Developer Creates 9M-Parameter On-Device Model for Mandarin Pronunciation Feedback
By
simedw
4mo ago· 6 min readen
100/100
Golden Brown
Bagelometer↗
Crisp on the outside, thoughtful on the inside. A keeper.
Score100Typehow-toSentimentpositive
Summary
A developer created a 9M-parameter CTC model trained on ~300 hours of transcribed Mandarin speech to grade pronunciation and tones. The tool addresses the author's personal struggle with Mandarin tones and the difficulty of self-correction without a teacher. The model is small enough to run on-device and provides pronunciation feedback, with a demo available online.
Key quotes
· 3 pulledMandarin pronunciation has been hard for me, so I took ~300 hours of transcribed speech and trained a small CTC model to grade my pronunciation.
Part of the problem is tones. They're fairly foreign to me, and I'm bad at hearing my own mistakes, which is deeply frustrating when you don't have a teacher.
A tiny on-device speech model to grade Mandarin pronunciation and tones.
A tiny on-device speech model to grade Mandarin pronunciation and tones.
