ute2e/tools/bridge-005-moss-preview.py

"""Generate an isolated five-step preview using the workshop's avatar and Junhao."""
from pathlib import Path
import hashlib
import html
import json
import shutil
import subprocess
import sys
import threading
import time
import argparse

import httpx
from sqlalchemy import select

ROOT = Path(__file__).resolve().parents[3]
sys.path.insert(0, str(ROOT / 'ai_person/ai_person_api'))
from app.config import get_settings
from app.persistence.database import get_session_factory
from app.persistence.agent_models import Agent, AgentAvatar, AgentVoice
from app.persistence.asset_models import Avatar, CapabilityAsset
from app.schemas.open_platform import GenerationCommand
from app.services.moss_tts_service import MossTtsService
from app.services.open_generation_worker import OpenGenerationWorker

REPORT = ROOT / 'unreal_tran/ute2e/reports/bridge-005-moss-junhao-v2-20260919'
WEB = ROOT / 'unreal_tran/unreal_tran_web'


def read_manifest(name):
    source = (WEB / 'src/features/virtual-training-scripts' / name).read_text('utf-8')
    return json.loads(source.split('export default', 1)[1].strip().rstrip(';'))


def digest(path):
    return hashlib.sha256(path.read_bytes()).hexdigest()


def main():
    REPORT.mkdir(parents=True, exist_ok=True)
    settings = get_settings()
    parser = argparse.ArgumentParser()
    parser.add_argument('--service-url', required=True, help='Renderer with the validated v2 avatar')
    args = parser.parse_args()
    settings = settings.model_copy(update={'service_base_url': args.service_url})
    with httpx.Client(trust_env=False, timeout=30) as client:
        widget = client.get('http://127.0.0.1:6180/api/auth/v1/system-config/public').json()['data']['digitalHumanWidget']
        with get_session_factory()() as db:
            agent = db.scalar(select(Agent).where(Agent.slug == widget['agentSlug'], Agent.is_delete == 0))
            relation = db.scalar(select(AgentAvatar).where(AgentAvatar.agent_id == agent.id).order_by(AgentAvatar.is_default.desc(), AgentAvatar.sort_order))
            avatar = db.get(Avatar, relation.avatar_id)
            voice = db.scalar(select(CapabilityAsset).where(CapabilityAsset.asset_code == 'builtin-voice-moss-junhao'))
            assert db.scalar(select(AgentVoice).where(AgentVoice.agent_id == agent.id, AgentVoice.capability_id == voice.id)), 'Junhao is not selected in the linked agent'
            avatar_id, voice_id = str(avatar.id), str(voice.id)
            provider_avatar = avatar.provider_avatar_id
            profile = {'agent': agent.agent_name, 'slug': agent.slug, 'avatar': avatar.avatar_name,
                       'providerAvatarId': provider_avatar, 'voice': voice.asset_name, 'engine': 'MOSS-TTS-Nano',
                       'speed': float(agent.voice_speed), 'width': 480, 'height': 640}
        manifest = client.get(str(settings.service_base_url).rstrip('/')+f'/api/v1/avatars/{provider_avatar}/assets/manifest.json')
        manifest.raise_for_status()
        profile['avatarManifestSha256'] = hashlib.sha256(manifest.content).hexdigest()
        expected = json.loads((ROOT / 'ai_person/ai_person_service/builtin_avatars/instructors-v2.json').read_text('utf-8'))['instructors'][0]
        assert profile['avatarManifestSha256'] == expected['bundle_sha256']['manifest.json'], 'Renderer does not have the approved v2 avatar'
    pack = next(p for p in read_manifest('narration-manifest.js')['scripts'] if p['code'] == 'TASK-VIRTUAL-005')
    old_pack = next(p for p in read_manifest('narration-video-manifest.js')['scripts'] if p['code'] == 'TASK-VIRTUAL-005')
    renderer = OpenGenerationWorker(None, settings, None, None)
    engine = MossTtsService(settings)
    rows = []
    try:
        for step in pack['steps']:
            number = step['number']
            stem = f'TASK-VIRTUAL-005-step-{number:02}-MOSS-Junhao'
            audio, video = REPORT / f'{stem}.wav', REPORT / f'{stem}.mp4'
            capture = REPORT / f'{stem}.webm'
            if video.exists():
                raise RuntimeError('Output exists; use a new report folder to preserve comparisons')
            print(f'Step {number}: synthesizing Junhao', flush=True)
            previous_audio = REPORT.parent / 'bridge-005-moss-junhao-20260919' / audio.name
