webml-kit
v0.3.0
Speech Recognition
Ready
AudioWorklet capture resampled to 16kHz mono. Chunks to in-browser Whisper.
Model
Hello! I am running locally in your browser with WebGPU. Ask me anything.
0 tok/s · TTFT: 0ms
Click to upload or pick a preset:
Model
TTS Model
Voice Style
Speaking Rate
Generates raw Float32Array waveform samples and plays via Web Audio API.
Device Specs & Limits (detectDevice)
Backend: --
Max Buffer: --
Architecture: --
Recommended Dtype: --
automatic-speech-recognition
PIPELINE_REGISTRY (18 Built-in Tasks)
| Task | Default Model | Dtype | Streaming | KV Cache |
|---|
| Model ID | Task | Downloads |
|---|---|---|
| Click Search Hub to explore compatible models. | ||
Browser Cached Models
| Cache Namespace | Files | Action |
|---|
GPURecovery State Machine
STATE: IDLE
WebGPU devices can be lost due to TDR resets, VRAM exhaustion, background tab suspension, or mobile OS reclamation.
GPURecovery provides exponential backoff auto-recovery with fallback to WASM.
[init] GPURecovery active. Watching WebGPU device (maxRetries: 3, baseDelay: 1000ms)...
webml() code