Why n_fft=400 by default in Transforms?

In MelSpectrogram, Spectrogram, GriffinLim, n_fft defaults to 400. Is there a reason for not setting it with a power of 2?

[vincentqb]

I will close this issue for now. Please feel free to re-open if there are more elements you would like to add to the discussion 😃

[vincentqb]

Quick run, and I don’t see significant differences 😃

In [1]: f = "steam-train-whistle-daniel_simon.wav"                                                                                                                                                          

In [2]: import torchaudio                                                                                                                                                                                   

In [3]: w, s = torchaudio.load(f)                                                                                                                                                                           

In [4]: %timeit torchaudio.transforms.Spectrogram(n_fft=400)(w)                                                                                                                                             
10.5 ms ± 376 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

In [5]: %timeit torchaudio.transforms.Spectrogram(n_fft=512)(w)                                                                                                                                             
11.1 ms ± 207 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)