推理部分，增加gradio的wav音频流式输出与TTS API

Author: upbit

GPT_SoVITS/inference_stream.py

@@ -0,0 +1,370 @@
+import os
+import tempfile, io, wave
+import gradio as gr
+import uvicorn
+import argparse
+from fastapi import FastAPI
+from fastapi.responses import StreamingResponse
+from pydub import AudioSegment
+from tools.i18n.i18n import I18nAuto
+from GPT_SoVITS.inference_webui import (
+    get_weights_names,
+    custom_sort_key,
+    change_choices,
+    change_gpt_weights,
+    change_sovits_weights,
+    get_tts_wav,
+)
+
+api_app = FastAPI()
+i18n = I18nAuto()
+
+# API mode Usage: python GPT_SoVITS/inference_stream.py --api
+parser = argparse.ArgumentParser(description="GPT-SoVITS Streaming API")
+parser.add_argument(
+    "-api",
+    "--api",
+    action="store_true",
+    default=False,
+    help="是否开启API模式(不开启则是WebUI模式)",
+)
+parser.add_argument(
+    "-s",
+    "--sovits_path",
+    type=str,
+    default="GPT_SoVITS/pretrained_models/s2G488k.pth",
+    help="SoVITS模型路径",
+)
+parser.add_argument(
+    "-g",
+    "--gpt_path",
+    type=str,
+    default="GPT_SoVITS/pretrained_models/s1bert25hz-2kh-longer-epoch=68e-step=50232.ckpt",
+    help="GPT模型路径",
+)
+parser.add_argument(
+    "-rw",
+    "--ref_wav",
+    type=str,
+    # default="./example/archive_ruanmei_8.wav",
+    help="参考音频路径",
+)
+parser.add_argument(
+    "-rt",
+    "--prompt_text",
+    type=str,
+    # default="我听不惯现代乐，听戏却极易入迷，琴弦拨动，时间便流往过去。",
+    help="参考音频文本",
+)
+parser.add_argument(
+    "-rl",
+    "--prompt_language",
+    type=str,
+    default="中文",
+    help="参考音频语种",
+)
+
+args = parser.parse_args()
+
+sovits_path = args.sovits_path
+gpt_path = args.gpt_path
+SoVITS_names, GPT_names = get_weights_names()
+
+EXAMPLES = [
+    [
+        "中文",
+        "根据过年的传说，远古时代有一隻凶残年兽，每到岁末就会从海底跑出来吃人。"
+        + "人们为了不被年兽吃掉，家家户户都会祭拜祖先祈求平安，也会聚在一起吃一顿丰盛的晚餐。"
+        + "后来人们发现年兽害怕红色、噪音与火光，便开始在当天穿上红衣、门上贴上红纸、燃烧爆竹声，藉此把年兽赶走。"
+        + "而这些之后也成为过年吃团圆饭、穿红衣、放鞭炮、贴春联的过年习俗。",
+    ],
+    [
+        "中文",
+        "神霄折戟录其二"
+        +"「嗯，好吃。」被附体的未央变得温柔了许多，也冷淡了很多。"
+        + "她拿起弥耳做的馅饼，小口小口吃了起来。第一口被烫到了,还很可爱地吐着舌头吸气。"
+        + "「我一下子有点接受不了, 需要消化消化。」用一只眼睛作为代价维持降灵的弥耳自己也拿了一个馅饼，「你再说一 遍?」"
+        + "「当年所谓的陨铁其实是神戟。它被凡人折断，铸成魔剑九柄。这一把是雾海魔剑。 加上他们之前已经收集了两柄。「然后你是?」"
+        + "「我是曾经的天帝之女，名字已经忘了。我司掌审判与断罪，用你们的话说，就是刑律。」"
+        + "因为光禄寺执掌祭祀典礼的事情，所以仪式、祝词什么的，弥耳被老爹逼得倒是能倒背如流。同时因为尽是接触怪力乱神，弥耳也是知道一些小]道的。 神明要是被知道了真正的秘密名讳，就只能任人驱使了。眼前这位未必是忘了。"
+        + "「所以朝廷是想重铸神霄之戟吗?」弥耳说服自己接受了这个设定，追问道。"
+        + "「我不知道。这具身体的主人并不知道别的事。她只是很愤怒,想要证明自己。」未央把手放在了胸口上。"
+        +"「那接下来,我是应该弄个什么送神仪式把你送走吗?」弥耳摸了摸绷带下已经失去功能的眼睛，「然后我的眼睛也会回来?」"
+    ],
+]
+
+
+# from https://huggingface.co/spaces/coqui/voice-chat-with-mistral/blob/main/app.py
+def wave_header_chunk(frame_input=b"", channels=1, sample_width=2, sample_rate=32000):
+    # This will create a wave header then append the frame input
+    # It should be first on a streaming wav file
+    # Other frames better should not have it (else you will hear some artifacts each chunk start)
+    wav_buf = io.BytesIO()
+    with wave.open(wav_buf, "wb") as vfout:
+        vfout.setnchannels(channels)
+        vfout.setsampwidth(sample_width)
+        vfout.setframerate(sample_rate)
+        vfout.writeframes(frame_input)
+
+    wav_buf.seek(0)
+    return wav_buf.read()
+
+
+def get_streaming_tts_wav(
+    ref_wav_path,
+    prompt_text,
+    prompt_language,
+    text,
+    text_language,
+    how_to_cut,
+    top_k,
+    top_p,
+    temperature,
+    ref_free,
+    byte_stream=True,
+):
+    chunks = get_tts_wav(
+        ref_wav_path=ref_wav_path,
+        prompt_text=prompt_text,
+        prompt_language=prompt_language,
+        text=text,
+        text_language=text_language,
+        how_to_cut=how_to_cut,
+        top_k=top_k,