            if previous_audio.exists():
                import wave
                shutil.copyfile(previous_audio, audio)
                with wave.open(str(audio)) as wav:
                    generated = {'duration': wav.getnframes() / wav.getframerate()}
            else:
                generated = engine.synthesize(step['text'], voice_code='Junhao', speed=profile['speed'])
                source = engine.output_path(generated['id'])
                shutil.copyfile(source, audio)
                source.unlink()
            print(f'Step {number}: rendering avatar, audio {generated["duration"]:.2f}s', flush=True)
            command = GenerationCommand(title=f'005 第{number}步 MOSS Junhao', text=step['text'],
                avatarId=avatar_id, voiceId=voice_id, speed=profile['speed'], width=480, height=640, subtitles=False)
            renderer.capture(command, provider_avatar, audio, None, capture, threading.Event())
            subprocess.run([settings.ffmpeg_binary, '-hide_banner', '-loglevel', 'error', '-y', '-i', str(capture),
                '-r', '25', '-c:v', 'libx264', '-preset', 'veryfast', '-crf', '20', '-pix_fmt', 'yuv420p',
                '-c:a', 'aac', '-b:a', '128k', '-movflags', '+faststart', str(video)], check=True)
            info = json.loads(subprocess.run([settings.ffprobe_binary, '-v', 'error', '-show_streams', '-show_format',
                '-of', 'json', str(video)], capture_output=True, text=True, check=True).stdout)
            v = next(s for s in info['streams'] if s['codec_type'] == 'video')
            a = next(s for s in info['streams'] if s['codec_type'] == 'audio')
            assert (v['width'], v['height'], v['codec_name']) == (480, 640, 'h264')
            assert a['codec_name'] == 'aac'
            assert abs(float(info['format']['duration']) - generated['duration']) < 0.7
            subprocess.run([settings.ffmpeg_binary, '-v', 'error', '-i', str(video), '-f', 'null', '-'], check=True)
            previous = next(s for s in old_pack['steps'] if s['number'] == number)
            old = WEB / 'public' / previous['url'].lstrip('/')
            if old.is_file(): shutil.copyfile(old, REPORT / f'original-step-{number:02}.mp4')
            row = {'number': number, 'title': step['expectedTitle'], 'text': step['text'], 'video': video.name,
                   'audio': audio.name, 'duration': float(info['format']['duration']), 'videoSha256': digest(video)}
            rows.append(row)
            (REPORT / 'result.json').write_text(json.dumps({'profile': profile, 'steps': rows, 'publishedBindingsChanged': False}, ensure_ascii=False, indent=2), 'utf-8')
            print(f'Step {number}: complete ({row["duration"]:.2f}s)', flush=True)
    finally:
        engine.close()
    cards = []
    for row in rows:
        cards.append(f'''<section><h2>第 {row['number']} 步 · {html.escape(row['title'])}</h2><p>{html.escape(row['text'])}</p>
        <div class="compare"><div><h3>新版 · MOSS Junhao</h3><video src="{row['video']}" controls preload="metadata"></video><p><a href="{row['audio']}">单独试听新语音</a></p></div>
        <div><h3>原版 · VITS</h3><video src="original-step-{row['number']:02}.mp4" controls preload="metadata"></video></div></div></section>''')
    page = '''<!doctype html><html lang="zh-CN"><meta charset="utf-8"><meta name="viewport" content="width=device-width"><title>005 · MOSS Junhao 讲解对比</title>
    <style>body{margin:32px auto;max-width:1040px;padding:0 20px;background:#edf5f3;color:#193e38;font:16px/1.7 system-ui}section{background:white;border-radius:16px;padding:24px;margin:24px 0}h2{font-size:20px}.compare{display:flex;gap:28px;flex-wrap:wrap}.compare>div{flex:1;min-width:220px}video{width:100%;max-width:360px;aspect-ratio:3/4;background:#17342f;border-radius:12px}a{color:#008363}</style>
    <h1>005 液压泵应急切换 · 新旧讲解对比</h1><p>教员1 · v2素材 · MOSS-TTS-Nano / Junhao · 1.0倍速 · 3:4。五个步骤沿用原讲解文稿，当前为独立预览，线上讲解未替换。</p>''' + ''.join(cards) + '''<script>document.addEventListener('play',e=>{document.querySelectorAll('video,audio').forEach(m=>{if(m!==e.target)m.pause()})},true)</script></html>'''
    (REPORT / 'index.html').write_text(page, 'utf-8')
    print('Preview ready: '+str(REPORT / 'index.html'), flush=True)


if __name__ == '__main__':
    main()