I noticed that the function inv_mel_spec in inference implements mel2spec conversion based on the Griffin_Lim vocoder, but the parameter spec_from_mel_scaling puzzled me.
Q1:The inverse transformation of mel spectra in the Librosa library does not seem to have this step, why is it needed here?
Q2:I tried to change the spec_from_mel_scaling to a different value, which seems to have an effect on the volume. So why is spec_from_mel_scaling set to 1000 instead of another value?
I noticed that the function inv_mel_spec in inference implements mel2spec conversion based on the Griffin_Lim vocoder, but the parameter spec_from_mel_scaling puzzled me.
Q1:The inverse transformation of mel spectra in the Librosa library does not seem to have this step, why is it needed here?
Q2:I tried to change the spec_from_mel_scaling to a different value, which seems to have an effect on the volume. So why is spec_from_mel_scaling set to 1000 instead of another value?