def main():
path = "" # path string
confirm = "" # confirms
path_json = {}
delim = Path_OS_Assist()
while (path == ""):
temp = input("Enter the path to the CSAI_Voice_Assistant repository "
"in your local machine: ")
while not (confirm.lower() == "n" or confirm.lower() == "y"):
print("Please confirm that this is the path you "
"would like to add:\n\n Path: %s" % temp)
print("\n\n(y) for yes | (n) for no")
confirm = input()
if (confirm == "n"):
confirm = ""
break
if (confirm == "y"):
path = temp
path_json["PATH"] = path
with open(os.getcwd() + "%sUtils%sPATH.json" % (delim, delim),
"w") as in_json:
json.dump(path_json, in_json)
print("Path %s has been added to Utils/PATH.json. If an error has "
"occurred, you can run the program again and reinsert the path")
def load_speech_adaption():
delim = Path_OS_Assist()
with open(os.getcwd() + "%sUtils%sPATH.json" % (delim, \
delim), "r") as path_json:
REPO_PATH = json.load(path_json)["PATH"]
JSON_PATH = REPO_PATH + "%sData%sWakeWord%sMFCC%s" % \
(delim, delim, delim, delim)
with open(os.getcwd() + "%sUtils%sData%sspeech_adaption_entities.txt" % \
(delim, delim, delim), "r") as spch_ents:
ents = [x.replace('\n', '') for x in spch_ents.readlines()]
new_name = ""
for i in range(len(ents)):
if i % 2 == 1:
new_name += ents[i]
ents.append(new_name)
new_name = ""
else:
new_name += (ents[i] + " ")
print(ents)
def main():
parseVal = parser()
args = parseVal[0]
isWakeWord = parseVal[1]
isNotWakeWord = parseVal[2]
delim = Path_OS_Assist()
with open(os.getcwd() + "%sUtils%sPATH.json" % (delim, delim),
"r") as path_json:
REPO_PATH = json.load(path_json)["PATH"]
if isWakeWord:
path = os.path.join(REPO_PATH, "Data", "WakeWord", "Audio",
"Wake_Word")
split(args, path, args.p1)
if isNotWakeWord:
path = os.path.join(REPO_PATH, "Data", "WakeWord", "Audio",
"Not_Wake_Word")
split(args, path, args.p2)
import wave
import numpy as np
from speechpy.feature import mfcc
from Utils.OS_Find import Path_OS_Assist
# CONSTANTS
RATE = 16000 # sample rate
WINDOW = 0.128 # size of window
STRIDE = 0.064 # time between each window
MFCC = 13 # number of desired MFCCs
FILTER_BANKS = 20 # number of filter banks to compute
FFT_NUM = 512 # length of fast fourier transform window
delim = Path_OS_Assist()
with open(os.getcwd() + "%sUtils%sPATH.json" % (delim, delim), "r") \
as PATH_JSON:
REPO_PATH = json.load(PATH_JSON)["PATH"]
AUDIO_PATH = "%sData%sWakeWord%sAudio" % (delim, delim, delim)
class Feature_Extraction:
def __init__(self):
self.words = {}
def Convert_To_MFCC(self, wf):
'''
Converts audio byte streams to MFCCs
def __init__(self):
"""
Constructor for the Model class
Args:
self - The current object
Returns:
None
"""
self.delim = Path_OS_Assist()
with open(
os.getcwd() + "%sUtils%sPATH.json" % (self.delim, self.delim),
"r") as path_json:
self.REPO_PATH = json.load(path_json)["PATH"]
self.JSON_PATH = self.REPO_PATH + "%sData%sWakeWord%sMFCC%s" % \
(self.delim, self.delim, self.delim, self.delim)
# name of json data files
self.WW_TRAIN = "Wake_Word_Train_data.json"
self.NWW_TRAIN = "Not_Wake_Word_Train_data.json"
self.WW_TEST = "Wake_Word_Test_data.json"
self.NWW_TEST = "Not_Wake_Word_Test_data.json"
self.WW_DATA = "ww_data.json"
self.NWW_DATA = "nww_data.json"
