長さの違う音声ファイルのうち、重なっている部分を探すやつ。WIP。

audio_overlay.py

import datetime
import sys

import librosa


def scores(longer, shorter):
    scores = []
    for i in range(len(longer) - len(shorter)):
        print('\r{}/{}\033[K'.format(i, len(longer) - len(shorter)), end='')
        sys.stdout.flush()
        scores.append((i, sum((shorter - longer[i:i+len(shorter)]) ** 2)))
    print('\r\033[K', end='')

    return scores


def find(a, b, sample_rate):
    print('resampling A...')
    a = librosa.resample(a[0], a[1], sample_rate)
    print('resampling B...')
    b = librosa.resample(b[0], b[1], sample_rate)
    print('done')

    print('calcurating scores...')
    ss = scores(a, b)

    for x in ss:
        print(x[0] / sample_rate, ',', x[1], file=sys.stderr)

    ss.sort(key=lambda x: x[1])
    print('done')

    candidates = []
    for shift, score in ss[:10]:
        time = datetime.timedelta(seconds=shift / sample_rate)
        print('{0}: {1}'.format(time, score))
        candidates.append(time)

    return candidates


if __name__ == '__main__':
    print('loading A...')
    a = librosa.load('a.mp3')
    print('done: {}'.format(datetime.timedelta(seconds=len(a[0])/a[1])))

    print('loading B...')
    b = librosa.load('b.mp3')
    print('done: {}'.format(datetime.timedelta(seconds=len(b[0])/b[1])))

    if a[1] != b[1]:
        print('sample rate deferent.')
        print('A: {}, B: {}'.format(a[1], b[1]))
        sys.exit(1)

    if len(a[0]) < len(b[0]):
        print('A is shorter than B.')
        print('A: {}, B: {}'.format(datetime.timedelta(seconds=len(a[0])/a[1]), datetime.timedelta(seconds=len(b[0])/b[1])))
        sys.exit(1)

    find(a, b, 512)