inference/hyperparams.yaml

tokenizer: !new:sentencepiece.SentencePieceProcessor

pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer
  loadables:
    lm: !ref <lm_model>
    tokenizer: !ref <tokenizer>
    normalizer: !ref <normalizer>
    asr: !ref <asr_model>

# Feature parameters
sample_rate: 16000
n_fft: 400
n_mels: 80
hop_length: 20

compute_features: !new:speechbrain.lobes.features.Fbank
  sample_rate: !ref <sample_rate>
  n_fft: !ref <n_fft>
  n_mels: !ref <n_mels>
  hop_length: !ref <hop_length>

####################### Model parameters ###########################
# Transformer
d_model: 256
nhead: 4
num_encoder_layers: 12
num_decoder_layers: 6
d_ffn: 2048
transformer_dropout: 0.1
activation: !name:torch.nn.GELU
output_neurons: 5000
vocab_size: 5000

# Outputs
blank_index: 0
label_smoothing: 0.1
pad_index: 0
bos_index: 1
eos_index: 2
unk_index: 0

# Decoding parameters
min_decode_ratio: 0.0
max_decode_ratio: 1.0
valid_search_interval: 10
valid_beam_size: 10
test_beam_size: 10
ctc_weight_decode: 0.3
lm_weight: 0.2

############################## models ################################
CNN: !new:speechbrain.lobes.models.convolution.ConvolutionFrontEnd
  input_shape: !!python/tuple [8, 10, 8]
  num_blocks: 2
  num_layers_per_block: 1
  out_channels: !!python/tuple [256, 256]
  kernel_sizes: !!python/tuple [3, 3]
  strides: !!python/tuple [2, 2]
  residuals: !!python/tuple [False, False]

Transformer:
  !new:speechbrain.lobes.models.transformer.TransformerASR.TransformerASR # yamllint disable-line rule:line-length
  input_size: 5120
  tgt_vocab: !ref <output_neurons>
  d_model: !ref <d_model>
  nhead: !ref <nhead>
  num_encoder_layers: !ref <num_encoder_layers>
  num_decoder_layers: !ref <num_decoder_layers>
  d_ffn: !ref <d_ffn>
  dropout: !ref <transformer_dropout>
  activation: !ref <activation>
  normalize_before: True

lm_model:
  !new:speechbrain.lobes.models.transformer.TransformerLM.TransformerLM # yamllint disable-line rule:line-length
  vocab: !ref <output_neurons>
  d_model: 576
  nhead: 6
  num_encoder_layers: 6
  num_decoder_layers: 0
  d_ffn: 1538
  dropout: 0.2
  activation: !name:torch.nn.GELU
  normalize_before: False

ctc_lin: !new:speechbrain.nnet.linear.Linear
  input_size: !ref <d_model>
  n_neurons: !ref <output_neurons>

seq_lin: !new:speechbrain.nnet.linear.Linear
  input_size: !ref <d_model>
  n_neurons: !ref <output_neurons>

encoder: !new:speechbrain.nnet.containers.LengthsCapableSequential
  input_shape: [null, null, !ref <n_mels>]
  compute_features: !ref <compute_features>
  normalize: !ref <normalizer>
  cnn: !ref <CNN>
  transformer_encoder: !ref <Tencoder>

asr_model: !new:torch.nn.ModuleList
  - [!ref <CNN>, !ref <Transformer>, !ref <seq_lin>, !ref <ctc_lin>]

decoder: !new:speechbrain.decoders.S2STransformerBeamSearch
  modules: [!ref <Transformer>, !ref <seq_lin>, !ref <ctc_lin>]
  bos_index: !ref <bos_index>
  eos_index: !ref <eos_index>
  blank_index: !ref <blank_index>
  min_decode_ratio: !ref <min_decode_ratio>
  max_decode_ratio: !ref <max_decode_ratio>
  beam_size: !ref <test_beam_size>
  ctc_weight: !ref <ctc_weight_decode>
  lm_weight: !ref <lm_weight>
  lm_modules: !ref <lm_model>
  temperature: 1.15
  temperature_lm: 1.15
  using_eos_threshold: False
  length_normalization: True

Tencoder:
  !new:speechbrain.lobes.models.transformer.TransformerASR.EncoderWrapper
  transformer: !ref <Transformer>

normalizer: !new:speechbrain.processing.features.InputNormalization
  norm_type: global
  update_until_epoch: 4



modules:
  normalizer: !ref <normalizer>
  encoder: !ref <encoder>
  decoder: !ref <decoder>
# define two optimizers here for two-stage training



log_softmax: !new:torch.nn.LogSoftmax
  dim: -1