# ################################ # Model: VGG2 + LSTM + time pooling # Augmentation: SpecAugment # Authors: Titouan Parcollet, Mirco Ravanelli, Peter Plantinga, Ju-Chieh Chou, # and Abdel HEBA 2020 # ################################ # Feature parameters (FBANKS etc) sample_rate: 16000 n_fft: 400 n_mels: 80 # Model parameters activation: !name:torch.nn.LeakyReLU dropout: 0.15 cnn_blocks: 3 cnn_channels: (128, 200, 256) inter_layer_pooling_size: (2, 2, 2) cnn_kernelsize: (3, 3) time_pooling_size: 4 rnn_class: !name:speechbrain.nnet.RNN.LSTM rnn_layers: 5 rnn_neurons: 1024 rnn_bidirectional: True dnn_blocks: 2 dnn_neurons: 1024 emb_size: 128 dec_neurons: 1024 # Outputs output_neurons: 500 # BPE size, index(blank/eos/bos) = 0 # Decoding parameters # Be sure that the bos and eos index match with the BPEs ones blank_index: 0 bos_index: 0 eos_index: 0 min_decode_ratio: 0.0 max_decode_ratio: 1.0 beam_size: 80 eos_threshold: 1.5 using_max_attn_shift: True max_attn_shift: 140 ctc_weight_decode: 0.0 temperature: 1.50 normalizer: !new:speechbrain.processing.features.InputNormalization norm_type: global compute_features: !new:speechbrain.lobes.features.Fbank sample_rate: !ref n_fft: !ref n_mels: !ref enc: !new:speechbrain.lobes.models.CRDNN.CRDNN input_shape: [null, null, !ref ] activation: !ref dropout: !ref cnn_blocks: !ref cnn_channels: !ref cnn_kernelsize: !ref inter_layer_pooling_size: !ref time_pooling: True using_2d_pooling: False time_pooling_size: !ref rnn_class: !ref rnn_layers: !ref rnn_neurons: !ref rnn_bidirectional: !ref rnn_re_init: True dnn_blocks: !ref dnn_neurons: !ref emb: !new:speechbrain.nnet.embedding.Embedding num_embeddings: !ref embedding_dim: !ref dec: !new:speechbrain.nnet.RNN.AttentionalRNNDecoder enc_dim: !ref input_size: !ref rnn_type: gru attn_type: location hidden_size: 1024 attn_dim: 1024 num_layers: 1 scaling: 1.0 channels: 10 kernel_size: 100 re_init: True dropout: !ref ctc_lin: !new:speechbrain.nnet.linear.Linear input_size: !ref n_neurons: !ref seq_lin: !new:speechbrain.nnet.linear.Linear input_size: !ref n_neurons: !ref log_softmax: !new:speechbrain.nnet.activations.Softmax apply_log: True asr_model: !new:torch.nn.ModuleList - [!ref , !ref , !ref , !ref , !ref ] tokenizer: !new:sentencepiece.SentencePieceProcessor # We compose the inference (encoder) pipeline. encoder: !new:speechbrain.nnet.containers.LengthsCapableSequential input_shape: [null, null, !ref ] compute_features: !ref normalize: !ref model: !ref decoder: !new:speechbrain.decoders.S2SRNNBeamSearcher embedding: !ref decoder: !ref linear: !ref bos_index: !ref eos_index: !ref min_decode_ratio: !ref max_decode_ratio: !ref beam_size: !ref eos_threshold: !ref using_max_attn_shift: !ref max_attn_shift: !ref temperature: !ref modules: normalizer: !ref encoder: !ref decoder: !ref pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer loadables: normalizer: !ref asr: !ref tokenizer: !ref