Spaces:
Sleeping
Sleeping
| ############################################################################ | |
| # Model: TTS with attention-based mechanism | |
| # Tokens: g2p + possitional embeddings | |
| # losses: MSE & BCE | |
| # Training: LJSpeech | |
| # ############################################################################ | |
| ################################### | |
| # Experiment Parameters and setup # | |
| ################################### | |
| seed: 1234 | |
| __set_seed: !apply:torch.manual_seed [!ref <seed>] | |
| # Folder set up | |
| # output_folder: !ref .\\results\\tts\\<seed> | |
| # save_folder: !ref <output_folder>\\save | |
| output_folder: !ref ./results/<seed> | |
| save_folder: !ref <output_folder>/save | |
| ################################ | |
| # Model Parameters and model # | |
| ################################ | |
| # Input parameters | |
| 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 | |
| input_encoder: !new:speechbrain.dataio.encoder.TextEncoder | |
| ################################ | |
| # Model Parameters and model # | |
| # Transformer Parameters | |
| ################################ | |
| d_model: 512 | |
| nhead: 8 | |
| num_encoder_layers: 3 | |
| num_decoder_layers: 3 | |
| dim_feedforward: 512 | |
| dropout: 0.1 | |
| # Decoder parameters | |
| # The number of frames in the target per encoder step | |
| n_frames_per_step: 1 | |
| decoder_rnn_dim: 1024 | |
| prenet_dim: 256 | |
| max_decoder_steps: 1000 | |
| gate_threshold: 0.5 | |
| p_decoder_dropout: 0.1 | |
| decoder_no_early_stopping: False | |
| blank_index: 0 # This special tokes is for padding | |
| # Masks | |
| lookahead_mask: !name:speechbrain.lobes.models.transformer.Transformer.get_lookahead_mask | |
| padding_mask: !name:speechbrain.lobes.models.transformer.Transformer.get_key_padding_mask | |
| ################################ | |
| # CNN 3-layers Prenet # | |
| ################################ | |
| # Encoder Prenet | |
| encoder_prenet: !new:module_classes.CNNPrenet | |
| # Decoder Prenet | |
| decoder_prenet: !new:module_classes.CNNDecoderPrenet | |
| ################################ | |
| # Positional Encodings # | |
| ################################ | |
| #encoder | |
| pos_emb_enc: !new:module_classes.ScaledPositionalEncoding | |
| input_size: !ref <d_model> | |
| max_len: 5000 | |
| #decoder | |
| pos_emb_dec: !new:module_classes.ScaledPositionalEncoding | |
| input_size: !ref <d_model> | |
| max_len: 5000 | |
| ################################ | |
| # S2S Transfomer # | |
| ################################ | |
| 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 | |
| ################################ | |
| # CNN 5-layers PostNet # | |
| ################################ | |
| decoder_postnet: !new:speechbrain.lobes.models.Tacotron2.Postnet | |
| # Linear transformation on the top of the decoder. | |
| stop_lin: !new:speechbrain.nnet.linear.Linear | |
| input_size: !ref <d_model> | |
| n_neurons: 1 | |
| # Linear transformation on the top of the decoder. | |
| mel_lin: !new:speechbrain.nnet.linear.Linear | |
| input_size: !ref <d_model> | |
| n_neurons: 80 | |
| modules: | |
| encoder_prenet: !ref <encoder_prenet> | |
| pos_emb_enc: !ref <pos_emb_enc> | |
| decoder_prenet: !ref <decoder_prenet> | |
| pos_emb_dec: !ref <pos_emb_dec> | |
| Seq2SeqTransformer: !ref <Seq2SeqTransformer> | |
| mel_lin: !ref <mel_lin> | |
| stop_lin: !ref <stop_lin> | |
| decoder_postnet: !ref <decoder_postnet> | |
| model: !new:torch.nn.ModuleList | |
| - [!ref <encoder_prenet>,!ref <pos_emb_enc>, | |
| !ref <decoder_prenet>, !ref <pos_emb_dec>, !ref <Seq2SeqTransformer>, | |
| !ref <mel_lin>, !ref <stop_lin>, !ref <decoder_postnet>] | |
| pretrained_model_path: ./model.ckpt | |
| # The pretrainer allows a mapping between pretrained files and instances that | |
| # are declared in the yaml. E.g here, we will download the file model.ckpt | |
| # and it will be loaded into "model" which is pointing to the <model> defined | |
| # before. | |
| pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer | |
| collect_in: !ref <save_folder> | |
| loadables: | |
| model: !ref <model> | |
| paths: | |
| model: !ref <pretrained_model_path> | |