AudioWeb

Whisper tiny speech-to-text, entirely in the browser, two independent ways:

• /wgsl – hand-written WGSL compute shaders on WebGPU (weights from safetensors).
• /ggml – own Whisper inference on plain GGML, compiled to WebAssembly (SIMD + pthreads).

Weights load automatically: /wgsl from openai/whisper-tiny, /ggml from narendra-orin/whisper-tiny-ggml. Record from the mic, pick a file, or use a bundled sample clip (from FLEURS, CC BY 4.0). /wgsl needs a WebGPU browser (Chrome/Edge).