: null
  multiprocessing_distributed: true
  cudnn_enabled: true
  cudnn_benchmark: false
  cudnn_deterministic: true
  collect_stats: false
  write_collected_feats: false
  max_epoch: 200
  patience: null
  val_scheduler_criterion:
  - valid
  - loss
  early_stopping_criterion:
  - valid
  - loss
  - min
  best_model_criterion:
  -   - valid
      - loss
      - min
  -   - train
      - loss
      - min
  keep_nbest_models: 5
  grad_clip: 1.0
  grad_clip_type: 2.0
  grad_noise: false
  accum_grad: 2
  no_forward_run: false
  resume: true
  train_dtype: float32
  use_amp: false
  log_interval: null
  pretrain_path: []
  pretrain_key: []
  num_iters_per_epoch: 1000
  batch_size: 20
  valid_batch_size: null
  batch_bins: 9000000
  valid_batch_bins: null
  train_shape_file:
  - exp/tts_stats_raw_phn_pypinyin_g2p_phone/train/text_shape.phn
  - exp/tts_stats_raw_phn_pypinyin_g2p_phone/train/speech_shape
  valid_shape_file:
  - exp/tts_stats_raw_phn_pypinyin_g2p_phone/valid/text_shape.phn
  - exp/tts_stats_raw_phn_pypinyin_g2p_phone/valid/speech_shape
  batch_type: numel
  valid_batch_type: null
  fold_length:
  - 150
  - 240000
  sort_in_batch: descending
  sort_batch: descending
  multiple_iterator: false
  chunk_length: 500
  chunk_shift_ratio: 0.5
  num_cache_chunks: 1024
  train_data_path_and_name_and_type:
  -   - dump/raw/tr_no_dev/text
      - text
      - text
  -   - dump/raw/tr_no_dev/wav.scp
      - speech
      - sound
  valid_data_path_and_name_and_type:
  -   - dump/raw/dev/text
      - text
      - text
  -   - dump/raw/dev/wav.scp
      - speech
      - sound
  allow_variable_data_keys: false
  max_cache_size: 0.0
  valid_max_cache_size: null
  optim: adam
  optim_conf:
      lr: 1.0
  scheduler: noamlr
  scheduler_conf:
      model_size: 512
      warmup_steps: 8000
  token_list:
  - 
  - 
  - d
  - sh
  - j
  - l
  - "\u3002"
  - zh
  - "\uFF0C"
  - i4
  - x
  - h
  - b
  - e
  - g
  - t
  - m
  - z
  - q
  - i1
  - i3
  - ch
  - u4
  - n
  - f
  - i2
  - r
  - k
  - s
  - e4
  - ai4
  - a1
  - c
  - p
  - ian4
  - uo3
  - ao3
  - ai2
  - ao4
  - an4
  - ong1
  - u3
  - ing2
  - en2
  - e2
  - u2
  - ui4
  - ian2
  - iou3
  - ang4
  - u1
  - iao4
  - uo4
  - eng2
  - a4
  - in1
  - eng1
  - ou3
  - ang1
  - ian1
  - ou4
  - ing1
  - uo1
  - an1
  - ian3
  - ie3
  - a3
  - ing4
  - an3
  - an2
  - "\xFC4"
  - iao3
  - ei4
  - ong2
  - en1
  - uei4
  - "\xFCan2"
  - ang2
  - ang3
  - iu4
  - iang4
  - ai3
  - ao1
  - ou1
  - eng4
  - iang3
  - en3
  - ai1
  - ong4
  - ie4
  - e3
  - ia1
  - uo2
  - ia4
  - "\xFC3"
  - uan1
  - er2
  - ei3
  - ei2
  - iang1
  - i
  - ing3
  - en4
  - "\xFC2"
  - uan3
  - e1
  - in2
  - iao1
  - in4
  - ie1
  - ong3
  - iang2
  - ie2
  - uan4
  - a2
  - ui3
  - eng3
  - uan2
  - "\xFCe4"
  - uai4
  - ou2
  - "\uFF1F"
  - "\xFCe2"
