app.py

import openai
from playsound import playsound
from gtts import gTTS
import speech_recognition as sr
import gradio as gr
import os

openai.api_key = os.environ['api_key']

def generate_response(prompt):
    prompt = (f"{prompt}")

    response = openai.ChatCompletion.create(
        model="gpt-3.5-turbo",
        messages=[
            {"role": "user", "content": f"{prompt}"},
        ])

    message = response.choices[0]['message']['content']
    return message

r = sr.Recognizer()
from pydub import AudioSegment
def transcribe(audio, lang):
    with sr.AudioFile(audio) as source: audio = r.record(source)
    text = r.recognize_google(audio, language=lang)
    text = generate_response(text)
    tts = gTTS(text=text, lang=lang)
    out = "tmp.mp3"
    tts.save(out)
    return out

with open('gradio_article.md') as f:
    article = f.read()

interface_options = {
    "title": "Smart GPT",
    "description": "Let's have a chat! Talk to me, and I'll respond in a jiffy!",
    "article": article,
    "layout": "horizontal",
    "theme": "default",
}

inputs = gr.Audio(source="microphone", type="filepath")
outputs = "audio"

lang = gr.Dropdown(choices=["en", "vi", "nl"], value="en", resettable=False)
if lang.value == "":
    lang.value = "en"

gr.Interface(fn=transcribe, inputs=[inputs, lang], outputs=outputs, live=False,
                      allow_clear=False, **interface_options).launch()

# TODO
# Custom voice
    # VALL-E
    # https://cloud.google.com/text-to-speech/custom-voice/docs/quickstart
    # Mozilla TTS
    # OpenSeq2Seq
    # Best VN: Vbee, FPT
    # Elevenlabs for English