Whisper, on your GPU, in this tab
Speech-to-text with OpenAI's Whisper (tiny or base) running on kiln's hand-written WebGPU kernels. Your audio never leaves your device.
This browser doesn't support WebGPU. Try a recent Chrome or Edge on desktop.
1 Load the model
2 Give it some speech (any length)
3 Transcript
Nothing yet.
- audio
- spectrogram
- encoder
- decoder
- speed
Everything above runs locally: the model weights are downloaded from Hugging Face and run on your GPU through WebGPU compute shaders written from scratch — no ONNX, no WASM runtime.