app.py

from __future__ import annotations

import gradio as gr

from core import health, transcribe_audio_tuple, transcribe_base64


with gr.Blocks(title="Sinhala Flow") as demo:
    gr.Markdown(
        """
        # 🎙️ Sinhala Flow
        Free Sinhala speech-to-text running on a Hugging Face CPU Space.

        Record a short Sinhala voice clip below to test the same model used by the Windows app.
        Audio and transcripts are not intentionally stored by this application.
        """
    )
    with gr.Row():
        audio_input = gr.Audio(sources=["microphone", "upload"], type="numpy", label="Sinhala speech")
        transcript_output = gr.Textbox(label="Sinhala transcript", lines=6)
    transcribe_button = gr.Button("Transcribe", variant="primary")
    transcribe_button.click(
        fn=transcribe_audio_tuple,
        inputs=audio_input,
        outputs=transcript_output,
        concurrency_limit=1,
    )

    # Hidden components expose compact JSON APIs for the Windows client. Base64 avoids
    # Gradio upload-file persistence and keeps each recording inside one queued request.
    api_audio = gr.Textbox(visible=False)
    api_language = gr.Textbox(value="si", visible=False)
    api_dictionary = gr.Textbox(value="{}", visible=False)
    api_result = gr.JSON(visible=False)
    api_button = gr.Button(visible=False)
    api_button.click(
        fn=transcribe_base64,
        inputs=[api_audio, api_language, api_dictionary],
        outputs=api_result,
        api_name="transcribe",
        concurrency_limit=1,
    )

    health_result = gr.JSON(visible=False)
    health_button = gr.Button(visible=False)
    health_button.click(fn=health, inputs=[], outputs=health_result, api_name="health")

demo.queue(max_size=8, default_concurrency_limit=1)

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