import time

import numpy as np
from faster_whisper import WhisperModel

from whisper_app import app, config, grammar, media_duck, typer


def load_model() -> None:
    try:
        app.log(f"Loading {config.config['model']} on {config.config['device']}...")
        model_dir = config._resolve_path(config.config.get("model_dir", "")) or None
        app.model = WhisperModel(
            config.config["model"],
            device=config.config["device"],
            compute_type=config.config["compute_type"],
            download_root=model_dir,
        )
        app.log("Model ready.")
        if config.config.get("grammar_check"):
            grammar.init(config.config.get("language") or "de", log=app.log)
    except Exception as e:
        app.log(f"Model load FAILED: {e}")


def stop_and_transcribe() -> None:
    if app.state != app.AppState.RECORDING:
        return
    set_state(app.AppState.TRANSCRIBING)
    try:
        _do_transcribe()
    except Exception as e:
        app.log(f"Transcription error: {e}")
    finally:
        set_state(app.AppState.IDLE)


def _do_transcribe() -> None:
    chunks = list(app.audio_chunks)

    if not chunks:
        return

    if app.model is None:
        app.log("Model not loaded yet — skipped.")
        return

    audio = np.concatenate(chunks, axis=0).flatten().astype(np.float32)
    duration = len(audio) / config.config["sample_rate"]
    rms = float(np.sqrt(np.mean(audio ** 2)))
    app.log(f"Audio: {duration:.1f}s  RMS: {rms:.5f}")

    if duration < 0.3 or rms < 0.0001:
        app.log("Too short or silent — skipped.")
        return

    target_rms = 0.05
    if rms > 0:
        audio = audio * (target_rms / rms)
    audio = np.clip(audio, -1.0, 1.0)

    lang = config.config["language"] if config.config["language"] else None
    prompt = config.get_initial_prompt()
    segments, _ = app.model.transcribe(
        audio, language=lang, beam_size=5, vad_filter=True,
        initial_prompt=prompt if prompt else None,
    )
    text = " ".join(s.text for s in segments).strip()
    text = config.apply_vocab(text)
    if config.config.get("grammar_check"):
        text = grammar.correct(text)
    app.log(f"Result: {repr(text)}")

    if text:
        time.sleep(0.15)
        typer.type_text(text)


def set_state(new_state: app.AppState) -> None:  # semi-public, used by main.py
    app.state = new_state
    if app.tray_icon:
        from whisper_app import tray
        tray.update_icon(new_state)
    if new_state == app.AppState.RECORDING:
        from whisper_app import overlay
        overlay.show()
        if config.config.get("media_duck"):
            media_duck.duck(config.config.get("duck_percent", 20))
    else:
        from whisper_app import overlay
        overlay.hide()
        media_duck.unduck()