def testLabelWav(self):
tmp_dir = self.get_temp_dir()
wav_data = self._getWavData()
wav_filename = os.path.join(tmp_dir, "wav_file.wav")
self._saveTestWavFile(wav_filename, wav_data)
input_name = "test_input"
output_name = "test_output"
graph_filename = os.path.join(tmp_dir, "test_graph.pb")
with tf.Session() as sess:
tf.placeholder(tf.string, name=input_name)
tf.zeros([1, 3], name=output_name)
with open(graph_filename, "wb") as f:
f.write(sess.graph.as_graph_def().SerializeToString())
labels_filename = os.path.join(tmp_dir, "test_labels.txt")
with open(labels_filename, "w") as f:
f.write("a\nb\nc\n")
label_wav.label_wav(wav_filename, labels_filename, graph_filename,
input_name + ":0", output_name + ":0", 3)
def get_result(self, filename):
# graph_name = './866_895.pb'
# labels_name = './conv_labels.txt'
wav = filename
input_name = 'wav_data:0'
output_name = 'labels_softmax:0'
how_many_labels = 1
result_dict = label_wav.label_wav(wav, self.labels, self.graph,
input_name, output_name,
how_many_labels)
return result_dict
def infer(wav):
## wav should be a name in string form that is under recorder
## 将需要识别结果的 wav 文件放到 recorder 里面
## 要求 必须是 无压缩 16000Hz 的单英文单词的 1s音频
## e.g. wav = '0b40aa8e_nohash_0.wav'
labels = nlp_path + '/speech_commands_train/conv_labels.txt' #xxx改变label生成位置
graph = nlp_path + '/graph/my_frozen_graph.pb' #freeze里面改变graph生成位置
input_name = 'wav_data:0'
output_name = 'labels_softmax:0'
how_many_labels = 1
wav = nlp_path + "/recorder/" + wav
return label_wav.label_wav(wav, labels, graph, input_name, output_name, how_many_labels)
def recog(request):
sound = AudioSegment.from_wav("documents/theRecog.wav")
sound = sound.set_frame_rate(16000)
sound.export("documents/theNewRecog.wav", format="wav")
result = label_wav(
wav='documents/theNewRecog.wav',
graph='my_frozen_graph.pb',
labels='conv_labels.txt',
input_name='wav_data:0',
output_name='labels_softmax:0',
how_many_labels=3,
)
notResult = ",".join(result)
strResult = "\"" + notResult + "\""
print(strResult)
return render(request, 'recorder.html', {'recogResult': strResult})
def recog(request):
sound = AudioSegment.from_wav("documents/the_" + ctx['rlt'] + "_Recog.wav")
sound = sound.set_frame_rate(16000)
sound.export("documents/the_" + ctx['rlt'] + "_NewRecog.wav", format="wav")
result = label_wav(
wav='documents/the_' + ctx['rlt'] + '_NewRecog.wav',
graph='my_frozen_graph.pb',
labels='conv_labels.txt',
input_name='wav_data:0',
output_name='labels_softmax:0',
how_many_labels=3,
)
notResult = ",".join(result)
strResult = "\"" + notResult + "\""
print(strResult)
return HttpResponse(strResult)
def post(self):
audio_webm = request.files['blob']
audio_webm.save(os.path.join(app.config['UPLOAD_FOLDER'], str(timestamp)+".webm"))
try:
if(get_chunk(str(timestamp)+".webm")=="silence"):
raise FileExistsError
else:
top = label_wav("uploads/chunk0.wav", "sp_train/conv_labels.txt", "model/my_frozen_graph.pb", "wav_data:0", "labels_softmax:0", 4)
print(top)
return make_response(jsonify(
top
), 200)
except FileExistsError:
return make_response(jsonify(
"silence"
), 200)
def listen_label(graph, labels, wav_file_path):
#listen for 3s
RECORDING = True
while (RECORDING):
inp = input("standby: ")
if inp == 'r':
print("\n*listening")
frames = []
#open stream
stream = p.open(format=FORMAT,
channels=CHANNELS,
rate=RATE,
input=True,
frames_per_buffer=CHUNK)
for i in range(0, int(RATE / CHUNK * RECORD_SECONDS)):
data = stream.read(CHUNK)
frames.append(data)
print("the chord is: ")
stream.stop_stream()
stream.close()
#write out file
WAVE_OUTPUT_FILENAME = "temp.wav"
wf = wave.open(WAVE_OUTPUT_FILENAME, 'wb')
wf.setnchannels(CHANNELS)
wf.setsampwidth(p.get_sample_size(FORMAT))
wf.setframerate(RATE)
wf.writeframes(b''.join(frames))
wf.close()
chord = lw.label_wav(wav_file_path, labels, graph, 'wav_data:0',
'labels_softmax:0', 1)
return chord
elif inp == 'q':
print("\n*exiting")
RECORDING = False
p.terminate()
def check_new_files():
while True:
next_wav_id = r.lpop('wav_id')
if not next_wav_id:
return
next_file = f'/data/{next_wav_id}.wav'
logger.info(f'processing {next_file}')
result, error = None, None
try:
result = label_wav.label_wav(next_file)
logger.info('result from label_wav: ' + str(result))
except Exception as e:
error = e
logger.info('error from label_wav: ' + str(error))
continue
for k, v in result.items():
label = k
label_prob = v
if label_prob > LABEL_PROB_THRESHOLD:
notifications.push_police_notification({
'wav_id': next_wav_id,
'label': label,
'label_prob': str(label_prob),
})
def listen(graph, labels, wav_file_path):
frames = []
stream = p.open(format=FORMAT,
channels=CHANNELS,
rate=RATE,
input=True,
frames_per_buffer=CHUNK)
for i in range(0, int(RATE / CHUNK * RECORD_SECONDS)):
data = stream.read(CHUNK)
frames.append(data)
stream.stop_stream()
stream.close()
WAVE_OUTPUT_FILENAME = "temp.wav"
wf = wave.open(WAVE_OUTPUT_FILENAME, 'wb')
wf.setnchannels(CHANNELS)
wf.setsampwidth(p.get_sample_size(FORMAT))
wf.setframerate(RATE)
wf.writeframes(b''.join(frames))
wf.close()
chord = lw.label_wav(wav_file_path, labels, graph, 'wav_data:0',
'labels_softmax:0', 1)
return str(chord)
