def realtime_detect(af, af_url, rf):
if not af: # no local file, download from url
# y, sr = sf.read(io.BytesIO(urlopen(af_url).read()))
af = 'audio.mp3'
with open(af, 'wb') as audio:
audio.write(urlopen(af_url).read())
y, sr = librosa.load(af)
start = 0
step = int(sr * 1)
feat = []
pos = 0
total = 0
feat = [feature_extraction.feature(y, sr)]
result = detect(rf, feat)
return result
def realtime_detect(af, af_url, rf):
if not af: # no local file, download from url
# y, sr = sf.read(io.BytesIO(urlopen(af_url).read()))
af = '/opt/audio_process/audio.mp3'
audio_f = urlopen(af_url, timeout=5)
if not audio_f:
return "url request timeout"
with open(af, 'wb') as audio:
audio.write(audio_f.read())
y, sr = librosa.load(af)
raw_feat = feature_extraction.feature(y, sr=sr)
feat = raw_feat
feat = feat.reshape(1, -1)
result = detect(rf, feat)
return result
def realtime_detect(af, af_url, model):
if not af: # no local file, download from url
# y, sr = sf.read(io.BytesIO(urlopen(af_url).read()))
af = 'audio.mp3'
with open(af, 'wb') as audio:
audio.write(urlopen(af_url).read())
try:
y, sr = librosa.load(af)
except Exception as e:
return e
raw_feat = feature_extraction.feature(y, sr)
feat = np.r_[raw_feat[1], raw_feat[3:6], raw_feat[12:14], raw_feat[16:18],
raw_feat[24:26], raw_feat[27:30]]
# reshape array for single sample
feat = feat.reshape(1, -1)
result = detect(model, feat)
return result
def calc_feature(date_str):
audio_list = glob(AUDIO_PATH + date_str + '*.mp3')
feature_file = os.path.join(os.path.dirname(feature_extraction.__file__), 'feature_data/neg_' + date_str + '.log')
print "saving features to {} ...".format(feature_file)
feat = {}
if len(audio_list) > 1:
# sample the audio_list
if len(audio_list) > 1500:
audio_list = random.sample(audio_list, 1500)
with open(feature_file, 'w') as f_file:
for f in audio_list:
print("processing: {}".format(f))
try:
y, sr = librosa.load(f)
feat['feature'] = list(feature_extraction.feature(y, sr=sr))
except Exception as e:
print(e)
continue
feat_json = json.dumps(feat)
f_file.write(feat_json+'\n')
import os
import librosa
import feature_extraction
import svm
from svmutil import *
features = []
labels = []
for file_name in os.listdir('test/'):
y, sr = librosa.load('test/'+file_name)
feat = feature_extraction.feature(y, sr=sr)
feat = feat.tolist()
features.append(feat)
labels.append(-1)
model = svm_load_model('audio.model')
pre, acc, val = svm_predict(labels, features, model)
print('accuracy: ', acc)
print(pre)
arg 1: audio folder
arg 2: feature filename
'''
if __name__ == "__main__":
ff = []
try:
if os.path.isdir(sys.argv[1]):
# ff = [os.path.join(sys.argv[1], f) for f in os.listdir(sys.argv[1]) if f.split('.')[1]=='mp3']
ff = [os.path.join(sys.argv[1], f) for f in os.listdir(sys.argv[1])]
feature_file = 'feature_data/' + sys.argv[2]
except Exception as e:
print(e)
feat = {}
# sample the audio-list
if len(ff) > 1500:
ff = random.sample(ff, 1500)
with open(feature_file, 'w') as f_file:
for f in ff:
print('processing: {}'.format(f))
y, sr = librosa.load(f)
try:
feat['feature'] = list(feature_extraction.feature(y, sr=sr))
except Exception as e:
print(e)
continue
feat_json = json.dumps(feat)
f_file.write(feat_json+'\n')
print('saved feature array to ' + feature_file)