hyperparams.yaml

################################
# Audio Parameters             #
################################
sample_rate: 22050
hop_length: 256
win_length: 1024
n_mel_channels: 80
n_fft: 1024
mel_fmin: 0.0
mel_fmax: 8000.0
power: 1
normalized: False
min_max_energy_norm: True
norm: "slaney"
mel_scale: "slaney"
dynamic_range_compression: True
mel_normalized: False
min_f0: 65  #(torchaudio pyin values)
max_f0: 2093 #(torchaudio pyin values)

positive_weight: 5.0
lexicon:
    - AA
    - AE
    - AH
    - AO
    - AW
    - AY
    - B
    - CH
    - D
    - DH
    - EH
    - ER
    - EY
    - F
    - G
    - HH
    - IH
    - IY
    - JH
    - K
    - L
    - M
    - N
    - NG
    - OW
    - OY
    - P
    - R
    - S
    - SH
    - T
    - TH
    - UH
    - UW
    - V
    - W
    - Y
    - Z
    - ZH
    - ' '
n_symbols: 42 #fixed depending on symbols in the lexicon +1 for a dummy symbol used for padding
padding_idx: 0

# Define model architecture
d_model: 512
nhead: 8
num_encoder_layers: 6
num_decoder_layers: 6
dim_feedforward: 2048
dropout: 0.2
blank_index: 0 # This special token is for padding
bos_index: 1
eos_index: 2
stop_weight: 0.45
stop_threshold: 0.5


###################PRENET#######################
enc_pre_net: !new:models.EncoderPrenet
dec_pre_net: !new:models.DecoderPrenet


encoder_emb: !new:torch.nn.Embedding
    num_embeddings: 128
    embedding_dim: !ref <d_model>
    padding_idx: !ref <blank_index>

pos_emb_enc: !new:models.ScaledPositionalEncoding
    d_model: !ref <d_model>

decoder_emb: !new:torch.nn.Embedding
    num_embeddings: 128
    embedding_dim: !ref <d_model>
    padding_idx: !ref <blank_index>

pos_emb_dec: !new:models.ScaledPositionalEncoding
    d_model: !ref <d_model>


Seq2SeqTransformer: !new:torch.nn.Transformer
    d_model: !ref <d_model>
    nhead: !ref <nhead>
    num_encoder_layers: !ref <num_encoder_layers>
    num_decoder_layers: !ref <num_decoder_layers>
    dim_feedforward: !ref <dim_feedforward>
    dropout: !ref <dropout>
    batch_first: True

postnet: !new:models.PostNet
    mel_channels: !ref <n_mel_channels>
    postnet_channels: 512
    kernel_size: 5
    postnet_layers: 5

mel_lin: !new:speechbrain.nnet.linear.Linear
    input_size: !ref <d_model>
    n_neurons: !ref <n_mel_channels>

stop_lin: !new:speechbrain.nnet.linear.Linear
    input_size: !ref <d_model>
    n_neurons: 1

mel_spec_feats: !name:speechbrain.lobes.models.FastSpeech2.mel_spectogram
    sample_rate: !ref <sample_rate>
    hop_length: !ref <hop_length>
    win_length: !ref <win_length>
    n_fft: !ref <n_fft>
    n_mels: !ref <n_mel_channels>
    f_min: !ref <mel_fmin>
    f_max: !ref <mel_fmax>
    power: !ref <power>
    normalized: !ref <normalized>
    min_max_energy_norm: !ref <min_max_energy_norm>
    norm: !ref <norm>
    mel_scale: !ref <mel_scale>
    compression: !ref <dynamic_range_compression>

modules:
    enc_pre_net: !ref <enc_pre_net>
    encoder_emb: !ref <encoder_emb>
    pos_emb_enc: !ref <pos_emb_enc>

    dec_pre_net: !ref <dec_pre_net>
    #decoder_emb: !ref <decoder_emb>
    pos_emb_dec: !ref <pos_emb_dec>

    Seq2SeqTransformer: !ref <Seq2SeqTransformer>
    postnet: !ref <postnet>
    mel_lin: !ref <mel_lin>
    stop_lin: !ref <stop_lin>
    model: !ref <model>

lookahead_mask: !name:speechbrain.lobes.models.transformer.Transformer.get_lookahead_mask
padding_mask: !name:speechbrain.lobes.models.transformer.Transformer.get_key_padding_mask

model: !new:torch.nn.ModuleList
    - [!ref <enc_pre_net>, !ref <encoder_emb>, !ref <pos_emb_enc>, !ref <dec_pre_net>, !ref <pos_emb_dec>, !ref <Seq2SeqTransformer>, !ref <postnet>, !ref <mel_lin>, !ref <stop_lin>]

label_encoder: !new:speechbrain.dataio.encoder.TextEncoder

pretrained_path: /content/

pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer
    loadables:
        model: !ref <model>
        label_encoder: !ref <label_encoder>
    paths:
        model: !ref <pretrained_path>/model.ckpt
        label_encoder: !ref <pretrained_path>/label_encoder.txt