run.py

"""Launches the Whisper Transcribe app behind an ngrok tunnel."""

import os
import sys
import threading

os.environ.setdefault("WHISPER_MODEL_SIZE", "large-v3")
os.environ.setdefault("HOST", "0.0.0.0")
os.environ.setdefault("PORT", "5000")

from backend import config                      # noqa: E402
from backend.app import app, logger             # noqa: E402
import backend.transcriber as T                # noqa: E402
from pyngrok import ngrok                       # noqa: E402


def _preload() -> None:
    T.cleanup_outputs()
    print(f"Pre-loading '{config.MODEL_SIZE}' model (first run downloads weights ~1-3 GB)…")
    try:
        T.get_model(config.MODEL_SIZE)
        print("Model ready.\n")
    except Exception as exc:
        logger.warning("Model pre-load failed: %s — will retry on first request.", exc)


def main() -> None:
    flask_thread = threading.Thread(
        target=lambda: app.run(host=config.HOST, port=config.PORT, threaded=True),
        daemon=True,
    )
    flask_thread.start()

    token = os.environ.get("NGROK_AUTHTOKEN", "").strip()
    if token:
        ngrok.set_auth_token(token)
    else:
        print(
            "WARNING: NGROK_AUTHTOKEN is not set — falling back to ngrok anonymous tier, "
            "which may be rate-limited."
        )

    try:
        public_url = ngrok.connect(config.PORT).public_url
    except Exception as exc:
        print("Could not open an ngrok tunnel:", exc)
        print("Set NGROK_AUTHTOKEN in .env and run again.")
        raise

    print("\n" + "=" * 64)
    print(f"   Whisper Transcribe is live → {public_url}")
    print("=" * 64 + "\n")

    # Pre-load in background so the URL is immediately usable
    threading.Thread(target=_preload, daemon=True).start()

    flask_thread.join()


if __name__ == "__main__":
    try:
        main()
    except KeyboardInterrupt:
        print("\nShutting down.")
    except Exception:
        logger.exception("Fatal error while starting the app")
        sys.exit(1)