Observed arrival · 2026-11-22
Koelony’s Cantonese models keep English in the transcript
A Hong Kong team publishes open speech-recognition models tuned to transcribe colloquial Cantonese in Traditional Chinese while preserving English words.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Hong Kong Cantonese speakers and speech-model builders
- Worth noticing
- The models are released under Apache-2.0, and the page lists GGUF, FP8 and NVFP4 builds.
Field notes
The page links to model weights, benchmarks and quantizations, and reports 77 GitHub stars. It identifies the releases as Apache-2.0 and lists GGUF, FP8 and NVFP4 builds. OraType, described as a free and open-source macOS voice-typing app, ships a GGUF model as its built-in Cantonese speech engine. Other capabilities, including text-to-speech and embeddings, appear in the roadmap as in training rather than released.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
◎NicheUnusually specific use
One card from the complete issue