audio load & resampling

# waveform = torch.from_numpy(waveform)   ----> numpy array를 torch tensor로 변환하는 것은 속도에 영향이 거의 없다.

# sr = 22050이면 --> torchaudio가 많이 느리다.
# sr = 16000  ---> origin sample_rate과 일치하면 모두 다 빠르다.
# sr = 8000

n_samples = 100
sr = 22050
s_time=time.time()
for i in range(n_samples):
    waveform,sr = librosa.load(A[i],sr=sr,res_type='kaiser_best')
print('librosa(kaiser_best): ',time.time()-s_time)


s_time=time.time()
for i in range(n_samples):
    waveform,sr = librosa.load(A[i],sr=sr,res_type='kaiser_fast')
print('librosa(kaiser_fast): ',time.time()-s_time)


s_time=time.time()
for i in range(n_samples):
    waveform,sr = librosa.load(A[i],sr=sr,res_type='polyphase')
print('librosa(polyphase): ',time.time()-s_time)



transform = torchaudio.transforms.Resample(orig_freq=16000, new_freq=sr)
s_time=time.time()
for i in range(n_samples):
    waveform,sr = torchaudio.load(A[i])
    waveform = transform(waveform)
print('torchaudio: ',time.time()-s_time)