def train():
"""
trains data on our model
:return: no return value
"""
speakers_file = "./speakers.txt"
try:
os.remove(speakers_file)
except OSError as e:
print("Error: " + str(e))
for f in os.listdir('./'):
if re.search('marf*', f):
os.remove(os.path.join('./', f))
name_list = speakers
folder_path = './train_data/'
recog = recognition.SpeakerRecognizer()
recog.debug = True
# setting classifier
recog.set_feature_option('-noise -raw -aggr -eucl')
for i in range(len(name_list)):
recog.speaker_name = name_list[i]
path = os.path.join(folder_path, name_list[i])
print("Speaker Name: {} speaker_data: {}".format(
name_list[i], name_list[i]))
recog.train_new_data(path)
def trainPiWho(self):
# TODO fix piwho and remove this
if False and (self.getPiWhoTrainingDataLastModified() > 0
and self.isPiWhoTrainingDataModified()):
print('start training')
recog = recognition.SpeakerRecognizer()
print('got recog')
try:
os.remove(self.piwho_speakers)
except:
pass
print('rm speakers')
for f in os.listdir(self.piwho_model):
if os.path.isfile(f) and "marf" in f and ".gzbin" in f:
os.remove(os.path.join(self.piwho_model, f))
print('rm marf')
print("training folder has {}".format(
os.listdir(self.piwho_training_data)))
for d in os.listdir(self.piwho_training_data):
if os.path.isdir(os.path.join(self.piwho_training_data, d)):
print("train {}".format(d))
recog.speaker_name = d
os.chdir(self.piwho_model)
recog.train_new_data(
os.path.join(self.piwho_training_data, d), d)
print('trained marf for {}'.format(d))
print('trained marf')
def find_speaker(flag):
recog = recognition.SpeakerRecognizer()
# Record voice until silence is detected
# save WAV file
#vad.record()
record(5, 'test.wav')
# use newly recorded file for recognition
name = []
name = recog.identify_speaker('test.wav')
dictn = recog.get_speaker_scores()
print(name[0])
print(dictn)
if flag == 1:
if float(dictn[name[0]]) < 0.5:
print('Congratulation !!! with distance ' + dictn[name[0]])
return "CONGRATULATIONS!"
else:
print('Sorry... with distance ' + dictn[name[0]])
return "Please retry."
elif flag == -1:
if float(dictn[name[0]]) < 0.4:
print('Congratulation !!! with distance ' + dictn[name[0]])
return "CONGRATULATIONS!"
else:
print('Sorry... with distance ' + dictn[name[0]])
return "Please retry."
def find_speaker():
r = sr.Recognizer()
with sr.Microphone() as source:
r.adjust_for_ambient_noise(source)
os.system("beep")
print("Speaker needs to be identified. Say something..")
audio = r.listen(source)
s = "results.wav"
with open(s, "wb") as f:
f.write(audio.get_wav_data())
recog = recognition.SpeakerRecognizer()
name = []
name = recog.identify_speaker('results.wav')
dictn = recog.get_speaker_scores()
i = float(dictn.get(name[0]))
#print("Voice close to "+name[0])
if (i > 0.55):
print("Voice not secure enough to proceed")
os.system("streamer -f jpeg -o INTRUDER.jpeg")
os.system(
"espeak -s 150 'Intruder alert , Intruder alert , Intruder alert .'"
