downsampling.py

import json
import os
from os.path import join as opj
from tqdm import tqdm
import sys
import torchaudio

downsample_rate = 16000

clean_train_path = 'original clean train path'
noisy_train_path = 'original noisy train path'

clean_test_path  = 'original clean test path'
noisy_test_path  = 'original noisy test path'

resample_path    = 'resample path'

def downsample_wave(clean_path, noisy_path, sample_rate=downsample_rate, phase='train'):
    
    clean_list = os.listdir(clean_path)
    noisy_list = os.listdir(noisy_path)
    
    clean_list.sort(); noisy_list.sort()
    
    clean_resample_path = opj(resample_path, 'clean_'+phase)
    noisy_resample_path = opj(resample_path, 'noisy_'+phase)
    
    if not os.path.isdir(clean_resample_path):
        os.makedirs(clean_resample_path)
        
    if not os.path.isdir(noisy_resample_path):
        os.makedirs(noisy_resample_path)

    tf = torchaudio.transforms.Resample(48000,sample_rate)
    
    for i in tqdm(range(len(clean_list))):
        
        clean, sr = torchaudio.load(opj(clean_path, clean_list[i]))
        noisy, sr = torchaudio.load(opj(noisy_path, noisy_list[i]))
        
        clean = tf(clean)
        noisy = tf(noisy)

        torchaudio.save(opj(clean_resample_path, clean_list[i]), clean, sample_rate)
        torchaudio.save(opj(noisy_resample_path, noisy_list[i]), noisy, sample_rate)

if __name__ == "__main__":
    print('--- Trainset Resampling ---')
    downsample_wave(clean_train_path, noisy_train_path, sample_rate=downsample_rate, phase='train')
    print('--- Testset Resampling ---')
    downsample_wave(clean_test_path, noisy_test_path, sample_rate=downsample_rate, phase='test')