What downloads
Public files, stored by your browser.
The model weights, tokenizer, configuration, and browser runtime assets are fetched when requested. Your input is not attached to those file requests.
RunnableTranscription · q8 CPU · fp32/q4 WebGPU
Whisper Tiny English · quantized
Transcribes a short English audio file locally using WebGPU when available or a slower CPU fallback.
Runtime facts
Know what will run before it downloads.
Run Any AI uses the public browser-ready model and exposes the important runtime choices. No Run Any AI backend receives the lab input or output.
- Model ID
- onnx-community/whisper-tiny.en
- Task
- Transcription
- Runtime
- WebGPU with WebAssembly fallback
- Accelerator
- CPU or WebGPU
- Quantization
- q8 CPU · fp32/q4 WebGPU
- Language
- English
- License
- Apache-2.0
- AI API key
- Not required
Use with care
Known limitations
- The first lab release accepts audio up to 60 seconds and targets English speech.
- Accuracy falls with noise, overlapping speakers, uncommon names, and strong accents.
- The CPU fallback can be substantially slower than WebGPU.
Test the model with representative data before relying on its output in a real workflow.