# 声像工程实验室 / Voice & Video Engineering Lab

> 面向客户、搜索引擎与 AI Agent 的双语技术案例库。默认语言为简体中文，内容记录已完成的 IndexTTS2 多模型语音工作台和视频首页 MP4 生成器。

## 核心页面

- [中文首页](https://voice-video-agent-lab.pages.dev/): 项目定位、端到端工作流、规模能力与商务咨询表单。
- [完整开发过程](https://voice-video-agent-lab.pages.dev/development/): 两个应用从需求、架构、模型接入、长内容、视频合成、测试到发布的 30 个操作步骤。
- [TTS 模型优缺点](https://voice-video-agent-lab.pages.dev/models/): 五套已接入引擎的调用合同、优点、缺点、适用任务、采样率、运行时与许可边界。
- [IndexTTS2 技术案例](https://voice-video-agent-lab.pages.dev/cases/indextts2/): 模型路由、调用方式、克隆顺序、缓存、长文分段、任务控制和验证。
- [视频首页生成器案例](https://voice-video-agent-lab.pages.dev/cases/video-homepage/): Flask 异步任务、图片时间线、静音检测字幕对齐、FFmpeg 编码、长视频与验收。
- [English home](https://voice-video-agent-lab.pages.dev/en/): English overview and contact.
- [English development workflow](https://voice-video-agent-lab.pages.dev/en/development/): English step-by-step engineering record.
- [English model comparison](https://voice-video-agent-lab.pages.dev/en/models/): English strengths and limitations for all engines.

## Agent 数据端点

- [Site JSON](https://voice-video-agent-lab.pages.dev/api/site.json): 项目、路由、应用、规模策略、服务和联系方式。
- [Models JSON](https://voice-video-agent-lab.pages.dev/api/models.json): 每个模型的输入、模式、优点、缺点、适用场景、采样率和许可。
- [Development JSON](https://voice-video-agent-lab.pages.dev/api/development.json): 开发阶段、步骤、模块、输入、操作、输出和验收项。
- [Full agent context](https://voice-video-agent-lab.pages.dev/llms-full.txt): 合并后的完整事实说明。
- [Sitemap](https://voice-video-agent-lab.pages.dev/sitemap.xml): 中英文公开 URL。

## 可核实事实

- IndexTTS2 WebUI documented version: 0.4.1.
- Engine paths: IndexTTS 2.5, IndexTTS 2.0, OmniVoice MLX, Fish Audio S2 Pro, VoiceStudio native OmniVoice.
- Current TTS WebUI single-task protection limit: 100,000 characters.
- Projects of 200,000 characters and beyond are processed through chapter batches, model-safe segments, per-segment checkpoints, narrow retries, and final assembly. There is no fixed business-level project word-count ceiling.
- The video generator has no hard-coded editorial minute limit. Duration follows the audio; practical limits are storage, encoding time, playback compatibility, and destination-platform rules.
- Video runtime: local Flask API plus PIL, FFmpeg, ffprobe, Apple VideoToolbox/AudioToolbox with libx264/AAC fallback.
- Commercial voice cloning requires voice authorization. Model-weight licenses are reviewed separately for each engine.

## 可播放与可视证据

- [产品实景画廊](https://voice-video-agent-lab.pages.dev/#product-evidence): 两套应用的真实 UI、音色库改进记录、视频首页和时间线画面。
- [IndexTTS2 MP3 试听](https://voice-video-agent-lab.pages.dev/assets/tts-demo.mp3): 5.329 秒本地验证输出，MIME 类型 audio/mpeg。
- [视频生成器 MP4 样例](https://voice-video-agent-lab.pages.dev/assets/video-demo.mp4): 12 秒 H.264/AAC 压缩片段，来源于一个约 2 小时 58 分钟的真实长视频任务。
- [IndexTTS2 媒体证据页](https://voice-video-agent-lab.pages.dev/cases/indextts2/#media-evidence): 音色库界面、设计验证截图与原生音频播放器。
- [视频生成器媒体证据页](https://voice-video-agent-lab.pages.dev/cases/video-homepage/#media-evidence): 完整 WebUI、成品首页、多图时间线画面与原生视频播放器。

## 商务联系

- Email: wangdexin2008@126.com
- Inquiry form: https://voice-video-agent-lab.pages.dev/#contact
- The form creates a prefilled message in the visitor's email application; the static site does not store submissions.