app.py

import os
import tempfile
from openai import OpenAI
from tts_voice import tts_order_voice
import edge_tts
import numpy as np
import anyio
import torch
import torchaudio
import gradio as gr
from scipy.io import wavfile
from scipy.io.wavfile import write

# 创建 KNN-VC 模型
knn_vc = torch.hub.load('bshall/knn-vc', 'knn_vc', prematched=True, trust_repo=True, pretrained=True, device='cpu')

# 初始化 language_dict
language_dict = tts_order_voice

# 异步文字转语音函数
async def text_to_speech_edge(text, language_code):
    voice = language_dict[language_code]
    communicate = edge_tts.Communicate(text, voice)
    with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp_file:
        tmp_path = tmp_file.name

    await communicate.save(tmp_path)

    return "语音合成完成：{}".format(text), tmp_path

# 声音更改函数
#def voice_change(audio_in, audio_ref):
    #samplerate1, data1 = wavfile.read(audio_in)
    #samplerate2, data2 = wavfile.read(audio_ref)
    #write("./audio_in.wav", samplerate1, data1)
    #write("./audio_ref.wav", samplerate2, data2)

    #query_seq = knn_vc.get_features("./audio_in.wav")
    #matching_set = knn_vc.get_matching_set(["./audio_ref.wav"])
    #out_wav = knn_vc.match(query_seq, matching_set, topk=4)
    #torchaudio.save('output.wav', out_wav[None], 16000)
    #return 'output.wav'
def voice_change(audio_in, audio_ref):
    samplerate1, data1 = wavfile.read(audio_in)
    samplerate2, data2 = wavfile.read(audio_ref)

    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_audio_in, \
         tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_audio_ref:
        audio_in_path = tmp_audio_in.name
        audio_ref_path = tmp_audio_ref.name
        write(audio_in_path, samplerate1, data1)
        write(audio_ref_path, samplerate2, data2)

    query_seq = knn_vc.get_features(audio_in_path)
    matching_set = knn_vc.get_matching_set([audio_ref_path])
    out_wav = knn_vc.match(query_seq, matching_set, topk=4)
    output_path = 'output.wav'
    torchaudio.save(output_path, out_wav[None], 16000)
    return output_path

# def voice_change(audio_in, audio_ref):
#     samplerate1, data1 = wavfile.read(audio_in)
#     samplerate2, data2 = wavfile.read(audio_ref)

#     # 强制匹配音频文件的长度
#     max_length = max(data1.shape[0], data2.shape[0])

#     if data1.shape[0] < max_length:
#         data1 = np.pad(data1, (0, max_length - data1.shape[0]), mode='constant')
#     if data2.shape[0] < max_length:
#         data2 = np.pad(data2, (0, max_length - data2.shape[0]), mode='constant')

#     with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_audio_in, \
#          tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_audio_ref:
#         audio_in_path = tmp_audio_in.name
#         audio_ref_path = tmp_audio_ref.name
#         wavfile.write(audio_in_path, samplerate1, data1)
#         wavfile.write(audio_ref_path, samplerate2, data2)

#     query_seq = knn_vc.get_features(audio_in_path)
#     matching_set = knn_vc.get_matching_set([audio_ref_path])
#     out_wav = knn_vc.match(query_seq, matching_set, topk=4)
#     output_path = 'output.wav'
#     torchaudio.save(output_path, torch.tensor(out_wav)[None], 16000)
#     return output_path
# 文字转语音（OpenAI）
def tts(text, model, voice, api_key):
    if len(text) > 300:
        raise gr.Error('您输入的文本字符多于300个，请缩短您的文本')
    if api_key == '':
        raise gr.Error('请填写您的 中转API Key')

    try:
        client = OpenAI(api_key=api_key, base_url='https://lmzh.top/v1')
        response = client.audio.speech.create(
            model=model,
            voice=voice,
            input=text,
        )
    except Exception as error:
        raise gr.Error(f"生成语音时出错：{error}")

    with tempfile.NamedTemporaryFile(suffix=".mp3", delete=False) as temp_file:
        temp_file.write(response.content)

