I’m trying to use the get_melspectrogram_layer from the Kapre library in a Keras model. However, I’m encountering the following error:

ValueError: Exception encountered when calling STFT.call().
Invalid dtype: complex64

Here’s the code I’m using:

input_shape = (int(16000*1.0), 1)

mel_spectogram_layer = get_melspectrogram_layer(input_shape=input_shape,
                              n_mels=128,
                              pad_end=True,
                              n_fft=512,
                              win_length=400,
                              hop_length=160,
                              sample_rate=16000,
                              return_decibel=True,
                              input_data_format='channels_last',
                              output_data_format='channels_last')

The error traceback points to an issue with the complex64 dtype:

Invalid dtype: complex64

Arguments received by STFT.call():
  • args=('<KerasTensor shape=(None, 16000, 1), dtype=float32, sparse=None, name=keras_tensor_13>',)
  • kwargs=<class 'inspect._empty'>

I have tried casting the output to float32 after the Mel-spectrogram layer, but the error persists. How can I resolve this issue? Is there a way to handle the complex64 dtype within the STFT or Mel-spectrogram layers in Kapre?

Any help would be appreciated!

I wanted to use the kapre library to preprocess the audio files into melspectograms and pass it to the CNN model that I have written