Nicheloom visits: 66,996 all-time · 9,621 in the last 7 days

The opportunity tracker for new startups.

Today: 1,339 launches · Niche of the day: game save editors and managers · See what launched today →

15MB transcription model that runs in the browser/device

Get picks like this daily. Today's launch numbers, the top launches, AI and tech news, and a daily opportunity pick, straight to your inbox.

Want to reach these readers? Sponsor an issue.

I've been obsessed with getting useful transcription models that can be used by websites directly. Starbright is <15MB in size and works even on modest phones, chromebooks, etc. At the moment, I've only got an English model but hopefully more languages can come soon.

HN · launched Nov 19, 2026 · 7 upvotes (top 58%) · 0 views

Niche: speech and voice ai models · 156 launches, 11 in the last 7 days ▲ +22% · Explore this niche →

Category and tags

  • Horizontal
  • Model & infra
  • Developer
  • AI-native

Visit website ↗ Source listing ↗

Could you build this?

Partial Running an existing compact speech model in the browser is vibe-codeable with WebAssembly/ONNX, but building a sub-15MB English model that works on weak devices is a model-compression and training task that needs ML expertise.

What it would actually take: The stack would be a quantized small ASR model (for example a distilled Whisper variant, Vosk-style model, or a custom CTC model) exported to ONNX or WebAssembly and run in a Web Worker with WebGPU or WASM fallback. The hard part is producing acceptable accuracy under about 15MB through pruning, quantization, and distillation, then evaluating on varied accents and noise. That requires speech-ML engineers and training data.

Discussion

6 comments analyzed.

Competitors mentioned: Whistle, Rift transcription

Concerns raised: 25x speed claim is unclear, claim seems too good to be true, model license requires request and forbids redistribution, no API or documentation

Feature requests: API and documentation, fix bold styling on settled text, keyword boosting for names

Competitors

Other products that read as similar to this one : 126 launches clear the similarity bar, closest 8 shown.

Attention rank: #42 of 127 (itself plus its competitors, highest first, normalized so YC and Product Hunt are compared fairly).

Launched 383 days after the earliest competitor.

The niche, labels, vibe code rating and competitor comparison on this page are analysis by Nicheloom. If you use them, please credit nicheloom.com. How you may use our data.