gradio_demo.py

import gradio as gr
import os
from SenseVoiceAx import SenseVoiceAx
from print_utils import rich_transcription_postprocess

max_len = 256

model_path = os.path.join("sensevoice_ax650", "sensevoice.axmodel")

assert os.path.exists(model_path), f"model {model_path} not exist"

pipeline = SenseVoiceAx(
    model_path,
    max_len=max_len,
    beam_size=3,
    language="auto",
    hot_words=None,
    use_itn=True,
    streaming=False,
)


def speech_to_text(audio_path, lang):
    """
    audio_path: 音频文件路径
    lang: 语言类型 "auto", "zh", "en", "yue", "ja", "ko"
    """
    if not audio_path:
        return "无音频"

    pipeline.choose_language(language=lang)
    asr_res = pipeline.infer(audio_path, print_rtf=False)

    return asr_res


def main():
    with gr.Blocks() as demo:
        with gr.Row():
            output_text = gr.Textbox(label="识别结果", lines=5)

        with gr.Row():
            audio_input = gr.Audio(
                sources=["upload"], type="filepath", label="录制或上传音频", format="mp3"
            )
            lang_dropdown = gr.Dropdown(
                choices=["auto", "zh", "en", "yue", "ja", "ko"],
                value="auto",
                label="选择音频语言",
            )

        audio_input.change(
            fn=speech_to_text, inputs=[audio_input, lang_dropdown], outputs=output_text
        )

    demo.launch(
        server_name="0.0.0.0",
    )


if __name__ == "__main__":
    main()