If I use librosa.feature.melspectrogram to extract melspectrogram from the raw wav , the result is quite different from use extract_mel_spectrogram_for_tts in datapipeline.py .
such as using same setting and wav data:
mel_spec1 = extract_mel_spectrogram_for_tts( ... )
mel_spec2 = librosa.feature.melspectrogram( ... )
while the result in a random example:
mel_spec1.max=0.08053161389832361
mel_spec1.min=0.0
mel_spec1.mean=0.02656768943313912
mel_spec1.std=0.021109905037463596
mel_spec2.max=209.5188483174194
mel_spec2.min=0.0
mel_spec2.mean=0.25262605107753267
mel_spec2.std=2.8514387417932685
Could someone help me to explain the reason of this difference ?
If I use librosa.feature.melspectrogram to extract melspectrogram from the raw wav , the result is quite different from use extract_mel_spectrogram_for_tts in datapipeline.py .
such as using same setting and wav data:
mel_spec1 = extract_mel_spectrogram_for_tts( ... )
mel_spec2 = librosa.feature.melspectrogram( ... )
while the result in a random example:
mel_spec1.max=0.08053161389832361
mel_spec1.min=0.0
mel_spec1.mean=0.02656768943313912
mel_spec1.std=0.021109905037463596
mel_spec2.max=209.5188483174194
mel_spec2.min=0.0
mel_spec2.mean=0.25262605107753267
mel_spec2.std=2.8514387417932685
Could someone help me to explain the reason of this difference ?