#!/usr/bin/env python3
"""
Regeneriert alle MP3s, deren Text sich gegenueber audio-texts.json.bak-pre-christo-kill
geaendert hat ODER deren Text einen Aussprache-Map-Key enthaelt (= braucht neue Aussprache).

Logik:
  1) Diff zwischen current und .bak-pre-christo-kill -> 'changed'
  2) Eintraege mit Aussprache-Map-Treffern ('Montafon', 'Brandner', ...) -> 'pron'
  3) Vereinigung -> regen-Liste

Voice-Routing wie regenerate-mission-audios.py.
Cache (audio-durations.json) wird fuer regenerierte Eintraege geloescht.

Usage:
  python regenerate-changed-audios.py --dry      # zeigt nur Liste
  python regenerate-changed-audios.py            # regeneriert wirklich
"""
import json, os, sys, subprocess, tempfile, shutil, time

KEY = os.environ.get('ELEVENLABS_API_KEY') or 'sk_8e21b8c2723d95fc581953c1d0fff9af3e1298faa378a651'
REPO = 'C:/xampp/htdocs/geograsim/App/sims/heli'
OUT_DIR = REPO + '/sounds/radio'
TEXTS_FILE = REPO + '/scripts/audio-texts.json'
BAK = TEXTS_FILE + '.bak-pre-christo-kill'
MAP_FILE = REPO + '/data/aussprache-map.json'
DUR_FILE = REPO + '/data/audio-durations.json'

VOICE_SARAH = 'EXAVITQu4vr4xnSDxMaL'
VOICE_BELLA = 'hpp4J3VqNfWAUOO0d1Us'

def voice_for(aid):
    if aid.startswith('r_mission_intro_') or aid.startswith('r_mission_start_') or aid.startswith('r_tower_'):
        return VOICE_SARAH, 'Sarah'
    return VOICE_BELLA, 'Bella'

def apply_pron(text, mapping):
    for k in sorted(mapping.keys(), key=len, reverse=True):
        text = text.replace(k, mapping[k])
    return text

def generate(aid, text, voice_id):
    out = OUT_DIR + '/' + aid + '.mp3'
    if os.path.exists(out):
        shutil.copy(out, out + '.bak-pre-christo-kill')
    body = json.dumps({
        'text': text,
        'model_id': 'eleven_multilingual_v2',
        'voice_settings': {'stability': 0.55, 'similarity_boost': 0.75, 'style': 0.1, 'use_speaker_boost': True}
    }, ensure_ascii=False)
    with tempfile.NamedTemporaryFile(mode='w', encoding='utf-8', suffix='.json', delete=False) as f:
        f.write(body); bpath = f.name
    r = subprocess.run(['curl','-sS','-w','%{http_code}','-o',out,
        '-X','POST','https://api.elevenlabs.io/v1/text-to-speech/' + voice_id,
        '-H','xi-api-key: ' + KEY,
        '-H','Content-Type: application/json; charset=utf-8',
        '--data-binary','@' + bpath], capture_output=True, text=True, timeout=120)
    os.unlink(bpath)
    if r.stdout.strip() == '200':
        return 'ok', os.path.getsize(out)
    return 'err-' + r.stdout, 0

def main():
    if not os.path.exists(BAK):
        print('FEHLER: ' + BAK + ' nicht vorhanden.')
        sys.exit(2)
    current = json.load(open(TEXTS_FILE, 'r', encoding='utf-8'))
    bak = json.load(open(BAK, 'r', encoding='utf-8'))
    pron = json.load(open(MAP_FILE, 'r', encoding='utf-8'))['pronunciations']
    durs = json.load(open(DUR_FILE, 'r', encoding='utf-8'))

    pron_keys = list(pron.keys())
    changed, pron_only = [], []
    for aid, txt in current.items():
        if aid.startswith('_'): continue
        if not isinstance(txt, str): continue
        old = bak.get(aid, '')
        if txt != old:
            changed.append(aid)
        elif any(k in txt for k in pron_keys):
            pron_only.append(aid)

    to_regen = changed + pron_only
    print(f'Diff-changed: {len(changed)}')
    print(f'Pron-Map-only: {len(pron_only)}  ({", ".join(pron_only) if pron_only else "-"})')
    print(f'Total Regen: {len(to_regen)}')

    if '--dry' in sys.argv:
        print('\n--dry-run, keine Audios generiert.')
        return

    print(f'\nStarte Regen ({len(to_regen)} MP3s)...')
    t0 = time.time()
    ok, fail = 0, 0
    for i, aid in enumerate(to_regen, 1):
        text = current[aid]
        tts = apply_pron(text, pron)
        voice, label = voice_for(aid)
        status, size = generate(aid, tts, voice)
        elapsed = time.time() - t0
        eta = (elapsed / i) * (len(to_regen) - i) if i > 0 else 0
        if status == 'ok':
            print(f'  [{i:3d}/{len(to_regen)}] OK  {aid:42s} {size//1024:>4}K [{label}]  ETA {eta:.0f}s')
            ok += 1
            durs['durations'].pop(aid, None)
        else:
            print(f'  [{i:3d}/{len(to_regen)}] ERR {aid:42s} {status}')
            fail += 1
        # Zwischenspeicherung jedes 20. Mal
        if i % 20 == 0:
            json.dump(durs, open(DUR_FILE, 'w', encoding='utf-8'), indent=2, ensure_ascii=False)

    json.dump(durs, open(DUR_FILE, 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
    print(f'\nFertig: {ok} OK, {fail} Fehler, {time.time()-t0:.0f}s gesamt')

if __name__ == '__main__':
    main()