Turn ChatGPT into Your own Voice AI with Python

For more information about the Vosk speech recognition engine used, refer to:

https://alphacephei.com/vosk/

Code used in this video:

# TinkerChatGPT by The Tinker Foundry

# Giving ChatGPT ears to hear and a voice to speak

# ----------------------------------------------------

# Dependencies on python libraries described below

# ----------------------------------------------------

# Uses Vosk speech recognition:

#    pip install vosk

#    pip install sounddevice

#    prerequisites: as described in https://alphacephei.com/vosk/install and also python module `sounddevice` (simply run command `pip install sounddevice`)

# Also uses text-to-speech (TTS):

#    pip install pyttsx3

# Of course, we use the openAI api for access to ChatGPT engine:

#    pip install openai

# ----------------------------------------------------

import argparse

import queue

import sys

import json

import sounddevice as sd

import pyttsx3

import openai

from vosk import Model, KaldiRecognizer, SetLogLevel

SetLogLevel(-1) # this disables annoying Vosk logs

q = queue.Queue()

# -------------------------------------------

# This segment of code is for text to speech

# -------------------------------------------

engine = pyttsx3.init()

voices = engine.getProperty('voices')

#print(voices)

engine.setProperty('voice', voices[1].id)

rate = engine.getProperty('rate')

#print(rate)

engine.setProperty('rate', rate-200)

# -------------------------------------------

def int_or_str(text):

    """Helper function for argument parsing."""

    try:

        return int(text)

    except ValueError:

        return text

def callback(indata, frames, time, status):

    """This is called (from a separate thread) for each audio block."""

    if status:

        print(status, file=sys.stderr)

    q.put(bytes(indata))

parser = argparse.ArgumentParser(add_help=False)

parser.add_argument(

    "-l", "--list-devices", action="store_true",

    help="show list of audio devices and exit")

args, remaining = parser.parse_known_args()

if args.list_devices:

    print(sd.query_devices())

    parser.exit(0)

parser = argparse.ArgumentParser(

    description=__doc__,

    formatter_class=argparse.RawDescriptionHelpFormatter,

    parents=[parser])

parser.add_argument(

    "-f", "--filename", type=str, metavar="FILENAME",

    help="audio file to store recording to")

parser.add_argument(

    "-d", "--device", type=int_or_str,

    help="input device (numeric ID or substring)")

parser.add_argument(

    "-r", "--samplerate", type=int, help="sampling rate")

parser.add_argument(

    "-m", "--model", type=str, help="language model; e.g. en-us, fr, nl; default is en-us")

args = parser.parse_args(remaining)

try:

    if args.samplerate is None:

        device_info = sd.query_devices(args.device, "input")

        # soundfile expects an int, sounddevice provides a float:

        args.samplerate = int(device_info["default_samplerate"])

       

    if args.model is None:

        model = Model(lang="en-us")

    else:

        model = Model(lang=args.model)

    if args.filename:

        dump_fn = open(args.filename, "wb")

    else:

        dump_fn = None

   

    # Main line taking microphone input and converting speech to text...

    with sd.RawInputStream(samplerate=args.samplerate, blocksize = 8000, device=args.device,

            dtype="int16", channels=1, callback=callback):

        print("#" * 80)

        print("Start chatting...")

        print("#" * 80)

        rec = KaldiRecognizer(model, args.samplerate)

        while True:

            data = q.get()

            if rec.AcceptWaveform(data):

                text = json.loads(rec.Result())["text"]

                if (text != ""): # if there is some valid speech to text string, then pass to openAI api (ChatGPT model)

                    print(text)

                    openai.api_key_path = "OPENAI_API_KEY.txt" # here, provide the path to you openAI key

                    response = openai.Completion.create(

                      model="text-davinci-003", # this is the ChatGPT3 model

                      prompt=text,

                      temperature=0.7,

                      max_tokens=200, # adjust this to control length of response; note the higher this number, the more credit is used up

                      top_p=1,

                      frequency_penalty=0,

                      presence_penalty=0

                    )

                    #print (response)

                    saytext = response["choices"][0]["text"] # extract the text portion of the response

                    print (saytext)

                    # use text to speech to communicate ChatGPT response

                    engine.say(saytext)

                    engine.runAndWait()

                    engine.stop()

            if dump_fn is not None:

                dump_fn.write(data)

except KeyboardInterrupt:

    print("\nDone")

    parser.exit(0)

except Exception as e:

    parser.exit(type(e).__name__ + ": " + str(e))