Nicheloom

Market intelligence for builders — see what's gaining traction before it's crowded.

I trained a 9M speech model to fix my Mandarin tones

Details

External ID
46832074
Source
HN
Company
—
Product
I trained a 9M speech model to fix my Mandarin tones
Website domain
simedw.com
Launched
Jan. 31, 2026
Cohort
—
Upvotes
469
Upvotes percentile
0.9920948616600791
Tags
—
Fetched at
Sept. 7, 2026, 9:25 p.m.
Updated at
Sept. 7, 2026, 9:25 p.m.

Description

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 ~300h (AISHELL + Primewords), quantized to INT8 (11 MB), runs 100% in-browser via ONNX Runtime Web.Grades per-syllable pronunciation + tones with Viterbi forced alignment.Try it here: https://simedw.com/projects/ear/

Enrichment

Theme
audio and signal processing tools
Vertical
Education
Function
Model & infra
Audience
B2C
AI stance
AI-native
Project type
Hobby / open-source project
Normalized one-liner
mandarin tone correction speech model
Manually corrected
False

Could you build this?

No Training custom acoustic speech models (Conformer-CTC) on 300+ hours of audio with Viterbi forced alignment and quantization requires specialized machine learning research.

What it would actually take: The stack involves PyTorch/NeMo for training a Conformer-CTC acoustic model on phonetic/tonal Mandarin corpora, implementing custom loss functions for tone deviation, exporting and quantizing to INT8 ONNX, and executing client-side forced alignment via ONNX Runtime Web and Web Audio APIs.

Discussion

20 comments analyzed.

Concerns raised: Accuracy issues with longer sentences and at normal speech pace, Difficulty distinguishing high/mid tone contrasts in tonal languages, Limited learning materials available for some languages like Cantonese, Accent becomes harder to fix after it's ingrained through practice

Feature requests: Add Thai language support, Improve accuracy for longer phrases and natural speech speed, Focus on smaller phrase accuracy before scaling to longer sentences

Competitors

Other products that read as similar to this one — 141 launches clear the similarity bar, closest 8 shown.

Attention rank: #5 of 142 (itself plus its competitors, highest first — normalized so YC and Product Hunt are compared fairly).

Launched 89 days after the earliest competitor.

Other launches for this product

Same idea, different domain

Nobody's really built a model & infra tool for Fintech yet.