One-page reference: the name untangle ('faster-whisper' is a server-side CTranslate2 library, NOT a browser engine; the real browser engines are whisper.cpp via WebAssembly and Transformers.js via WebGPU; plus the built-in Web Speech API, which sends audio to Google by default); the in-browser pipeline (microphone -> 16 kHz mono -> Silero VAD gate -> Whisper -> caption, all in the tab); WebAssembly-vs-WebGPU (CPU vs GPU, ~100x, cross-origin isolation COOP+COEP for Wasm threads); the cost ledger (model download 31-182 MB quantized, memory, battery); the real-time-factor check (RTF below 1.0 keeps up live); Moonshine for tight live loops; and the client-vs-server decision table.
Download free PDF
The book · Volume 6 of 7
AI for Video: A Complete Guide to Intelligence in Video Products: From Models and Inference to Computer Vision, Speech, Search, and Generative Video
This cheat sheet maps browser speech recognition: whisper.cpp in WebAssembly versus Transformers.js on WebGPU, the VAD-gated pipeline, model download and memory cost, and real-time factor. Volume 6 covers this in depth: speech recognition and diarization, deployment topologies including on-device, model artifacts and the ML basics behind the choice. Written by Nikolay Sapunov, CEO at Fora Soft.
Running Whisper in the browser and unsure it will keep up in real time?
Tell us your target devices and languages, and we will benchmark in-browser ASR and pick the engine, model size and fallback to the server. 250+ video products shipped since 2005.
You get a realistic estimate, technical recommendations and next steps. Free, no obligations.