  - in3
  - uang3
  - uang1
  - iu2
  - en
  - a
  - ao2
  - ua4
  - un1
  - ui1
  - uei2
  - iong4
  - uang2
  - v3
  - ui2
  - iao2
  - uang4
  - "\xFC1"
  - ei1
  - o2
  - er4
  - iou2
  - iou4
  - "\uFF01"
  - ua1
  - "\xFCan4"
  - iu3
  - un4
  - "\xFCan3"
  - uen2
  - "\xFCn4"
  - iu1
  - un3
  - uen4
  - er3
  - "\xFCn1"
  - un2
  - "\xFCn2"
  - o4
  - o1
  - ua2
  - uei1
  - uei3
  - ia3
  - iong3
  - ua3
  - ia
  - "\xFCe1"
  - v4
  - "\xFCan1"
  - iong1
  - ia2
  - uai1
  - iong2
  - iou1
  - uai3
  - "\xFCe3"
  - uen1
  - uen3
  - uai2
  - o3
  - er
  - ve4
  - io1
  - "\xFCn3"
  - u
  - ou
  - o
  - ang
  - ueng1
  - v2
  - uo
  - ao
  - ueng4
  - ua
  - ei
  - uen
  - an
  - '2'
  - ueng3
  - iang
  - "\xFC"
  - ie
  - "\uFF30"
  - "\uFF22"
  - ai
  - 
  odim: null
  model_conf: {}
  use_preprocessor: true
  token_type: phn
  bpemodel: null
  non_linguistic_symbols: null
  cleaner: null
  g2p: pypinyin_g2p_phone
  feats_extract: fbank
  feats_extract_conf:
      fs: 24000
      fmin: 80
      fmax: 7600
      n_mels: 80
      hop_length: 300
      n_fft: 2048
      win_length: 1200
  normalize: global_mvn
  normalize_conf:
      stats_file: exp/tts_stats_raw_phn_pypinyin_g2p_phone/train/feats_stats.npz
  tts: transformer
  tts_conf:
      embed_dim: 0
      eprenet_conv_layers: 0
      eprenet_conv_filts: 0
      eprenet_conv_chans: 0
      dprenet_layers: 2
      dprenet_units: 256
      adim: 512
      aheads: 8
      elayers: 6
      eunits: 1024
      dlayers: 6
      dunits: 1024
      positionwise_layer_type: conv1d
      positionwise_conv_kernel_size: 1
      postnet_layers: 5
      postnet_filts: 5
      postnet_chans: 256
      use_masking: true
      bce_pos_weight: 5.0
      use_scaled_pos_enc: true
      encoder_normalize_before: false
      decoder_normalize_before: false
      reduction_factor: 1
      init_type: xavier_uniform
      init_enc_alpha: 1.0
      init_dec_alpha: 1.0
      eprenet_dropout_rate: 0.0
      dprenet_dropout_rate: 0.5
      postnet_dropout_rate: 0.5
      transformer_enc_dropout_rate: 0.1
      transformer_enc_positional_dropout_rate: 0.1
      transformer_enc_attn_dropout_rate: 0.1
      transformer_dec_dropout_rate: 0.1
      transformer_dec_positional_dropout_rate: 0.1
      transformer_dec_attn_dropout_rate: 0.1
      transformer_enc_dec_attn_dropout_rate: 0.1
      use_guided_attn_loss: true
      num_heads_applied_guided_attn: 2
      num_layers_applied_guided_attn: 2
      modules_applied_guided_attn:
      - encoder-decoder
      guided_attn_loss_lambda: 10.0
  pitch_extract: null
  pitch_extract_conf: {}
  pitch_normalize: null
  pitch_normalize_conf: {}
  energy_extract: null
  energy_extract_conf: {}
  energy_normalize: null
  energy_normalize_conf: {}
  required:
  - output_dir
  - token_list
  distributed: true