A TTS engine that speaks Cantonese, Mandarin, and Tibetan — from SourceForge to GitHub

What it does Ekho converts Chinese text to speech, supporting Cantonese, Mandarin, and Tibetan. It’s a C++ engine originally developed for the eGuideDog accessibility project, now living on GitHub after a SourceForge exile at revision r2418.

The interesting bit The voice architecture is deliberately modular: you can swap out the entire pinyin folder with new recordings, delete the index files, and let Ekho rebuild its voice database on next run. There’s even documentation for recording your own voice — a rarity in TTS projects that usually treat speakers as black boxes.

Key highlights

• Supports Cantonese, Mandarin, and Tibetan (per repo topics)
• Voice data ships separately; must be downloaded from SourceForge or a distro package
• Voice replacement workflow: swap folder, delete pinyin.index and pinyin.voice, restart
• Custom voice creation documented (in Chinese) via eGuideDog’s site
• Forked from SourceForge at r2418, patched to r2478

Caveats

• Voice files not included in repository; installation is a two-step process
• Build instructions exist but are in a separate INSTALL file, not the README
• Project appears to be in maintenance mode; last significant activity unclear from sources

Verdict Worth a look if you need offline Chinese TTS with Cantonese or Tibetan support, or want to experiment with voice cloning using your own recordings. Pass if you need a batteries-included, actively developed engine with modern deep-learning voices.