Show HN: I trained a 9M speech model to fix my Mandarin tones
Published: (January 30, 2026 at 07:51 PM EST)
1 min read
Source: Hacker News
Source: Hacker News
Model Overview
Built this because tones are killing my spoken Mandarin and I can’t reliably hear my own mistakes.
It’s a 9M Conformer‑CTC model trained on ~300 h (AISHELL + Primewords), quantized to INT8 (11 MB), and runs 100 % in‑browser via ONNX Runtime Web.
The model grades per‑syllable pronunciation + tones using a Viterbi force algorithm.