Show HN: Teach a speech model (en) a name by saying it once, on your Mac
Source: Show HN ·
I wanted to see if a speech model could spell a name better if it heard it, not just read it. Most STT tools handle custom words as text (keyterm list, fixing transcript after with llm, etc). So I trained an adapter to try it, specifically a LoRA on Qwen3-ASR 1.7b (see benchmark). This LoRA transcribes words only after it hears a recording of each of your custom words. To make this model easier to use in day-to-day life, we also built Detail Keeper. It's a Mac app where you can put this adapter to use by specifying difficult-to-transcribe-correctly words and then having them picked up correctl
Frequently Asked Questions
What is Show HN: Teach a speech model (en) a name by saying it once, on your Mac?
Show HN: Teach a speech model (en) a name by saying it once, on your Mac is a tool discovery curated from Show HN. I wanted to see if a speech model could spell a name better if it heard it, not just read it. Most STT tools handle custom words as text (keyterm list, fixing transcript after with
How can I access or experience this?
You can experience this content directly on its original host (Show HN) by clicking the "Open Original" button.
Where does this content come from?
Curated from Show HN by the OnSuggest team to surface the most interesting and serendipitous content on the web.
More like this
All Developer tools discoveries →Show HN: Tyler RSS – Open-source and feature-packed desktop RSS reader
Feature-packed, lightweight, beautiful, open-source RSS reader. Tyler is a free RSS reader for Mac, Windows and Linux, with offline articles, full-text search, and no account.
Show HN: Ryph – Self-Hosted, Open-Source Coding Agents for Incident Response
🤖 Self-hosted coding agents for your team: on call for your incidents, in your chat, on your forge. - ryph-hq/ryph
Show HN: Slate Terminal
A native macOS terminal. Add your own panels, commands and layouts, or ask your agent to build them. For Apple silicon and macOS 27 or later.
Show HN: Coding agent skills, tested on 15 backend tasks with hidden checks
Skills that stop AI agents from writing backend code that passes the demo and breaks in production. Critical mistakes: 25/150 runs without, 2/150 with. Claude Code, Codex, Cursor. - globalmpc/backe...
Show HN: Backwords – Let AI teach you how to speak in reverse
Let AI teach you how to speak in reverse
Show HN: Did prediction markets do any predicting this midterms?
I've heard so much about how prediction markets are the guiding light in this dark world. But when put head to head against polls or Nate Silver's FLIPR model, they are really a wash. Especially looking at flipped or close-by-polls races, there is no alpha compared to a traditional poll-based model. Was this obvious to everyone else too? Can we treat them like the entertainment they are and tax them for the terrible ads? Comments URL: https://news.ycombinator.com/item?id=50401451 Points: 19 # Comments: 8