A Chat Assistant for senior. Using Python, Naver Clova, Wake Word and ChatGPT 3.5
It can understand what you say on Mic. And answer like a human on Speaker.
It can be your friend. And If you want to change ChatGPT's prompt, then you can make other Chat Assistant easily.
-
Wake Chat Assistant : If you say "Hey", then It works using.
-
Talk Something : After saying "Hey", Say anything on Mic within 3 secs, The Chat Assistant understand what you said.
- Execute
while response != "": response = robot.gpt_send_anw(response) ans = response speaking(ans)
-
Make Answer properly : It is going to make answer something using ChatGPT. And speak out using Speaker.
-
Special Function
- Reset : If you say "Reset", Then It remove the user's data. And ask you about new user's data.
- Turn off : If you say "Turn off", Then It quit the process.
if response == "reset": speaking("ok. Reset mode") name_ini() elif response == "turn off": speaking("ok. turn off mode") return False
- Clone this Repository on your PC.
git clone https://github.com/HyungkyuKimDev/Chat_Assistant.git
- Install requirements.txt file on your Terminal.
pip install -r requirements.txt
- change Directory to The directory where you clone This Repository.
cd [The dircetory where you clone] - execute chatbot_voice.py
python3 chatbot_voice.py
stream = sd.InputStream(
samplerate=RATE, channels=CHANNELS, dtype='int16')
stream.start()
owwModel = Model(
wakeword_models=["../models/hey.tflite"], inference_framework="tflite")
n_models = len(owwModel.models.keys())
# Main loop for wake word detection
while True:
# Get audio
audio_data, overflowed = stream.read(CHUNK)
if overflowed:
print("Audio buffer has overflowed")
audio_data = np.frombuffer(audio_data, dtype=np.int16)
# Feed to openWakeWord model
prediction = owwModel.predict(audio_data)
common = False
# Process prediction results
for mdl in owwModel.prediction_buffer.keys():
scores = list(owwModel.prediction_buffer[mdl])
if scores[-1] > 0.2: # Wake word detected
print(f"wake word dectected {mdl}!")
mdl = ""
scores = [0] * n_models
audio_data = np.array([])
common = True
if common:
speaking("yes sir!")- Reference : dscripka/openWakeWord
def mic(time):
import requests
import sounddevice as sd
from scipy.io.wavfile import write
# Recording Voice
fs = 44100
seconds = time # time for recording
myRecording = sd.rec(int(seconds * fs), samplerate=fs, channels=4) # channels는 마이크 장치 번호
print("recording start")
# Find mic channel => python -m sounddevice
sd.wait()
write('sampleWav.wav', fs, myRecording)
# Voice To Text Using Naver Cloud : CLOVA Speech Recognition
## Set
lang = "Eng"
url = "https://naveropenapi.apigw.ntruss.com/recog/v1/stt?lang=" + lang
## Recorded Voice File
data_voice = open('sampleWav.wav', 'rb')
## headers
headers = {
"X-NCP-APIGW-API-KEY-ID": client_id,
"X-NCP-APIGW-API-KEY": client_secret,
"Content-Type": "application/octet-stream"
}
## VTT Output
response = requests.post(url, data=data_voice, headers=headers)
result_man = str(response.text)
result = list(result_man)
count_down = 0
say_str = []
for i in range(0, len(result) - 2):
if count_down == 3:
say_str.append(result[i])
if response.text[i] == "\"":
if count_down == 3:
break
else:
count_down += 1
anw_str = ''.join(map(str, say_str))
print(anw_str)
return anw_str- Reference : Naver Cloud : CLOVA Speech Recognition
def speaking(anw_text):
# NAVER CLOVA : CLOVA Voice
encText = urllib.parse.quote(anw_text)
data = f"speaker=djoey&volume=0&speed=0&pitch=0&format=mp3&text=" + encText
urls = "https://naveropenapi.apigw.ntruss.com/tts-premium/v1/tts"
requests = urllib.request.Request(urls)
requests.add_header("X-NCP-APIGW-API-KEY-ID", client_id)
requests.add_header("X-NCP-APIGW-API-KEY", client_secret)
response = urllib.request.urlopen(requests, data=data.encode('utf-8'))
rescodes = response.getcode()
if (rescodes == 200):
response_body = response.read()
with open('./ResultMP3.mp3', 'wb') as f:
f.write(response_body)
# speaker output
filename = "ResultMP3.mp3"
dst = "test.wav"
sound = AudioSegment.from_mp3(filename)
sound.export(dst, format="wav")
# data, fs = sf.read(filename, dtype='')
pl("test.wav")
else:
print("404 error")
# Remove Audio data
os.remove("ResultMP3.mp3")
os.remove("test.wav")- Reference : NAVER CLOVA : CLOVA Voice
class Robot():
memory_size = 100
with open('./user_value.json', 'r') as f:
data = json.load(f)
nameValue = data["user_name"]
manWomanValue = data["user_value"]
def set_memory_size(self, memory_size):
self.memory_size = memory_size
def gpt_send_anw(self, question: str):
self.gpt_standard_messages = [{"role": "assistant",
"content": f"You're a assistant robot for senior in USA. Your name is robot. "
f"Your being purpose is support. So Please answer politely in english and under 5 seconds. "
f"please be a good friend to your patient. "
f"Your patient's name is {self.nameValue} and {self.manWomanValue} is an old person."},
{"role": "user", "content" : question}]
response = openai.ChatCompletion.create(
model="gpt-3.5-turbo",
messages=self.gpt_standard_messages,
temperature=0.8
)
answer = response['choices'][0]['message']['content']
self.gpt_standard_messages.append({"role": "user", "content": question})
self.gpt_standard_messages.append({"role": "assistant", "content": answer})
return answerHyungkyu Kim
Project Link: HyungkyuKimDev/Chat_Assistant
- Wake Word Part developed
- Robot Engineer