app.py

import os
import warnings
import whisper
import gradio as gr 
import openai
from gtts import gTTS

# Load model outside high-frequency function
model = whisper.load_model("base")
device = model.device

# Separate OpenAI API functionality into a function
def call_openai_api(text):
    openai.api_key = os.getenv("OPENAI_API_KEY")
    result = openai.Completion.create(
        model="gpt-3.5-turbo-instruct",
        prompt=text,
        max_tokens=500,
        temperature=0
    )
    return result["choices"][0]["text"]

def transcribe(audio):
    # load audio and pad/trim it to fit 30 seconds
    audio = whisper.load_audio(audio)
    audio = whisper.pad_or_trim(audio)

    # make log-Mel spectrogram and move to the same device as the model
    mel = whisper.log_mel_spectrogram(audio).to(device)

    # detect the spoken language
    with warnings.catch_warnings():
        warnings.simplefilter("ignore")
        _, _ = model.detect_language(mel)

    # decode the audio
    options = whisper.DecodingOptions(fp16=False)
    result = whisper.decode(model, mel, options)
    result_text = result.text

    # Call OpenAI API for response
    out_result = call_openai_api(result_text)

    # Generate speech audio
    speech = gTTS(out_result)
    speech.save("test.mp3")

    return [result_text, out_result, "test.mp3"]

output_1 = gr.Textbox(label="Speech to Text")
output_2 = gr.Textbox(label="ChatGPT Output")
output_3 = gr.Audio(type="filepath", autoplay=True)

demo = gr.Interface(
    title = 'Voice to Text & Voice reply using OpenAI (KF)', 
    fn=transcribe, 
    inputs=gr.Audio(sources=["microphone"], type="filepath"),
    outputs=[output_1,  output_2, output_3]
    )

if __name__ == "__main__":
    demo.launch()