    return temp_file.name

# Gradio 前端设计
app = gr.Blocks()

with app:
    gr.Markdown("# <center>OpenAI TTS + 3秒实时AI变声+需要使用中转key</center>")
    gr.Markdown("### <center>中转key购买地址https://buy.sipola.cn</center>")
    with gr.Tab("TTS"):
        with gr.Row(variant='panel'):
            api_key = gr.Textbox(type='password', label='API Key', placeholder='请在此填写您的API Key')
            model = gr.Dropdown(choices=['tts-1','tts-1-hd'], label='请选择模型（tts-1推理更快，tts-1-hd音质更好）', value='tts-1')
            voice = gr.Dropdown(choices=['alloy', 'echo', 'fable', 'onyx', 'nova', 'shimmer'], label='请选择一个说话人', value='alloy')
        with gr.Row():
            with gr.Column():
                inp_text = gr.Textbox(label="请填写您想生成的文本中英文皆可", placeholder="请输入ai生成的文案，不要超过300字，最好200字左右", lines=5)
                btn_text = gr.Button("一键开启真实拟声吧", variant="primary")
            with gr.Column():
                inp1 = gr.Audio(type="filepath", label="OpenAI TTS真实拟声", interactive=False)
                inp2 = gr.Audio(type="filepath", label="请上传AI变声的参照音频（决定变声后的语音音色）")
                btn1 = gr.Button("一键开启AI变声吧", variant="primary")
            with gr.Column():
                out1 = gr.Audio(type="filepath", label="AI变声后的专属音频")
            btn_text.click(tts, [inp_text, model, voice, api_key], inp1)
            btn1.click(voice_change, [inp1, inp2], out1)
    # with gr.Tab("⚡ Edge TTS"):
    #     with gr.Row():
    #         input_text = gr.Textbox(lines=5, placeholder="请输入ai生成的文案，不要超过300字，最好200字左右", label="请填写您想生成的文本中英文皆可")
    #         default_language = list(language_dict.keys())[15]
    #         language = gr.Dropdown(choices=list(language_dict.keys()), value=default_language, label="请选择文本对应的语言")
    #         btn_edge = gr.Button("一键开启真实拟声吧", variant="primary")
    #         output_text = gr.Textbox(label="输出文本", visible=False)
    #         output_audio = gr.Audio(type="filepath", label="Edge TTS真实拟声")

    #     with gr.Row():
    #         inp_vc = gr.Audio(type="filepath", label="请上传AI变声的参照音频决定变声后的语音音色")
    #         btn_vc = gr.Button("一键开启AI变声吧", variant="primary")
    #         out_vc = gr.Audio(type="filepath", label="AI变声后的专属音频")

    #     btn_edge.click(lambda text, lang: anyio.run(text_to_speech_edge, text, lang), [input_text, language], [output_text, output_audio])
    #     btn_vc.click(voice_change, [output_audio, inp_vc], out_vc)
    with gr.Tab("Edge TTS"):
        with gr.Row():
            with gr.Column():
                input_text = gr.Textbox(label="请填写您想生成的文本中英文皆可",placeholder="请输入ai生成的文案，不要超过300字，最好200字左右",lines=5)
                btn_edge = gr.Button("一键开启真实拟声吧", variant="primary")
            with gr.Column():
                default_language = list(language_dict.keys())[15]
                language = gr.Dropdown(choices=list(language_dict.keys()), value=default_language, label="请选择文本对应的语言")
                output_audio = gr.Audio(type="filepath", label="TTS真实拟声")
                output_text = gr.Textbox(label="输出文本", visible=False)
        with gr.Row():
            with gr.Column():
                inp_vc = gr.Audio(type="filepath", label="请上传AI变声的参照音频决定变声后的语音音色")
                btn_vc = gr.Button("一键开启AI变声吧", variant="primary")
            with gr.Column():
                out_vc = gr.Audio(type="filepath", label="AI变声后的专属音频")
        btn_edge.click(lambda text, lang: anyio.run(text_to_speech_edge, text, lang), [input_text, language], [output_text, output_audio])
        btn_vc.click(voice_change, [output_audio, inp_vc], out_vc)
    gr.Markdown("### <center>注意获取中转API Key [here](https://buy.sipola.cn).</center>")
    gr.Markdown("### <center>ai文案生成可使用中转key，请访问 [here](https://ai.sipola.cn).</center>")
    gr.HTML('''
        <div class="footer">
         <center><p>Power by sipola </p></center>
        </div>
    ''')

app.launch(show_error=True)