app.py

import gradio as gr
import whisper

# =========================
# Load Whisper model ONCE
# =========================
model = whisper.load_model("small")


# =========================
# ASR function
# =========================
def transcribe(audio_path, src_lang, tgt_lang):
    if audio_path is None:
        return "No audio provided", None

    result = model.transcribe(audio_path)
    text = result.get("text", "").strip()

    if not text:
        text = "No speech detected"

    # Audio output is intentionally None (voice cloning later)
    return text, None


# =========================
# UI
# =========================
with gr.Blocks(title="Multilingual Voice Clone") as demo:
    gr.Markdown("# 🎙 Multilingual Voice Cloning Demo")
    gr.Markdown("**Phase 2: Speech-to-Text (Whisper ASR)**")

    with gr.Row():
        audio_input = gr.Audio(
            sources=["microphone"],
            type="filepath",
            label="Speak (Input Speech)"
        )

        gr.Audio(
            sources=["upload"],
            type="filepath",
            label="Upload Voice Sample (used later)",
            interactive=False
        )

    with gr.Row():
        src_lang = gr.Dropdown(
            ["Marathi", "Hindi", "English"],
            value="English",
            label="Input Language (for later)"
        )

        tgt_lang = gr.Dropdown(
            ["Hindi", "English", "Tamil"],
            value="Hindi",
            label="Output Language (for later)"
        )

    btn = gr.Button("Generate Voice")

    text_out = gr.Textbox(
        label="Transcribed Text",
        placeholder="Speech-to-text output will appear here"
    )

    audio_out = gr.Audio(
        label="Cloned Voice Output (coming in Phase 4)"
    )

    btn.click(
        fn=transcribe,
        inputs=[audio_input, src_lang, tgt_lang],
        outputs=[text_out, audio_out]
    )

demo.launch()