Skip to content

Instantly share code, notes, and snippets.

@vivek081166
Last active April 9, 2019 08:41
Show Gist options
  • Save vivek081166/5c9178610d7ee657d1134efd2f28359d to your computer and use it in GitHub Desktop.
Save vivek081166/5c9178610d7ee657d1134efd2f28359d to your computer and use it in GitHub Desktop.
import os
import pickle
from glob import iglob
import numpy as np
import librosa
DATA_AUDIO_DIR = './audio'
TARGET_SR = 8000
OUTPUT_DIR = './output'
OUTPUT_DIR_TRAIN = os.path.join(OUTPUT_DIR, 'train')
OUTPUT_DIR_TEST = os.path.join(OUTPUT_DIR, 'test')
AUDIO_LENGTH = 10000
class_ids = {
'normal': 0,
'murmur': 1,
'extrahls': 2,
'artifact': 3,
'unlabelled': 4,
}
def extract_class_id(wav_filename):
if 'normal' in wav_filename:
return class_ids.get('normal')
elif 'murmur' in wav_filename:
return class_ids.get('murmur')
elif 'extrahls' in wav_filename:
return class_ids.get('extrahls')
elif 'artifact' in wav_filename:
return class_ids.get('artifact')
elif 'unlabelled' in wav_filename:
return class_ids.get('unlabelled')
else:
return class_ids.get('unlabelled')
def read_audio_from_filename(filename, target_sr):
audio, _ = librosa.load(filename, sr=target_sr, mono=True)
audio = audio.reshape(-1, 1)
return audio
def convert_data():
for i, wav_filename in enumerate(iglob(os.path.join(DATA_AUDIO_DIR, '**/**.wav'), recursive=True)):
class_id = extract_class_id(wav_filename)
audio_buf = read_audio_from_filename(wav_filename, target_sr=TARGET_SR)
# normalize mean 0, variance 1
audio_buf = (audio_buf - np.mean(audio_buf)) / np.std(audio_buf)
original_length = len(audio_buf)
print(i, wav_filename, original_length, np.round(np.mean(audio_buf), 4), np.std(audio_buf))
if original_length < AUDIO_LENGTH:
audio_buf = np.concatenate((audio_buf, np.zeros(shape=(AUDIO_LENGTH - original_length, 1))))
print('PAD New length =', len(audio_buf))
elif original_length > AUDIO_LENGTH:
audio_buf = audio_buf[0:AUDIO_LENGTH]
print('CUT New length =', len(audio_buf))
output_folder = OUTPUT_DIR_TRAIN
if i // 50 == 0:
output_folder = OUTPUT_DIR_TEST
output_filename = os.path.join(output_folder, str(i) + '.pkl')
out = {'class_id': class_id,
'audio': audio_buf,
'sr': TARGET_SR}
with open(output_filename, 'wb') as w:
pickle.dump(out, w)
if __name__ == '__main__':
convert_data()
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment