waveファイルを無音区間で分割して吐き出すやつ

wave_split.py

from struct import unpack
from scipy.io import wavfile
import scipy.io
import numpy as np
import os, sys
import glob


threshold = 0
silence_min_frames = 1024

def split_wav(wavfilepath, output_dir):
    print("-----------------------------------------")
    sr, rdata = wavfile.read(wavfilepath, mmap=True)
    n_ch = rdata.shape[1]
    assert(n_ch == 2)
    n_frames = rdata.shape[0]
    length_sec = n_frames / sr
    print(f"{wavfilepath} | Fs = {sr} | ch = {n_ch} | frames = {n_frames} | length = {length_sec}")

    silence_area = []

    def is_silence(rdata, idx):
        return rdata[idx][0] <= threshold and rdata[idx][1] <= threshold

    # 簡単化のため、先頭と末尾に十分な量の無音区間を追加
    zero_area = np.zeros((silence_min_frames,2), dtype = rdata.dtype)
    data = np.vstack((zero_area, rdata))
    data = np.vstack((data, zero_area))
    n_frames += silence_min_frames * 2

    print("START")

    # データは必ず無音区間から始まる
    under_silence = True
    silence_start = 0

    # 無音区間の検出
    for i in range(1, n_frames):
        if is_silence(data, i):
            if under_silence:
                # 無音継続中
                pass
            else:
                # 無音区間開始
                under_silence = True
                silence_start = i
        else:
            if under_silence:
                # 無音区間終了
                silence_end = i - 1
                if silence_end - silence_start >= silence_min_frames:
                    silence_area.append( (silence_start, silence_end) )
                    print(f"silence_area : {silence_start} to {silence_end}")
                silence_start = -1
                under_silence = False
            else:
                # 音声区間継続
                pass
    # 最後は必ず無音区間
    silence_area.append( (silence_start, n_frames-1) )
    print(f"silence_area : {silence_start} to {n_frames-1}")

    # チェック
    if len(silence_area) == 1:
        raise RuntimeError("This is silence file") # これは完全な無音ファイル
    if len(silence_area) == 2:
        raise RuntimeError("This file does not need to be split") # そもそも分割する必要がないファイル
    
    # 音声区間に変換
    audio_area = []
    for i in range(0, len(silence_area)-1):
        audio_start = silence_area[i][1]+1
        audio_end = silence_area[i+1][0]-1
        audio_area.append( (audio_start, audio_end) )
        # print(f"audio_area #{i} : {audio_start} to {audio_end}")

    # 出力
    for i, area in enumerate(audio_area):
        name,ext = os.path.splitext( os.path.basename(wavfilepath) )
        output_fname = f"{output_dir}/{name} part{i+1:0=2}{ext}"
        wavfile.write(output_fname, sr, data[area[0]:area[1]+1])
        print(f"output : {output_fname} | {area[0]} to {area[1]}")

    print("END")
    print("-----------------------------------------")



def main():
    target_dir = sys.argv[1]
    output_dir = f"{target_dir}/split_data"
    os.makedirs(output_dir, exist_ok=True)
    for path in glob.glob(f"{target_dir}/*.wav"):
        split_wav(path, output_dir)

main()