● runs 100% in your browser · no server

Slow speech that stays correct.

Each word is re-synthesized as taught, not stretched — vowels held long, stops kept crisp, sounds isolated then blended. Pick a word, a teaching mode, and a speed.

Loading voice model… 0%

—

Speed2.0× slower

1×1.5×2×2.5×3×

Voice: Piper en_US-lessac-medium (VITS) via onnxruntime-web. Sound-out isolates each phoneme with class-aware timing (vowels stretch, stops held), then blends — all stitched in-browser over one ONNX model. Phonemes + segmentation pre-baked.