)
exit(0)
alert("Hello there !!")
def test_create_database_entry(self):
recog = recognition.SpeakerRecognizer()
recog._create_entry('Test', 'Test.wav')
self.assertTrue(os.path.isfile('./speakers.txt'))
try:
os.remove('./speakers.txt')
except IOError:
raise
def train():
recog = recognition.SpeakerRecognizer()
recog.debug = True
for i in range(len(speakers)):
recog.speaker_name = speakers[i]
path = os.path.join(folder_path, data_folders[i])
print("speaker name: {} speaker_data: {}".format(
speakers[i], data_folders[i]))
recog.train_new_data(path)
def test_directory_as_constructor_argument(self):
recog = recognition.SpeakerRecognizer(DATA_DIR_PATH)
self.assertEqual(recog.dirpath, os.path.abspath(DATA_DIR_PATH))
self.assertEqual(recog.debug, False)
self.assertEqual(recog.filepath, None)
self.assertEqual(recog.feature, ' -endp -lpc -cheb')
self.assertEqual(recog.last_recognized_file, None)
self.assertEqual(recog.last_trained_file, None)
def train_speaker(name, *args):
recog = recognition.SpeakerRecognizer('{0}{1}'.format(model_path_dir,name))
recog.speaker_name = name
if(len(args)==1):
recog.train_new_data(args[0])
else:
recog.train_new_data()
print recog.get_speakers()
def test_default_constructor(self):
recog = recognition.SpeakerRecognizer()
self.assertEqual(recog.dirpath, os.getcwd())
self.assertEqual(recog.debug, False)
self.assertEqual(recog.filepath, None)
self.assertEqual(recog.feature, ' -endp -lpc -cheb')
self.assertEqual(recog.last_recognized_file, None)
self.assertEqual(recog.last_trained_file, None)
def test_correct_entry_in_databse_file(self):
recog = recognition.SpeakerRecognizer()
recog._create_entry('Test', 'ABC.wav')
entry = '0,Test,ABC.wav|\n'
with open('./speakers.txt', 'r') as reader:
self.assertEqual(entry, reader.readline())
try:
os.remove('./speakers.txt')
except IOError:
raise
def test_convert_file_into_8khz(self):
recog = recognition.SpeakerRecognizer()
recog._convert_file(os.path.join(DATA_DIR_PATH, 'arctic_a0001.wav'),
os.path.join(DATA_DIR_PATH, 'converted.wav'))
self.assertTrue(
os.path.isfile(os.path.join(DATA_DIR_PATH, 'converted.wav')))
try:
os.remove(os.path.join(DATA_DIR_PATH, 'converted.wav'))
except IOError:
raise
def train():
# use recently added file from the directory 'recordings'
recog = recognition.SpeakerRecognizer('/home/pi/jasper/recordings/')
# set speaker lable for training
recog.speaker_name = 'Aditya'
# this will print some useful logs for debugging
# you should see a message saying "done training with [filename].wav"
recog.debug = True
# train model
recog.train_new_data()
def test_subprocess_function(self):
recog = recognition.SpeakerRecognizer()
filename = os.path.join(DATA_DIR_PATH, 'arctic_a0001.wav')
cmd = 'cp ' + filename + ' ' + os.path.join(DATA_DIR_PATH, 'copy.wav')
recog._start_subprocess(cmd)
self.assertTrue(os.path.isfile(os.path.join(DATA_DIR_PATH,
'copy.wav')))
try:
os.remove(os.path.join(DATA_DIR_PATH, 'copy.wav'))
except IOError:
raise
def find_speaker():
recog = recognition.SpeakerRecognizer()
#recording the voice
#saving the recording
vad.record
#using the newly recorded file to test
name = []
name = recog.identify_speaker()
return name
def test_identify_recently_added_file(self):
recog = recognition.SpeakerRecognizer(DATA_DIR_PATH)
recog.speaker_name = 'Test'
recog.train_new_data()
recog.last_trained_file = None
recog.speaker_name = 'Test2'
recog.train_new_data()
name = None
name = recog.identify_speaker()
self.assertNotEqual(name, None)
