{"slug": "show-hn-teach-a-speech-model-en-a-name-by-saying-it-once-on-your-mac", "title": "Show HN: Teach a speech model (en) a name by saying it once, on your Mac", "summary": "A developer released a LoRA adapter for Qwen3-ASR 1.7B that transcribes custom names after hearing a single spoken recording of each, packaged in a free local Mac app called Detail Keeper. On the project's open benchmark of audiobooks with 1,100 name mentions, the adapter spelled taught names correctly 89.6% of the time versus 88.2% for Eleven Labs Scribe with keyterms, a difference the author called not measurable, though it overwrites more untaught names and alters slightly more ordinary words. Detail Keeper requires an Apple M-series Mac on OS 26 or later, downloads about 3.6GB of models, works only in English, and the adapter is published on Hugging Face at huggingface.co/virgil-coach/own-names-qwen3-asr.", "body_md": "I wanted to see if a speech model could spell a name better if it heard it, not just read it.\n\nMost STT tools handle custom words as text (keyterm list, fixing transcript after with llm, etc).\n\nSo 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.\n\nTo 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 correctly in your recordings.\n\nWe've found this helpful in two specific cases: 1. Words STT has never heard: invented characters and places, codenames, unusual surnames. 2. Your spelling of a word it does know: Kathryn and not Catherine, Jon and not John.\n\nI wasn't able to achieve both good recognition and recognizing many words (1k+) with the model alone.\n\nDetail Keeper first narrows each passage to the likeliest taught names by sound. Then the adapter writes the passage, and the app checks it against a plain transcript.\n\nOn our open benchmark (audiobooks, 1,100 name mentions) it spells taught names as well as Eleven Labs Scribe with keyterms: 89.6% vs 88.2%, no measurable difference. The only downside is that it overwrites a few more untaught names and changes slightly more ordinary words.\n\nEverything, including where we lose, is at detailkeeper.com/benchmark.\n\nNote: It's free, local and nothing leaves your Mac, but it does require an Apple M series+ mac and OS 26+ currently. Also worth noting that it downloads about 3.6GB of models and only works in English right now.\n\nIf you're interested, you can also find the adapter itself on HF: huggingface.co/virgil-coach/own-names-qwen3-asr\n\nComments URL: [https://news.ycombinator.com/item?id=50383785](https://news.ycombinator.com/item?id=50383785)\n\nPoints: 2\n\n# Comments: 0", "url": "https://wpnews.pro/news/show-hn-teach-a-speech-model-en-a-name-by-saying-it-once-on-your-mac", "canonical_source": "https://detailkeeper.com/", "published_at": "2026-11-05 20:06:33+00:00", "updated_at": "2026-11-05 20:21:18.239455+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "ai-tools", "ai-products", "natural-language-processing"], "entities": ["Qwen3-ASR 1.7B", "Detail Keeper", "Eleven Labs Scribe", "Hugging Face", "Apple"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/show-hn-teach-a-speech-model-en-a-name-by-saying-it-once-on-your-mac", "markdown": "https://wpnews.pro/news/show-hn-teach-a-speech-model-en-a-name-by-saying-it-once-on-your-mac.md", "text": "https://wpnews.pro/news/show-hn-teach-a-speech-model-en-a-name-by-saying-it-once-on-your-mac.txt", "jsonld": "https://wpnews.pro/news/show-hn-teach-a-speech-model-en-a-name-by-saying-it-once-on-your-mac.jsonld"}}