LibriSpeech Annotation Corrector

libri_annotator.py

from glob import glob
from playsound import playsound

from multiprocessing import Process
import os, sys, time

def open_text_editor(txt_file):
    # Only Windows 😎
    # Probably use gedit for Ubuntu?
    os.system('notepad++ %s' % txt_file)
    return

if __name__ == '__main__':
    audio_files = sorted(glob('wav/*.wav'))
    if not audio_files:
        sys.exit('No audio files found.')
    # i = 0
    i = int(input('Start annotating from index: '))

    while i < len(audio_files):
        audio_file = audio_files[i]
        audio_player = Process(target=playsound, args=(audio_file, ))
        txt_file = 'txt/' + os.path.basename(audio_file).replace('.wav', '.txt')
        if not os.path.isfile(txt_file):
            print('No txt for:', audio_file)
            continue
        print('>> Current:', i)
        text_editor = Process(target=open_text_editor, args=(txt_file,))
        
        text_editor.start()
        time.sleep(0.5)
        audio_player.start()
        
        while True:
            command = input('Command: ')
            audio_player.terminate()
            if command == 'n': # Next audio
                i += 1
                break
            if command == 'r': # Replay audio
                audio_player = Process(target=playsound, args=(audio_file, ))
                audio_player.start()
                continue
            if command == 'p': # Previous audio
                if i > 0:
                    i -= 1
                break
            if command == 'q': # Quit Annotating
                print('Quitting... Remember: Last Audio Index:', i)
                sys.exit()
        # text_editor.terminate()
    print('SUCCESS: Completed Annotating %d samples of data!!!' % len(audio_files))

LibriSpeech_Annotator.md

Steps:

1. Place this script inside your librispeech dataset folder.
2. This should the directory structure of the dataset folder:

libri_dataset
|-libri_annotator.py
|-txt
 |- sample1.txt
 |- sample2.txt
|-wav
 |-sample1.wav
 |-sample2.wav

3. Ensure Notepad++ is installed and set in PATH
4. pip install playsound
5. Run python libri_annotator.py from inside the libri_dataset folder.

How to use:

• Press n for next sample
• Press p for previous sample
• Press r for replay
• Press q to quit

Tips:

• Dock the Notepad++ window to the right-side and the command-line to the left side (for ease of annotation)
• For Linux, you might have to use something like gedit instead of notepad++