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()