self.assertEqual(name[0], 'Test')
self.assertEqual(name[1], 'Test2')
def test_training_an_entire_folder(self):
recog = recognition.SpeakerRecognizer()
recog.speaker_name = 'Test'
recog.train_new_data(os.path.join(DATA_DIR_PATH))
model = glob.glob('*.gzbin')
self.assertTrue(os.path.isfile('./' + ''.join(model)))
self.assertTrue(os.path.isfile('./speakers.txt'))
try:
os.remove('./' + ''.join(model))
os.remove('./speakers.txt')
except IOError:
raise
def test_training_with_specified_file(self):
recog = recognition.SpeakerRecognizer()
recog.speaker_name = 'Test'
recog.train_new_data(os.path.join(DATA_DIR_PATH, 'arctic_a0001.wav'))
model = glob.glob('*.gzbin')
self.assertTrue(os.path.isfile('./' + ''.join(model)))
self.assertTrue(os.path.isfile('./speakers.txt'))
try:
os.remove('./' + ''.join(model))
os.remove('./speakers.txt')
except IOError:
raise
def test_get_speaker_from_database(self):
recog = recognition.SpeakerRecognizer(DATA_DIR_PATH)
recog.speaker_name = 'Test'
recog.train_new_data()
recog = recognition.SpeakerRecognizer(DATA_DIR_PATH)
recog.speaker_name = 'Test1'
recog.train_new_data()
speakers = []
speakers = recog.get_speakers()
self.assertEqual(speakers[0], 'Test')
self.assertEqual(speakers[1], 'Test1')
model = glob.glob('*.gzbin')
self.assertTrue(os.path.isfile('./' + ''.join(model)))
self.assertTrue(os.path.isfile('./speakers.txt'))
try:
os.remove('./' + ''.join(model))
os.remove('./speakers.txt')
except IOError:
raise
def __init__(
self,
mqtt_hostname=os.environ.get('mqtt_hostname', 'mosquitto'),
mqtt_port=os.environ.get('mqtt_port', 1883),
hotword_model=os.environ.get('hotword_model',
'resources/snowboy.umdl'),
hotword=os.environ.get('hotword', 'snowboy'),
sensitivity=os.environ.get('hotword_sensitivity', '0.5'),
listen_to=os.environ.get('hotword_listen_to', 'default'),
):
SnipsMqttServer.__init__(self, mqtt_hostname, mqtt_port)
self.hotword = hotword
self.hotword_model = hotword_model
self.listen_to = listen_to
self.subscribe_to = 'hermes/hotword/+/toggleOff,hermes/hotword/+/toggleOn,hermes/audioServer/+/audioFrame'
# PIWHO
self.piwho_model = os.path.dirname(
os.path.abspath(__file__)) + '/piwho/'
self.piwho_speakers = self.piwho_model + '/speakers.txt'
self.piwho_training_data = os.path.dirname(
os.path.abspath(__file__)) + '/piwho/data'
self.recog = None
self.clientList = []
self.allowedClientList = listen_to.split(',')
self.client_buffer = {}
self.client_recognition = {}
self.client_talking = {}
self.record = {}
self.messageCount = 0
self.detection = snowboydecoder.HotwordDetector(
hotword_model, sensitivity=sensitivity)
for client in self.allowedClientList:
self.client_buffer[client] = AudioBuffer(
self.detection.detector.NumChannels() *
self.detection.detector.SampleRate() * 20)
self.client_recognition[client] = AudioBuffer(
self.detection.detector.NumChannels() *
self.detection.detector.SampleRate() * 2)
self.client_talking[client] = False
self.record[client] = False
self.trainPiWho()
if (self.piwho_enabled()):
self.recog = recognition.SpeakerRecognizer()
def handle(self, text, mic):
recog = recognition.SpeakerRecognizer('/home/pi/jasper-client/recordings/')
name = []
name = recog.identify_speaker()
if name[0] == 'Aditya':
mic.say("Good morning "+ name)
else:
if name[0] == 'unknown':
gen = gd.identify_gender(recog.get_recently_added_file())
if gen == 'M':
mic.say("Good morning sir")
elif gen == 'F':
mic.say("Good morning madam")
def find_speaker():
recog = recognition.SpeakerRecognizer()
