Show HN: Speech-to-text in the browser with Whisper and Parakeet, no server

A developer released browser-whisper, an open-source JavaScript library that runs speech-to-text entirely in the browser with no server, using WebGPU, WebCodecs (via mediabunny) and Transformers.js for zero-server inference. The library supports eight models, including whisper-tiny, whisper-base and whisper-small (multilingual, 99 languages), English-only moonshine-tiny, moonshine-base, distil-whisper-small and parakeet-ctc-0.6b, and parakeet-tdt-0.6b-v3, which covers 25 European languages with auto-detection and runs on WebGPU only. Users can pass any BCP-47 language code to skip auto-detection for slightly faster and more accurate transcription, and the bundled Parakeet models from NVIDIA are licensed CC BY 4.0, converted to ONNX and quantized by istupakov, ysdede and onnx-community.

js import { BrowserWhisper } from 'browser-whisper'; const whisper = new BrowserWhisper { model: 'whisper-base' } ; for await const { text, start, end } of whisper.transcribe file { console.log ${start}-${end}s : ${text} ; } Try it in the browser Native WebGPU https://developer.mozilla.org/en-US/docs/Web/API/WebGPU API audio transcription. Connects directly to WebCodecs https://developer.mozilla.org/en-US/docs/Web/API/WebCodecs API mediabunny https://mediabunny.dev and Transformers.js https://huggingface.co/docs/transformers.js for zero-server inference. Models & language support | Model | Languages | language | |---|---|---| | whisper-tiny | Multilingual · 99 languages | ✓ | | whisper-base | Multilingual · 99 languages | ✓ | | whisper-small | Multilingual · 99 languages | ✓ | | moonshine-tiny | English only | — | | moonshine-base | English only | — | | distil-whisper-small | English only | — | | parakeet-ctc-0.6b | English only | — | | parakeet-tdt-0.6b-v3 | 25 European languages, auto-detected WebGPU only | — | Model credits: Parakeet TDT 0.6B v3 https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3 by NVIDIA, licensed under CC BY 4.0 https://creativecommons.org/licenses/by/4.0/ . Changes: converted to ONNX and quantized by istupakov https://huggingface.co/istupakov/parakeet-tdt-0.6b-v3-onnx / ysdede https://huggingface.co/ysdede/parakeet-tdt-0.6b-v3-onnx ; runs with parakeet.js https://github.com/ysdede/parakeet.js MIT . Parakeet CTC 0.6B https://huggingface.co/nvidia/parakeet-ctc-0.6b by NVIDIA, licensed under CC BY 4.0 https://creativecommons.org/licenses/by/4.0/ . Changes: converted to ONNX and quantized by onnx-community https://huggingface.co/onnx-community . Whisper by OpenAI MIT . Moonshine by Useful Sensors MIT . Distil-Whisper by Hugging Face MIT . Lite-Whisper by efficient-speech Apache-2.0 . ONNX exports by onnx-community https://huggingface.co/onnx-community . All third-party notices https://github.com/tanpreetjolly/browser-whisper/blob/main/THIRD PARTY NOTICES.md . Pass any BCP-47 code as language to skip auto-detection — slightly faster and more accurate when the language is known.