Where is Basis-MelGAN-dataset folder? about convtasnet4basismelgan HOT 1 CLOSED

import torch

import os, sys
import audio_torch as audio
import random
import numpy as np
import hparams as hp
import glob
from tqdm import tqdm
from functools import partial
from multiprocessing import cpu_count
from concurrent.futures import ProcessPoolExecutor

import warnings
warnings.filterwarnings('ignore')

DATASET = "aishell3"  # aishell3 or biaobei


def _process_utterance(in_path, out_path, index):
    wav = torch.Tensor(audio.load_wav(in_path)) #, hp.sample_rate, encode=False))
    noi = audio.add_noise(wav, quantization_channel=hp.quantization_channel)
    mix = wav.float() + noi

    wav_name = f"{index}.wav.npy"
    noi_name = f"{index}.noi.npy"
    mix_name = f"{index}.mix.npy"

    if wav.size(0) > 44100:
        np.save(os.path.join(out_path, wav_name), wav.numpy(), allow_pickle=False)
        np.save(os.path.join(out_path, noi_name), noi.numpy(), allow_pickle=False)
        np.save(os.path.join(out_path, mix_name), mix.numpy(), allow_pickle=False)

if __name__ == "__main__":

    if len(sys.argv) == 1:
        wav_path = "data_wav_path"
    else:
        wav_path = sys.argv[1]

    max_workers = int(cpu_count()/4)+1
    print(max_workers)
    executor = ProcessPoolExecutor(max_workers=max_workers)
    futures = []
    os.makedirs(hp.dataset_path, exist_ok=True)
        
    paths = glob.glob(wav_path+"/*")
    length = len(paths)
    for i in tqdm(range(length)):
        path = paths[i]
        index = path.split("/")[-1]
        futures.append(executor.submit(partial(_process_utterance, os.path.join(path), hp.dataset_path, index)))
    [future.result() for future in tqdm(futures)]

    with open("dataset.txt", "w+", encoding="utf-8") as f:
        for filename in glob.glob("dataset/*.wav.mix.npy"):
            f.write(wav_path + "/"+ filename.split("/")[-1].replace(".mix.npy","")+"\n")