Runs in the browser · Speech to text
Run Whisper Large v3 Turbo in your browser
Whisper Large v3 Turbo downloads 537.4 MB over WebGPU (q4f16), or 1.01 GB over WebAssembly (uint8), to transcribe speech to text entirely in the tab with Transformers.js. No install, no server.
Speech to text · onnx-community/whisper-large-v3-turbo
- Parameters
- 808.88M
- Pipeline task
- automatic-speech-recognition
Reading
The second-largest speech to text download in the catalog (537.4 MB).
0.66 MB/M params against a 1.91 MB/M speech to text median: light for its parameter count.
The larger build here clears 1 GB (1.01 GB); browser cache and memory limits make this a real download to budget for, not an instant load.
The quant ladder spans 5.7x: 537.4 MB (q4f16) to 3.01 GB (fp32).
Will it run in your browser?
Checking for WebGPU support…
All measured variants
| Variant | Size |
|---|---|
| q4f16 WebGPU pick | 537.4 MB |
| bnb4 | 680.2 MB |
| q4 | 723.9 MB |
| uint8 WASM pick | 1.01 GB |
| fp16 | 1.51 GB |
| q8 | 2.02 GB |
| fp32 | 3.01 GB |
Sizes measured from the HuggingFace API file tree for onnx-community/whisper-large-v3-turbo, not estimated.
Use it with Transformers.js
import { pipeline } from "@huggingface/transformers";
const pipe = await pipeline("automatic-speech-recognition", "onnx-community/whisper-large-v3-turbo", {
device: "webgpu", // usually falls back to "wasm"; wrap in try/catch for production
dtype: "q4f16", // use "uint8" for the WASM build
});
const result = await pipe(/* your input */);
Requires npm install @huggingface/transformers (or the CDN build).
Sources
Same job, different size
FAQ
WebGPU or WASM for Whisper Large v3 Turbo, in practice?
Chrome, Edge, and Safari 26+ run the WebGPU build (q4f16) on the GPU. Firefox has WebGPU on Windows and Apple Silicon Macs but not everywhere yet. Any browser without a WebGPU adapter falls back automatically to the WebAssembly build (uint8) on CPU: same model, slower to load, slower to run.
What does q4f16 mean for Whisper Large v3 Turbo?
The WebGPU build here uses q4f16: 4-bit weights with some layers kept at 16-bit for stability: WebGPU's usual smallest clean build. The WASM fallback uses uint8: 8-bit integers: about a quarter the size of fp32, with a smaller quality trade than 4-bit.
Sizes measured 2026-08-01 from the HuggingFace API. Last validated 2026-08-03. See all browser models or the methodology.