hyperparams.yaml

# ############################################################################
# Model: Small CRDNN model for Voice Activity Detection
# Author:  Mirco Ravanelli, 2021
# ############################################################################

# Feature parameters
sample_rate: 16000
time_resolution: 0.01 # in seconds (e.g,, 0.01 = 10 ms)
n_fft: 400
n_mels: 40

# Model parameters
activation: !name:torch.nn.LeakyReLU
dropout: 0.15
cnn_blocks: 2
cnn_channels: (16, 32)
cnn_kernelsize: (3, 3)
rnn_layers: 2
rnn_neurons: 32
rnn_bidirectional: True
dnn_blocks: 1
dnn_neurons: 16
output_neurons: 1
device: 'cpu' # or 'cuda'

# Feature/Model objects
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 <time_resolution> * 1000 # in ms

mean_var_norm: !new:speechbrain.processing.features.InputNormalization
    norm_type: sentence

cnn: !new:speechbrain.nnet.containers.Sequential
    input_shape: [null, null, !ref <n_mels>]
    norm1: !name:speechbrain.nnet.normalization.LayerNorm
    cnn1: !name:speechbrain.lobes.models.CRDNN.CNN_Block
        channels: 16
        kernel_size: (3, 3) 
    cnn2: !name:speechbrain.lobes.models.CRDNN.CNN_Block
        channels: 32
        kernel_size: (3, 3)

rnn: !new:speechbrain.nnet.RNN.GRU
    input_shape: [null, null, 320]
    hidden_size: !ref <rnn_neurons>
    num_layers: !ref <rnn_layers>
    bidirectional: True

dnn: !new:speechbrain.nnet.containers.Sequential
    input_shape: [null, null, !ref <rnn_neurons> * 2]
    dnn1: !name:speechbrain.lobes.models.CRDNN.DNN_Block
        neurons: !ref <dnn_neurons>
    dnn2: !name:speechbrain.lobes.models.CRDNN.DNN_Block
        neurons: !ref <dnn_neurons>
    lin: !name:speechbrain.nnet.linear.Linear
        n_neurons: !ref <output_neurons>
        bias: False


model: !new:torch.nn.ModuleList
    - [!ref <cnn>, !ref <rnn>, !ref <dnn>]
      
modules:
    compute_features: !ref <compute_features>
    model: !ref <model>
    cnn: !ref <cnn>
    rnn: !ref <rnn>
    dnn: !ref <dnn>
    mean_var_norm: !ref <mean_var_norm>

pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer
    loadables:
        model: !ref <model>
        mean_var_norm: !ref <mean_var_norm>