transformer.yaml

# https://yaml.org/type/float.html
# network architecture
model:
    cmvn_file:  
    cmvn_file_type: "json"
    # encoder related
    encoder: transformer
    encoder_conf:
        output_size: 256    # dimension of attention
        attention_heads: 4
        linear_units: 2048  # the number of units of position-wise feed forward
        num_blocks: 12      # the number of encoder blocks
        dropout_rate: 0.1
        positional_dropout_rate: 0.1
        attention_dropout_rate: 0.0
        input_layer: conv2d # encoder input type, you can chose conv2d, conv2d6 and conv2d8
        normalize_before: true

    # decoder related
    decoder: transformer
    decoder_conf:
        attention_heads: 4
        linear_units: 2048
        num_blocks: 6
        dropout_rate: 0.1
        positional_dropout_rate: 0.1
        self_attention_dropout_rate: 0.0
        src_attention_dropout_rate: 0.0

    # hybrid CTC/attention
    model_conf:
        ctc_weight: 0.3
        ctc_dropoutrate: 0.0
        ctc_grad_norm_type: batch
        lsm_weight: 0.1     # label smoothing option
        length_normalized_loss: false

data:
  train_manifest: data/manifest.train
  dev_manifest: data/manifest.dev
  test_manifest: data/manifest.test-clean

collator:
  vocab_filepath: data/lang_char/train_960_unigram5000_units.txt
  unit_type: spm
  spm_model_prefix: data/lang_char/train_960_unigram5000
  feat_dim: 83
  stride_ms: 10.0
  window_ms: 25.0
  sortagrad: 0 # Feed samples from shortest to longest ; -1: enabled for all epochs, 0: disabled, other: enabled for 'other' epochs 
  batch_size: 30 
  maxlen_in: 512  # if input length  > maxlen-in, batchsize is automatically reduced
  maxlen_out: 150  # if output length > maxlen-out, batchsize is automatically reduced
  minibatches: 0 # for debug
  batch_count: auto
  batch_bins: 0 
  batch_frames_in: 0
  batch_frames_out: 0
  batch_frames_inout: 0
  augmentation_config: conf/augmentation.json
  num_workers: 0
  subsampling_factor: 1
  num_encs: 1


training:
  n_epoch: 120
  accum_grad: 2
  log_interval: 100
  checkpoint:
    kbest_n: 50
    latest_n: 5

optim: adam
optim_conf:
  global_grad_clip: 5.0
  weight_decay: 1.0e-06
scheduler: warmuplr     # pytorch v1.1.0+ required
scheduler_conf:
  lr: 0.004
  warmup_steps: 25000
  lr_decay: 1.0

decoding:
  batch_size: 1
  error_rate_type: wer
  decoding_method: attention  # 'attention', 'ctc_greedy_search', 'ctc_prefix_beam_search', 'attention_rescoring'
  lang_model_path: data/lm/common_crawl_00.prune01111.trie.klm
  alpha: 2.5
  beta: 0.3
  beam_size: 10
  cutoff_prob: 1.0
  cutoff_top_n: 0
  num_proc_bsearch: 8
  ctc_weight: 0.5 # ctc weight for attention rescoring decode mode.
  decoding_chunk_size: -1 # decoding chunk size. Defaults to -1.
      # <0: for decoding, use full chunk.
      # >0: for decoding, use fixed chunk size as set.
      # 0: used for training, it's prohibited here. 
  num_decoding_left_chunks: -1  # number of left chunks for decoding. Defaults to -1.
  simulate_streaming: False  # simulate streaming inference. Defaults to False.