[Python] ai音色翻译器

之前看到了ai voice有些许兴趣便玩了玩，用花儿不哭大佬制作的rvc推理包做了几个模型，可能因为训练量和杂音的原因导致最后使用的效果不佳，后关注到花儿不哭大佬新的GPT-SoVITS项目。这个由于是文字转语音，训练出来的效果好上很多，但是是tts，没法用做变声器，而且推理时间相对rvc变声器过长。因此发现可以以语音转文字然后翻译再转为音频，以当做一个翻译器来用。

项目使用需有GPT-SoVITS的后端接口所以选择了箱庭XTer的推理包，给了相应接口给与调用

https://www.yuque.com/xter/zibxlp/kkicvpiogcou5lgp

模型训练：

https://github.com/RVC-Boss/GPT-SoVITS

翻译使用谷歌翻译，调用对应api。参考编写

https://www.52pojie.cn/thread-1730501-1-1.html
备用api参考：
https://www.52pojie.cn/forum.php?mod=viewthread\&tid=1903090\&highlight=�ȸ跭��

由于语音转文字使用的是python的vosk库需去网址下载对应的模型文件（感觉识别准确度不够高）

https://alphacephei.com/vosk/models

<blockquote>import sounddevice as sd
import vosk
import queue
import json
import os
import sys
import requests
import pyaudio
import requests
from requests.exceptions import JSONDecodeError
from bs4 import BeautifulSoup


# 初始化队列
q = queue.Queue()


# 设置音频输入设备的采样率
samplerate = 16000


# 确保模型路径正确
model_path = "E:/aivoice/vosk-model-cn-0.22"  # 替换为你的模型文件夹路径，例如 "vosk-model-small-cn-0.3"


# 检查模型路径是否存在
if not os.path.exists(model_path):
    print(f"模型路径 {model_path} 不存在，请检查路径是否正确。")
    sys.exit(1)


# 加载Vosk模型
model = vosk.Model(model_path)




def translate_text(text,source_lang='',target_lang='ja'):                     #接口1:https://findmyip.net/api/translate.php
    # url = f"https://findmyip.net/api/translate.php?text={text}"
    url =f"https://findmyip.net/api/translate.php?text={text}&source_lang={source_lang}&target_lang={target_lang}"
    response = requests.get(url)


    try:
        data = response.json()
        print(data)
        if response.status_code == 200:
            if data['code']== 200:
                translation = data['data']['translate_result']
                print("translation",translation)
                return translation
            elif data['code'] == 400:
                return "出错"
            else:
                return "内部接口错误，请联系开发者"
        else:
            return "内部接口错误，请联系开发者"    
    except JSONDecodeError as e:
            return "出错"
    except requests.RequestException as e:
            return "出错"    




