Hi,
I am applying the pyaudio spectrogram to a 512 sample audio sample.
An example below:-
waveform = 512 samples
specgram = torchaudio.transforms.Spectrogram(hop_length=64)(waveform)
I thought this would generate 512/64=8 Hops, so 8 SFTs for the spectrogram, however it generates 9.
waveform = 512 samples
specgram = torchaudio.transforms.Spectrogram(hop_length=128)(waveform)
I thought this would generate 512/128=4 SFTs for the spectrogram, however it generates 5.
I guess it may start at the -hop_length/2 and finishes at 512 + hop_length/2
Pad is set to zero by default. If it is set to 32, it increases the number of SFTs by 1 as expected.