An example of a neural network that uses a custom LogMelSpectrogram layer

model.py

def ConvModel(n_classes, sample_rate=16000, duration=4,
              fft_size=_FFT_SIZE, hop_size=_HOP_SIZE, n_mels=_N_MEL_BINS):
    n_samples = sample_rate * duration
    
    # Accept raw audio data as input
    x = Input(shape=(n_samples,), name='input', dtype='float32')
    # Process into log-mel-spectrograms. (This is your custom layer!)
    y = LogMelSpectrogram(sample_rate, fft_size, hop_size, n_mels)(x)
    # Normalize data (on frequency axis)
    y = BatchNormalization(axis=2)(y)
    
    y = Conv2D(32, (3, n_mels), activation='relu')(y)
    y = BatchNormalization()(y)
    y = MaxPool2D((1, y.shape[2]))(y)

    y = Conv2D(32, (3, 1), activation='relu')(y)
    y = BatchNormalization()(y)
    y = MaxPool2D(pool_size=(2, 1))(y)

    y = Flatten()(y)
    y = Dense(64, activation='relu')(y)
    y = Dropout(0.25)(y)
    y = Dense(n_classes, activation='softmax')(y)

    return Model(inputs=x, outputs=y)


model = ConvModel(11)
model.compile(optimizer='adam', 
              loss='sparse_categorical_crossentropy', 
              metrics=['sparse_categorical_accuracy'])
model.summary()


# Model: "model"
# _________________________________________________________________
# Layer (type)                 Output Shape              Param #   
# =================================================================
# input (InputLayer)           [(None, 64000)]           0         
# _________________________________________________________________
# log_mel_spectrogram_4 (LogMe (None, 124, 64, 1)        0         
# _________________________________________________________________
# batch_normalization (BatchNo (None, 124, 64, 1)        256       
# _________________________________________________________________
# conv2d (Conv2D)              (None, 122, 1, 32)        6176      
# _________________________________________________________________
# batch_normalization_1 (Batch (None, 122, 1, 32)        128       
# _________________________________________________________________
# max_pooling2d (MaxPooling2D) (None, 122, 1, 32)        0         
# _________________________________________________________________
# conv2d_1 (Conv2D)            (None, 120, 1, 32)        3104      
# _________________________________________________________________
# batch_normalization_2 (Batch (None, 120, 1, 32)        128       
# _________________________________________________________________
# max_pooling2d_1 (MaxPooling2 (None, 60, 1, 32)         0         
# _________________________________________________________________
# flatten (Flatten)            (None, 1920)              0         
# _________________________________________________________________
# dense (Dense)                (None, 64)                122944    
# _________________________________________________________________
# dropout (Dropout)            (None, 64)                0         
# _________________________________________________________________
# dense_1 (Dense)              (None, 11)                715       
# =================================================================
# Total params: 133,451
# Trainable params: 133,195
# Non-trainable params: 256
# _________________________________________________________________