app.py

import gradio as gr
from huggingface_hub import InferenceClient

model_id = "JacobLinCool/whisper-large-v3-turbo-common_voice_16_1-zh-TW-2"
client = InferenceClient(model_id)


def transcribe_audio(audio: str) -> str:
    out = client.automatic_speech_recognition(audio)
    return out.text


with gr.Blocks() as demo:
    gr.Markdown("# TWASR: Chinese (Taiwan) Automatic Speech Recognition.")
    gr.Markdown("Upload an audio file or record your voice to transcribe it to text.")
    gr.Markdown(
        "First load may take a while to initialize the model, following requests will be faster."
    )

    with gr.Row():
        audio_input = gr.Audio(
            label="Audio", type="filepath", show_download_button=True
        )
        text_output = gr.Textbox(label="Transcription")

    transcribe_button = gr.Button("Transcribe with Inference API")
    transcribe_button.click(
        fn=transcribe_audio, inputs=[audio_input], outputs=[text_output]
    )

    gr.Examples(
        [
            ["./examples/audio1.mp3"],
            ["./examples/audio2.mp3"],
        ],
        inputs=[audio_input],
        outputs=[text_output],
        fn=transcribe_audio,
        cache_examples=True,
        cache_mode="lazy",
        run_on_click=True,
    )

    gr.Markdown(
        f"Current model: {model_id}. For more information, visit the [model hub](https://huggingface.co./{model_id})."
    )

if __name__ == "__main__":
    demo.launch()