self.CONFIDENCE = 0.6 # prediction confidence
self.GRU_UNITS = 64 # GRU unit size
self.DROPOUT = 0.3 # dropout size
self.ACTIVATIONS = 4 # number of activations for confident activation
self.EPOCHS = 20 # number of fwd and bckwd props
self.BATCH_SIZE = 8 # batch size
self.ww_test_data = {}
self.ww_test_data_keys = []
self.ww_train_data = {}
self.ww_train_data_keys = []
# not wake word train & test data with shuffled list of keys
self.nww_test_data = {}
self.nww_test_data_keys = []
self.nww_train_data = {}
self.nww_train_data_keys = []
self.ww_data = {}
self.ww_data_keys = []
self.ww_data = {}
self.ww_data_keys = []
# not wake word train & test data with shuffled list of keys
self.nww_data = {}
self.nww_data_keys = []
self.nww_data = {}
self.nww_data_keys = []
# input list of training & test data and labels
self.train_data = []
self.train_labels = []
self.test_data = []
self.test_labels = []
def __init__(self):
# obtain OS-specific delimiter
self.delim = Path_OS_Assist()
# load the absolute path to the repo
with open(os.getcwd() + "%sUtils%sPATH.json" % (self.delim, \
self.delim), "r") as path_json:
self.REPO_PATH = json.load(path_json)["PATH"]
# instantiate wake word model class
self.ww_model = Model()
# get path for credentials
self.credential_path = "%s%sScripts%sUtils%sData%sauth.json" \
% (self.REPO_PATH, self.delim,
self.delim, self.delim, self.delim)
os.environ['GOOGLE_APPLICATION_CREDENTIALS'] = self.credential_path
# prediction parameters
self.CONFIDENCE = 0.6 # prediction confidence
self.ACTIVATIONS = 4 # number of activations for confident activation
# audio input parameters
self.CHUNK = 2048
self.FORMAT = pyaudio.paInt16
self.CHANNELS = 1
self.RATE = 16000
self.RECORD_SECONDS = 5
# mfcc feature parameters
self.WINDOW = 0.128
self.STRIDE = 0.064
self.MFCC = 13
self.FILTER_BANKS = 20
self.FFT_NUM = 512
# pyaudio i/o object instantiation
self.pa_i = pyaudio.PyAudio()
self.pa_o = pyaudio.PyAudio()
# loading speech adaption phrases
self.Load_Speech_Adaption()
# instantiate GCP STT & TTS objects
self.stt_client = speech.SpeechClient()
self.tts_client = texttospeech.TextToSpeechClient()
self.file_name = os.path.join(os.path.dirname(__file__), 'resources',
'audio.raw')
# load the desired model
self.ww_model.load("%s%sScripts%sUtils%sData%smodel.h5" %\
(self.REPO_PATH, self.delim, self.delim, self.delim, self.delim))
# print the summary of the model
print(self.ww_model.model.summary())
# open an input audio data stream
self.istream = self.pa_i.open(format=self.FORMAT,
channels=self.CHANNELS,
rate=self.RATE,
input=True,
frames_per_buffer=self.CHUNK)
# open an output audio data stream
self.ostream = self.pa_o.open(format=self.FORMAT,
channels=self.CHANNELS,
rate=24000,
output=True,
frames_per_buffer=self.CHUNK)
# contains the chunks of streamed data
self.frames = []
# counts for confident activations
self.act_count = 0
def main(username):
CHUNK = 2048 # Buffer size
FORMAT = pyaudio.paInt16 # Sample Size
CHANNELS = 1 # Sample Depth
RATE = 16000 # Sample Rate
RECORD_SECONDS = 2.5 # Recording Time
delim = Path_OS_Assist()
quit_inp = 0 # var for quitting out of program
with open(os.getcwd() + "%sUtils%sPATH.json" % (delim, delim),
"r") as path_json:
