Hi,
So I initialize my melspectrogram as follows:
transform = torchaudio.transforms.MelSpectrogram(sample_rate=8000, n_mels=80, win_length=200, hop_length=80, center=False)
Then here’s how I use it:
x_in.shape == [1,5360]
x_out = transform(x_in)
x_out.shape == [1, 80, 63]
However, based on my (introductory) understanding of Fourier Transform, I thought the output length is supposed to be (input_length - win_length) / hop_length + 1 = (5360 - 200)./ 80 + 1 = 65.5
So shouldn’t be either 65 or 66, instead of 63? Or do I just not understand FT?
Thanks,
Kevin