Names, names, names

[keunwoochoi]

• For STFT, unlike in Kapre where I used Spectrogram, I think STFT is a better name - it's more precisely correct.
  • I'm fine with following most of the librosa argument names. Here's librosa one at the moment

    librosa.core.stft(y, n_fft=2048, hop_length=None, win_length=None, window='hann', center=True, dtype=<class 'numpy.complex64'>, pad_mode='reflect')

I personally found hop_length and win_length a bit verbose and prefer n_hop, n_win. But probably 'length of hop' sounds more correct for English speakers than 'number of hop'. The current choice - hop - actually sounds good too.

• For Mel-spectrogram, probably just as it is - Melspectrogram?

Let's keep this thread for discussion names.

[ksanjeevan]

I kinda liked what you had in the gist where STFT as the transform with real and imaginary parts and then depending on what norm is chosen you can get a power or amplitude spectrogram.

On the argument names I also found the _length too verbose. Am completely fine with either n_ or leaving the intuitive ones like hop or pad simple.

Also +1 Melspectrogram

[keunwoochoi]

Ok high-level stuff first. For Layer and Functional, I'll list up tentatively.

• STFT and stft
• Melspectrogram and melspectrogram
• AmplitudeToDb and amplitude_to_db
• DbToAmplitude and db_to_amplitude
  • ewww I wish we could avoid ruining the cases ofdB, but ..decibel seems too verbose. What do you think guys?