librispeech initial processing

librispeech.py

import os
import glob
from shutil import copyfile
from subprocess import Popen

base_dir = '/Tmp/sotelo/data/librispeech/'
raw_dir = os.path.join(
    base_dir, 'downloads/LibriSpeech/train-clean-100')
wav_dir = os.path.join(base_dir, 'wav')

audio_files = [
    os.path.join(root, name)
    for root, dirs, files in os.walk(raw_dir)
    for name in files
    if name.endswith((".flac"))]

text_files = [
    os.path.join(root, name)
    for root, dirs, files in os.walk(raw_dir)
    for name in files
    if name.endswith((".txt"))]

audio_files = sorted(audio_files)
text_files = sorted(text_files)

all_text = []
for text_file in text_files:
    with open(text_file, 'r') as f:
        text = f.readlines()
    all_text += text

files_codes = [x.strip().split(" ")[0] for x in all_text]
all_text = [" ".join(x.strip().split(" ")[1:]) for x in all_text]

audio_codes = [x.strip().split("/")[-1].split(".")[0] for x in audio_files]

assert files_codes == audio_codes

proc_txt = []
for file_name, txt in zip(files_codes, all_text):
    proc_txt.append('( ' + file_name + ' "' + txt + '" )\n')

with open(os.path.join(base_dir, 'utts.data'), 'w') as f:
    f.writelines(proc_txt)

convert_cmd = 'ffmpeg -i {} {}'

for audio_code, audio_file in zip(audio_codes, audio_files):
    wav_file = os.path.join(wav_dir, audio_code + '.wav')
    ! {convert_cmd.format(audio_file, wav_file)}