# Record voice until silence is detected.
# save WAV file
r = sr.Recognizer()
with sr.Microphone() as source:
r.adjust_for_ambient_noise(source)
vad.record()
# use newly recorded file.
name = []
name = recog.identify_speaker()
dictn=recog.get_speaker_scores()
print (dictn)
return name
def test_identify_specified_file(self):
recog = recognition.SpeakerRecognizer()
name = None
name = recog.identify_speaker(
os.path.join(DATA_DIR_PATH, 'arctic_a0001.wav'))
self.assertNotEqual(name, None)
self.assertEqual(name[0], 'Test')
model = glob.glob('*.gzbin')
self.assertTrue(os.path.isfile('./' + ''.join(model)))
self.assertTrue(os.path.isfile('./speakers.txt'))
try:
os.remove('./' + ''.join(model))
os.remove('./speakers.txt')
except IOError:
raise
def sayHello(mic):
recog = recognition.SpeakerRecognizer('/home/pi/jasper/recordings/')
name = []
name = recog.identify_speaker()
# I have trained the model with 30 voice clips with lable 'Aditya'
# These sound clips contain only my voice.
if 'Aditya' in name:
mic.say("Good morning " + name)
# Then trained the SAME model with 40 random voice clips with lable 'unknown'
# excluding my voice.
# This makes it easy to distinguish between two speakers.
elif 'unknown' in name:
gender = gd.identify_gender(recog.get_recently_added_file())
if 'M' in gender:
mic.say('Good morning Sir')
elif 'F' in gender:
mic.say('Good morning Madam')
def test_speaker_scores(self):
recog = recognition.SpeakerRecognizer(DATA_DIR_PATH)
recog.speaker_name = 'Test'
recog.train_new_data()
recog.last_trained_file = None
recog.speaker_name = 'Test2'
recog.train_new_data(os.path.join(DATA_DIR_PATH, 'arctic_a0001.wav'))
recog.identify_speaker()
dictn = recog.get_speaker_scores()
self.assertTrue(dictn)
model = glob.glob('*.gzbin')
self.assertTrue(os.path.isfile('./' + ''.join(model)))
self.assertTrue(os.path.isfile('./speakers.txt'))
try:
os.remove('./' + ''.join(model))
os.remove('./speakers.txt')
except IOError:
raise
def determine_speaker():
"""
takes speaker input data from the microphone and determines who was speaking
:return: a sorted list of scores for each team member from out distance classifier
"""
recog = recognition.SpeakerRecognizer()
# for now, set to classification by euclidean distance
recog.set_feature_option('-noise -raw -aggr -eucl')
name = recog.identify_speaker('./speak_detect.wav')
print(name)
scores = recog.get_speaker_scores()
sorted_scores = {k: float(v) for k, v in scores.iteritems()}
sort_s = sorted(sorted_scores.items(), key=operator.itemgetter(1))
# print full dictionary of scores
for k, v in sort_s:
print("%s: %f" % (k, v))
return sort_s
def test_is_good_wav_file(self):
recog = recognition.SpeakerRecognizer()
filepath = os.path.join(DATA_DIR_PATH, 'arctic_a0001.wav')
self.assertTrue(recog._is_good_wave(filepath))
def test_get_recently_added_file_in_directory(self):
recog = recognition.SpeakerRecognizer(DATA_DIR_PATH)
filename = None
filename = recog.get_recently_added_file()
self.assertNotEqual(filename, None)
def test_feature_option(self):
recog = recognition.SpeakerRecognizer()
recog.set_feature_option('-lpc -cheb')
self.assertEqual(recog.feature, ' -lpc -cheb ')
def test_speaker_name(self):
recog = recognition.SpeakerRecognizer()
recog.speaker_name = 'Test'
self.assertEqual(recog.speaker_name, 'Test')
def test_recently_added_file_exception(self):
recog = recognition.SpeakerRecognizer('/dev/null/')
self.assertRaises(ValueError, recog.get_recently_added_file)
