Norynta Data › Language data index › Cantonese
Cantonese (粵語) speech data for AI
How much open Cantonese speech data exists, where the gaps are, and how to get consented Cantonese data collected.
Cantonese is relatively well covered, but open data is mostly short read-aloud sentences. Natural conversation, call-centre and domain audio, children, older speakers and regional accents are still hard to find.
Source: Mozilla Common Voice language statistics, fetched 2026-11-05. Figures cover the open Common Voice corpus only.
Cantonese orders
We are not taking new buyer orders for Cantonese at the moment. Other languages: tell us what you need.
Speak Cantonese?
粵語
Join the contributor waitlist and we'll email you paid Cantonese recording and writing projects when they open. You record on your phone and are paid once your work is accepted.
Questions
Where can I get Cantonese speech data for AI training?
Mozilla Common Voice has 204 validated hours of Cantonese (as of 2026-11-05). Norynta is not taking new buyer orders for Cantonese at the moment.
How much open Cantonese speech data is available?
As of 2026-11-05, Mozilla Common Voice lists 204 validated hours and 310 recorded hours of Cantonese from 1,196 speakers, ranking 28 of 445 languages by validated hours.
Can I get paid to record Cantonese?
Yes. Native Cantonese speakers can join Norynta's contributor waitlist at collect.norynta.com/join. We email paid recording and writing projects in Cantonese when they open, and pay once the work is accepted.
Languages with similar coverage
- Russian 259 h
- Chinese (China) 240 h
- Tamil 235 h
- Abkhaz 217 h
- Portuguese 185 h
- Polish 177 h
- Thai 173 h
- Central Kurdish 139 h