Fix Issue #471

[tuxzz]

Fix Issue #471

[mthrok]

@vincentqb Some jobs failed at some artifact upload step. Tests are passing, so I think this one is good.

[mthrok]

I did quick experiments and it indeed is faster for long audio;

import time

import torch
import torchaudio.functional
path = 'test/assets/whitenoise_1min.mp3'
audio, sr = torchaudio.load(path)
audio = audio.repeat(1, 10)

kwargs = {
    'n_fft': 12,
    'hop_length': 4,
    'win_length': 12,
    'window': torch.hann_window(12),
    'center': True,
    'pad_mode': 'reflect',
    'normalized': True,
    'onesided': True,
}

stft = torch.stft(audio, **kwargs)
t0 = time.process_time()
torchaudio.functional.istft(stft, **kwargs)
t1 = time.process_time()
print(t1 - t0)

$ git checkout tuxzz-master
Previous HEAD position was d7c79b3 Add test transform and functional AmplitudeToDB (#463)
Switched to branch 'tuxzz-master'
$ python run_istft.py
7.318665638999999
$ git checkout upstream/master
HEAD is now at d7c79b3 Add test transform and functional AmplitudeToDB (#463)
$ python run_istft.py
8.697154317

[vincentqb]

@tuxzz -- Thanks for the PR!
@mthrok -- Thanks for adding the performance comparison!

LGTM