On device speech-to-text for real-time transcription

Run fast, accurate speech-to-text directly on your devices with Shunya Labs’ ONNX-based English ASR model

Tiny ONNX model, big performance

Lightweight

Small enough to fit on edge devices and existing servers without a hardware refresh

Fast

Transcribes as people speak, with sub-100 ms latency for partials

Accurate

Trained on high entropy data to achieve industry best 3.10% WER

The fastest way to add voice AI to your products

One platform for speech in and speech out—secure by design, built to scale.