POW

Practical Open Weights

Cohere STT WebGPU makes browser transcription feel practical

Today's useful browser AI story is not just that speech-to-text can run locally. It is that the experience is now simple enough to explain: your web app can use the device's graphics chip to process audio faster, keep the work in the browser, and return text without sending voice data away.

Published on May 2, 2026
Newsletter archive

The easiest way to think about WebGPU is this: it gives a browser app access to the part of your device that is very good at doing many small calculations at once. For speech-to-text, that helps the app process audio more efficiently and turn spoken words into text with less waiting. Because it runs in the browser, the same pattern can be integrated into almost any web browser app that is built for WebGPU, instead of requiring a separate desktop tool.

WebGPU, Simply

WebGPU lets a browser app use the GPU for the heavy lifting

In plain terms, WebGPU helps a website use your device's graphics chip for demanding work. That matters for speech-to-text because the GPU can handle lots of audio-related calculations in parallel, which makes in-browser transcription more responsive and more realistic for product use.

See the browser demo

Why It Matters

This can fit into almost any browser-based product

The important product point is not only speed. A team can integrate this kind of speech-to-text flow into a web browser app such as a dashboard, assistant, note-taker, or support tool, as long as the app is built to use WebGPU and the browser supports it. That makes private speech features easier to deliver without asking users to install something separate.

Explore browser AI

Language Coverage

Cohere Transcribe currently supports 14 transcription languages

The supported languages are English, German, French, Italian, Spanish, Portuguese, Greek, Dutch, Polish, Vietnamese, Chinese, Arabic, Japanese, and Korean. That mix gives browser-based transcription a practical footprint across several major regional and international use cases.

Review the language list

Cohere Transcribe adds a clear product angle to that technical shift. It supports 14 transcription languages today, and it shows why privacy-first browser AI is becoming easier to ship: faster local processing, simpler user flows, and speech features that can live directly inside familiar web products.

Read the Cohere WebGPU guide
Ask the AI for help