OpenAI Whisper and ChatGPT 语音助手

OpenAI Whisper and ChatGPT ASR Gradio Web UI

一 环境准备

1.1 python

gradio==3.19.1
gTTS==2.3.1
openai==0.27.0
openai-whisper==20230124

1.2 windows

使用以下命令安装 ffmpeg

choco install ffmpeg

需要科学上网，否则连接超时

二 导入所需要的包

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

三 加载模型

openai.api_key='输入你自己的openai-key'
model = whisper.load_model("base")

四 定义openai和whisper接口

def chatgpt_api(input_text):
    messages = [
    {"role": "system", "content": "you are great!"}]
    
    if input_text:
        messages.append(
            {"role": "user", "content": input_text},
        )
        chat_completion = openai.ChatCompletion.create(
            model="gpt-3.5-turbo", messages=messages
        )
    
    reply = chat_completion.choices[0].message.content
    return reply

def transcribe(audio):
    language = "zh-CN"
    audio = whisper.load_audio(audio)
    audio = whisper.pad_or_trim(audio)
    mel = whisper.log_mel_spectrogram(audio).to(model.device)

    _, probs = model.detect_language(mel)

    options = whisper.DecodingOptions(fp16 = False)
    result = whisper.decode(model, mel, options)
    result_text = result.text
    
    out_result = chatgpt_api(result_text)
    
    audioobj = gTTS(text = out_result, 
                    lang = language, 
                    slow = False)
    
    audioobj.save("Aria.mp3")

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

五 生成Gradio Web UI

output_1 = gr.Textbox(label="Speech to Text")
output_2 = gr.Textbox(label="ChatGPT Output")
output_3 = gr.Audio("Aria.mp3")

gr.Interface(
    title = 'OpenAI Whisper and ChatGPT ASR Gradio Web UI', 
    fn=transcribe, 
    inputs=[
        gr.inputs.Audio(source="microphone", type="filepath")
    ],

    outputs=[
        output_1,  output_2, output_3
    ],
    live=True).launch()

参考：https://github.com/bhattbhavesh91/voice-assistant-whisper-chatgpt