Open model
browser-whisper
by Tanpreet Jolly
A JavaScript library that runs Whisper speech recognition in the browser with WebGPU and a WebAssembly fallback, so audio and video files can be transcribed locally with no server.
- Web
Speech engine it runs
- Whisper by OpenAIWhisper large-v3-turbo, English: 3.92%Whisper large-v3-turbo, No noise: 3.78%Whisper large-v3-turbo, 20 dB: 4.2%Whisper large-v3-turbo, 10 dB: 5.51%Whisper large-v3-turbo, 5 dB: 13.92%Whisper large-v3-turbo, 0 dB: 66.09%
Word error rates from our own tests on public recordings: English in a quiet room and with background talk. Lower is better.
Key figures
Every figure about browser-whisper
Each figure is a row of a research dataset. Open it for the exact words of its source or the exact request it was read from.
Other open models and engines
- ABR NiagaraApplied Brain Research
- annyangTal Ater
- ARK-ASRAutoArk (Edge0)
- ASR4ALLFUTO
- AsrServeQuantatirsk
- Audar-ASRAudarAI
- Audio8 ASREdge0
- AuroraASRAuroraAI-Research
- balbal-v1Eray Yapağcı
- Breeze ASRMediaTek Research
- Cactus WhistleCactus Compute
- CMU Sphinx (PocketSphinx)Carnegie Mellon University
- CN-MultiDialect-ASRASLP-lab
- Cohere TranscribeCohere
- Confucius4-R2T2NetEase Youdao
- Coqui STTCoqui
- CrispASRCrispStrobe
- CrisperWhisperNyra Health
- Distil-WhisperHugging Face
- DolphinDataoceanAI
- Duyu-smallKaanAydinli
- ESPnetESPnet developers
- expo-speech-recognitionjamsch
- faster-whisperSYSTRAN
- FireRedASRFireRedTeam
- Fluid Speech MiniFluidVoice Labs
- FluidAudioFluid Inference
- FunASRModelScope
- Gemma 4Google DeepMind
- GigaAMSaluteDevices (Sber)