def extract_lpcc_feat(audiopath):
songXlpcc = []
songYlabel = []
audioData, fs = wavread(audiopath)
if "/train/" in audiopath:
lab_path = audiopath.replace("/train/", "/lab/")[:-3] + "lab"
elif "/test/" in audiopath:
lab_path = audiopath.replace("/test/", "/lab/")[:-3] + "lab"
else:
lab_path = audiopath.replace('/valid/', '/lab/')[:-3] + "lab"
lab_file = open(lab_path, 'r')
lab_content = lab_file.readlines()
for segment in lab_content:
list_segment = segment.split(" ")
start = float(list_segment[0])
end = float(list_segment[1])
label = list_segment[2][:-1]
segmentData = audioData[int(start * fs):int(end * fs)]
temp_path = 'tempSegment.wav'
wavwrite(temp_path, segmentData, fs)
segmentLPCC = calcLPCC(temp_path)
os.remove(temp_path)
for lpcc_item in segmentLPCC:
songXlpcc.append(lpcc_item)
songYlabel.append(label)
return songXlpcc, songYlabel
def extract_feat_from_wav(wav_path):
if '/train/' in wav_path:
lab_path = wav_path.replace('/train/', '/lab/')[:-3] + 'lab'
elif '/test/' in wav_path:
lab_path = wav_path.replace('/test/', '/lab/')[:-3] + 'lab'
elif '/valid/' in wav_path:
lab_path = wav_path.replace('/valid/', '/lab/')[:-3] + 'lab'
else:
lab_path = 'null'
label_file = open(lab_path, 'r')
labels = label_file.readlines()
label_file.close()
audioData, fs = wavread(wav_path)
song_mfcc_X = []
song_label_Y = []
for item_label in labels:
startTime, endTime, labelY = item_label.split(' ')
startTime = float(startTime)
endTime = float(endTime)
labelY = labelY[:-1]
audio_part_data = audioData[int(startTime * fs):int(endTime * fs)]
mfcc = calcMFCC(audio_part_data, fs)
for mfcc_item in mfcc:
song_mfcc_X.append(mfcc_item)
song_label_Y.append(labelY)
return song_mfcc_X, song_label_Y
def singing_voice_detection(audio_path):
predict_song_label = []
true_song_label = []
if '/train/' in audio_path:
lab_path = audio_path.replace('/train/', '/lab/')[:-3] + 'lab'
elif '/test/' in audio_path:
lab_path = audio_path.replace('/test/', '/lab/')[:-3] + 'lab'
elif '/valid/' in audio_path:
lab_path = audio_path.replace('/valid/', '/lab/')[:-3] + 'lab'
else:
lab_path = 'null'
label_file = open(lab_path, 'r')
labels = label_file.readlines()
label_file.close()
audioData, fs = wavread(audio_path)
for item_label in labels:
startTime, endTime, labelY = item_label.split(' ')
startTime = float(startTime)
endTime = float(endTime)
labelY = labelY[:-1]
true_song_label.append(labelY)
audio_part_data = audioData[int(startTime * fs):int(endTime * fs)]
segment_mfcc = []
mfcc = calcMFCC(audio_part_data, fs)
for item_mfcc in mfcc:
segment_mfcc.append(item_mfcc)
models = [
'Models/dt0.58.pkl', 'Models/NB0.59.pkl', 'Models/NC0.57.pkl',
'Models/NNP0.61.pkl', 'Models/sgd0.54.pkl'
]
all_pre = []
for model in models:
# print model
clf = joblib.load(model)
predictY = clf.predict(segment_mfcc)
all_pre.append(predictY)
voteRes = voteIt(all_pre)
if voteRes.count('sing') > voteRes.count('nosing'):
segmentLabel = 'sing'
else:
segmentLabel = 'nosing'
predict_song_label.append(segmentLabel)
return predict_song_label, true_song_label
import numpy
from baseZhang import calcMFCC, wavread
from pydub import AudioSegment
lab = '01 - 01 Les Jardins Japonais.lab'
label = open(lab, 'r')
labels = label.readlines()
label.close()
# print labels[0]
startTime, endTime, labelY = labels[0].split(' ')
startTime = float(startTime)
endTime = float(endTime)
labelY = labelY[:-1]
# print labelY
song = AudioSegment.from_file(lab.replace('.lab', '.ogg'))
song.set_channels(1)
song = song.export(lab.replace('.lab', '.wav'), 'wav')
audioData, fs = wavread(lab.replace('.lab', '.wav'))
# print len(audioData)
# print fs
part1 = audioData[int(startTime * fs):int(endTime * fs)]
mfcc = calcMFCC(part1, fs)
print numpy.shape(mfcc)