+        top_p=top_p,
+        temperature=temperature,
+        ref_free=ref_free,
+        stream=True,
+    )
+
+    if byte_stream:
+        yield wave_header_chunk()
+        for chunk in chunks:
+            yield chunk
+    else:
+        # Send chunk files
+        i = 0
+        format = "wav"
+        for chunk in chunks:
+            i += 1
+            file = f"{tempfile.gettempdir()}/{i}.{format}"
+            segment = AudioSegment(chunk, frame_rate=32000, sample_width=2, channels=1)
+            segment.export(file, format=format)
+            yield file
+
+
+def webui():
+    with gr.Blocks(title="GPT-SoVITS Streaming Demo") as app:
+        gr.Markdown(
+            value=i18n(
+                "流式输出演示，分句推理后推送到组件中。由于目前bytes模式的限制，采用<a href='https://github.com/gradio-app/gradio/blob/gradio%404.17.0/demo/stream_audio_out/run.py'>stream_audio_out</a>中临时文件的方案输出分句。这种方式相比bytes，会增加wav文件解析的延迟。"
+            ),
+        )
+
+        gr.Markdown(value=i18n("模型切换"))
+        with gr.Row():
+            GPT_dropdown = gr.Dropdown(
+                label=i18n("GPT模型列表"),
+                choices=sorted(GPT_names, key=custom_sort_key),
+                value=gpt_path,
+                interactive=True,
+            )
+            SoVITS_dropdown = gr.Dropdown(
+                label=i18n("SoVITS模型列表"),
+                choices=sorted(SoVITS_names, key=custom_sort_key),
+                value=sovits_path,
+                interactive=True,
+            )
+            refresh_button = gr.Button(i18n("刷新模型路径"), variant="primary")
+            refresh_button.click(
+                fn=change_choices, inputs=[], outputs=[SoVITS_dropdown, GPT_dropdown]
+            )
+            SoVITS_dropdown.change(change_sovits_weights, [SoVITS_dropdown], [])
+            GPT_dropdown.change(change_gpt_weights, [GPT_dropdown], [])
+
+        gr.Markdown(value=i18n("*请上传并填写参考信息"))
+        with gr.Row():
+            inp_ref = gr.Audio(
+                label=i18n("请上传3~10秒内参考音频，超过会报错！"), value=args.ref_wav, type="filepath"
+            )
+            with gr.Column():
+                ref_text_free = gr.Checkbox(
+                    label=i18n("开启无参考文本模式。不填参考文本亦相当于开启。"),
+                    value=False,
+                    interactive=True,
+                    show_label=True,
+                )
+                gr.Markdown(i18n("使用无参考文本模式时建议使用微调的GPT"))
+                prompt_text = gr.Textbox(label=i18n("参考音频的文本"), value=args.prompt_text)
+            prompt_language = gr.Dropdown(
+                label=i18n("参考音频的语种"),
+                choices=[
+                    i18n("中文"),
+                    i18n("英文"),
+                    i18n("日文"),
+                    i18n("中英混合"),
+                    i18n("日英混合"),
+                    i18n("多语种混合"),
+                ],
+                value=args.prompt_language,
+            )
+
+            def load_text(file):
+                with open(file.name, "r", encoding="utf-8") as file:
+                    return file.read()
+
+            load_button = gr.UploadButton(i18n("加载参考文本"), variant="secondary")
+            load_button.upload(load_text, load_button, prompt_text)
+
+        gr.Markdown(
+            value=i18n(
+                "*请填写需要合成的目标文本。中英混合选中文，日英混合选日文，中日混合暂不支持，非目标语言文本自动遗弃。"
+            )
+        )
+        with gr.Row():
+            text = gr.Textbox(
+                label=i18n("需要合成的文本"), value="", lines=5, interactive=True
+            )
+            text_language = gr.Dropdown(
+                label=i18n("需要合成的语种"),
+                choices=[
+                    i18n("中文"),
+                    i18n("英文"),
+                    i18n("日文"),
+                    i18n("中英混合"),
+                    i18n("日英混合"),
+                    i18n("多语种混合"),
