hey HN, i built an npm library - browser-whisper. it's an abstraction over libraries that lets you transcribe audio/videos completely in the browser using local models, with no api key and no server
it currently supports two runtimes - transformers.js and parakeet.js. think of it like a shadcn ecosystem but for in-browser STT models - you can plug n play this feature into your app with just 3 lines of code. in simple words, it's infra shipped with third-party models, and super fast to run in the browser, powered by WebGPU.
the latest parakeet model, has low enough latency to build simple Wispr Flow-style apps completely free of cost and in browser
repo - github.com/tanpreetjolly/browser-whisper (MIT License). i'd love feedback.
2 comments
[ 3.3 ms ] story [ 10.0 ms ] threadit currently supports two runtimes - transformers.js and parakeet.js. think of it like a shadcn ecosystem but for in-browser STT models - you can plug n play this feature into your app with just 3 lines of code. in simple words, it's infra shipped with third-party models, and super fast to run in the browser, powered by WebGPU.
the latest parakeet model, has low enough latency to build simple Wispr Flow-style apps completely free of cost and in browser
repo - github.com/tanpreetjolly/browser-whisper (MIT License). i'd love feedback.