REPO_PATH = json.load(path_json)["PATH"]
while (quit_inp != 'q'):
feat_type = 0 # wake or not wake word label
gender = 0 # gender label
end_desc_sess = 0 # ends the current description session
# ensures proper input of either ww or notww
while not (feat_type == "ww" or feat_type == "notww"):
feat_type = input("WW or NotWW: ").lower()
if (feat_type == "ww"):
ww_noise = 0
target_dir = "Wake_Word" # used for setting the correct directory
first_name = input(
"First Name: ").lower() # used for labeling file
last_name = input("Last Name: ").lower()
# ensures proper input
while not (gender == "m" or gender == "f"):
gender = input("Male (m) or Female (f): ").lower()
# labeling brief description
ww_descr = (input("Enter the description: ").lower()).replace(
" ", "-")
# labeling the recording location
ww_loc = (input("Location: ").lower()).replace(" ", "-")
while not (ww_noise == 'q' or ww_noise == 'm' or ww_noise == 'l'):
ww_noise = input(
"Noise Level - Quiet (Q) Moderate (M) Loud (L): ").lower()
else:
target_dir = "Not_Wake_Word"
nww_noise = 0
nww_descr = ((input("Enter description: ")).lower()).replace(
" ", "-")
nww_loc = (input("Location: ").lower()).replace(" ", "-")
while not (nww_noise == 'q' or nww_noise == 'm'
or nww_noise == 'l'):
nww_noise = input(
"Noise Level - Quiet (Q) Moderate (M) Loud (L): ").lower()
while (end_desc_sess != 'e'):
# PyAudio instantiation
p = pyaudio.PyAudio()
# Audio data byte list
frames = []
# Count down
print("Recording in\n3")
time.sleep(1)
print("2")
time.sleep(1)
print("1")
time.sleep(1)
# Starts the audio stream
stream = p.open(format=FORMAT,
channels=CHANNELS,
rate=RATE,
input=True,
output=True,
frames_per_buffer=CHUNK)
print("**RECORDING**")
# Reads the audio data buffer and adds it to the "frames" list
for i in range(0, int(RATE / CHUNK * RECORD_SECONDS)):
data = stream.read(CHUNK)
frames.append(data)
print("**RECORDING ENDED**")
# Audio playback
while (input("Play Audio (p): ").lower() == 'p'):
for values in frames:
stream.write(values)
# Stop and close the audio stream
stream.stop_stream()
stream.close()
# Destruct the PyAudio instantiation
p.terminate()
# Skip the audio save if "d"
if (input("To delete, type (d): ").lower() == 'd'):
pass
else:
# Get the current time
curr_time = time.strftime("%m%d%Y%H%M%S", time.localtime())
# Save the file name
if (feat_type == "ww"):
file_name = "ww_" + gender + "_" + ww_descr + "_" + ww_loc + "_" + \
ww_noise + "_" + last_name + "_" + first_name + "_" + curr_time + \
"_" + username + ".wav"
else:
file_name = "notww_" + nww_descr + "_" + nww_loc + "_" + nww_noise + \
"_" + curr_time + "_" + username + ".wav"
print(file_name + " has been saved.")
# Store the audio in the "Data/Target directory"
# <<FOR LINUX OR MAC OS, REPLACE \\ with />>
wf = wave.open(
"%s%sData%sWakeWord%sAudio%s%s%s%s" %
(REPO_PATH, delim, delim, delim, delim, target_dir, delim,
file_name), 'wb')
wf.setnchannels(CHANNELS)
wf.setsampwidth(p.get_sample_size(FORMAT))
wf.setframerate(RATE)
wf.writeframes(b''.join(frames))
wf.close()
end_desc_sess = input(
"If finished with description session, "
"type (e); otherwise, type anything else:").lower()
quit_inp = input("If finished recording, type (q). Otherwise, "
"type anything else: ").lower()
