Fixed sound threshold level (librosa.effects.split)

圖四 Silence Removal using librosa effect.split function

import librosa

import librosa.display
import numpy as np
import matplotlib.pyplot as plt
import os

import glob
import re
speechFileList = sorted(glob.glob('./C/*.wav'))

for speechFile in speechFileList:
  y, sr = librosa.load(speechFile,sr=None)
  m = re.match(r'\./C/(.+)\.wav', speechFile)
  filename = m.group(1)
  print(filename)
  #file.write(filename + '\n')
  path = 'segment/{}'.format(filename)
  print(path)
  if not os.path.isdir(path):
    os.makedirs(path)
  librosa.display.waveplot(y,sr)
  plt.savefig('{}/waveform_origin.png'.format(path))
  plt.clf()
  m = re.match(r'\./C/(.+)\.wav', speechFile)  
  ts = librosa.effects.split(y,top_db=18, ref=np.max)
  i = 1
  log_file = '{}/log.txt'.format(path)
  with open(log_file,'w') as file:
    for start_i, end_i in ts:
      print('chunk {} in file {}'.format(i, filename))
      file.write('chunk {} in file {}\n'.format(i, filename))
      #print('time: {}s'.format(float(end_i-start_i+1)/sr))
      file.write('time: {}s\n'.format(float(end_i-start_i+1)/sr))
      plt.subplot(len(ts),1,i)
      librosa.display.waveplot(y[start_i:end_i],sr)
      librosa.output.write_wav('{}/segment{}.wav'.format(path,i),y[start_i:end_i],sr)
      i = i+1
  plt.autoscale()
  plt.savefig('{}/waveform.png'.format(path))
  plt.clf()