def google_translate_text(text,source_lang='zh-CN',target_lang='ja'):         #接口2:https://translate.google.com
    url = "https://www.google.com.hk/async/translate"
    # word="你好"
    payload = f"async=translate,sl:{source_lang},tl:{target_lang},st:{text},id:1672056488960,qc:true,ac:true,_id:tw-async-translate,_pms:s,_fmt:pc"
    payload_encoded = payload.encode('utf-8')
    headers = { 
    'sec-ch-ua': '"Not?A_Brand";v="8", "Chromium";v="108", "Google Chrome";v="108"', 
    'DNT': '1', 
    'sec-ch-ua-mobile': '?0', 
    'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/108.0.0.0 Safari/537.36', 
    'sec-ch-ua-arch': '"x86"', 
    'sec-ch-ua-full-version': '"108.0.5359.125"', 
    'Content-Type': 'application/x-www-form-urlencoded;charset=UTF-8', 
    'sec-ch-ua-platform-version': '"10.0.0"', 
    'sec-ch-ua-full-version-list': '"Not?A_Brand";v="8.0.0.0", "Chromium";v="108.0.5359.125", "Google Chrome";v="108.0.5359.125"', 
    'sec-ch-ua-bitness': '"64"', 
    'sec-ch-ua-model': '', 
    'sec-ch-ua-wow64': '?0', 
    'sec-ch-ua-platform': '"Windows"', 
    'Accept': '*/*', 
    'X-Client-Data': 'CKW1yQEIhbbJAQiktskBCMS2yQEIqZ3KAQjb08oBCLD+ygEIlaHLAQjv8swBCN75zAEI5PrMAQjxgM0BCLKCzQEI7ILNAQjIhM0BCO+EzQEIt4XNAQ==', 
    'Sec-Fetch-Site': 'same-origin', 
    'Sec-Fetch-Mode': 'cors', 
    'Sec-Fetch-Dest': 'empty', 
    'host': 'www.google.com.hk', 
    'Cookie': '1P_JAR=2022-12-26-12; NID=511=eVLI1bG9nhyOZtqU14JBHm5Be00epdxfR4XmfQeehYyIkzgpXi6dbpNY75ZMVyS7aOjoM2oZ5WdoR8eNq6wi1-e_J0NeoyI0dtsHW-_8Ik4PGrqvuGHdcvVC03zTOEK2TY1FZL85Wimo_ZPIE3hGIrmGPSiel6-rRRW9lD30UPs'
    } 
    response = requests.request("POST", url, headers=headers, data=payload_encoded) 
    
    html=response.text
    soup = BeautifulSoup(html, 'html.parser')
    target_text = soup.find('span', {'id': 'tw-answ-target-text'})
    if target_text:
        print(target_text.get_text())
        return target_text.get_text()
    return None




    # return response


# 定义回调函数
def callback(indata, frames, time, status):
    if status:
        print(status, file=sys.stderr)
    q.put(bytes(indata))


# 定义发送文字到文本到语音服务并播放音频的函数
def text_to_speech(text):
    stream_url = f'http://127.0.0.1:5000/tts?character=holi&text={text}&stream=true'
    
    # 初始化pyaudio
    p = pyaudio.PyAudio()


    # 打开音频流
    stream = p.open(format=p.get_format_from_width(2),
                    channels=1,
                    rate=32000,
                    output=True)


    # 使用requests获取音频流
    response = requests.get(stream_url, stream=True)


    # 读取数据块并播放
    for data in response.iter_content(chunk_size=1024):
        stream.write(data)


    # 停止和关闭流
    stream.stop_stream()
    stream.close()


    # 终止pyaudio
    p.terminate()


# 打开音频流
with sd.RawInputStream(samplerate=samplerate, blocksize=8000, device=2, dtype='int16',
                       channels=1, callback=callback):                      #device是所使用的录入设备，需要自行调整，可通过代码输出查看所有设备列表进行更改
    print('#' * 80)
    print('按 Ctrl+C 结束')
    print('#' * 80)
    
    rec = vosk.KaldiRecognizer(model, samplerate)
    last_partial = ""
    while True:
        data = q.get()
        if rec.AcceptWaveform(data):
            result = rec.Result()
            result_json = json.loads(result)
            recognized_text = result_json['text']
            print(f"识别结果: {recognized_text}")
            last_partial = ""
            


            # 将识别结果发送到文本到语音服务并播放音频
            if recognized_text:
                # translated_text=translate_text(recognized_text)                                          #调用接口1
                translated_text=google_translate_text(recognized_text,'zh-CN','ja')                        #调用接口2  前一个是输入语言（需跟vosk模型一样），后一个是翻译后的语言
                
                text_to_speech(translated_text)




        else:
            partial_result = rec.PartialResult()
            partial_result_json = json.loads(partial_result)
            new_partial = partial_result_json['partial']
            
            # 分割新旧部分的不同内容
            new_words = new_partial[len(last_partial):].strip().split()
            if new_words:
                print(f"部分识别结果: {' '.join(new_words)}")
                
            last_partial = new_partial</blockquote>