I have a basic and probably stupid question. What is the point of using a WaveNet Vocoder to invert the mel spectrogram feature representation into time-domain waveform since we can just use audio.inv_mel_spectrogram and get the audio (voice output) directly from the encoder part?
I have a basic and probably stupid question. What is the point of using a WaveNet Vocoder to invert the mel spectrogram feature representation into time-domain waveform since we can just use audio.inv_mel_spectrogram and get the audio (voice output) directly from the encoder part?