+                ],
+                value=i18n("中文"),
+            )
+            how_to_cut = gr.Radio(
+                label=i18n("怎么切"),
+                choices=[
+                    i18n("不切"),
+                    i18n("凑四句一切"),
+                    i18n("凑50字一切"),
+                    i18n("按中文句号。切"),
+                    i18n("按英文句号.切"),
+                    i18n("按标点符号切"),
+                ],
+                value=i18n("按标点符号切"),
+                interactive=True,
+            )
+
+        gr.Markdown(value=i18n("* 参数设置"))
+        with gr.Row():
+            with gr.Column():
+                top_k = gr.Slider(
+                    minimum=1,
+                    maximum=100,
+                    step=1,
+                    label=i18n("top_k"),
+                    value=5,
+                    interactive=True,
+                )
+                top_p = gr.Slider(
+                    minimum=0,
+                    maximum=1,
+                    step=0.05,
+                    label=i18n("top_p"),
+                    value=1,
+                    interactive=True,
+                )
+                temperature = gr.Slider(
+                    minimum=0,
+                    maximum=1,
+                    step=0.05,
+                    label=i18n("temperature"),
+                    value=1,
+                    interactive=True,
+                )
+            inference_button = gr.Button(i18n("合成语音"), variant="primary")
+
+        gr.Markdown(value=i18n("* 结果输出(等待第2句推理结束后会自动播放)"))
+        with gr.Row():
+            audio_file = gr.Audio(
+                value=None,
+                label=i18n("输出的语音"),
+                streaming=True,
+                autoplay=True,
+                interactive=False,
+                show_label=True,
+            )
+
+        inference_button.click(
+            get_streaming_tts_wav,
+            [
+                inp_ref,
+                prompt_text,
+                prompt_language,
+                text,
+                text_language,
+                how_to_cut,
+                top_k,
+                top_p,
+                temperature,
+                ref_text_free,
+            ],
+            [audio_file],
+        ).then(lambda: gr.update(interactive=True), None, [text], queue=False)
+
+        with gr.Row():
+            gr.Examples(
+                EXAMPLES,
+                [text_language, text],
+                cache_examples=False,
+                run_on_click=False,  # Will not work , user should submit it
+            )
+
+    app.queue().launch(
+        server_name="0.0.0.0",
+        inbrowser=True,
+        share=False,
+        server_port=8080,
+        quiet=True,
+    )
+
+
+@api_app.get("/")
+async def tts(
+    text: str,  # 必选参数
+    language: str = i18n("中文"),
+    top_k: int = 5,
+    top_p: float = 1,
+    temperature: float = 1,
+):
+    ref_wav_path = args.ref_wav
+    prompt_text = args.prompt_text
+    prompt_language = args.prompt_language
+    how_to_cut = i18n("按标点符号切")
+
+    return StreamingResponse(
+        get_streaming_tts_wav(
+            ref_wav_path=ref_wav_path,
+            prompt_text=prompt_text,
+            prompt_language=prompt_language,
+            text=text,
+            text_language=language,
+            how_to_cut=how_to_cut,
+            top_k=top_k,
+            top_p=top_p,
+            temperature=temperature,
+            ref_free=False,
+            byte_stream=True,
+        ),
+        media_type="audio/x-wav",
+    )
+
+
+def api():
+    uvicorn.run(
+        app="inference_stream:api_app", host="127.0.0.1", port=8080, reload=True
+    )
+
+
+if __name__ == "__main__":
+    # 模式选择，默认是webui模式
+    if not args.api:
+        webui()
+    else:
+        api()