On-device deployment for enterprise voice AI

On-Device currently supports Text to Speech and Speech to Text, optimized for constrained edge hardware.

Yes. Speech generation runs entirely on the device, with no network round trips. Fully offline configurations are supported, and no audio or speech content ever touches the cloud.

Modern CPU and ARM-based chips, entry-level GPUs, and NPUs, designed for tight memory and storage budgets. Share your target device and we'll match it against our existing benchmarks.

On-device text to speech supports more than 30 languages, including English, Mandarin, Hindi, Spanish, Japanese, German, and French. Models can be fine-tuned for additional languages depending on the deployment.

Deployments ship with a fixed set of pre-built voices, and custom branded voices can be developed in collaboration with our Voice Design team. Instant voice cloning is not supported on-device.

We've benchmarked our on-device models across a wide range of hardware and can share results from tested devices closest to your target, giving you realistic latency, footprint, and quality expectations without a hardware engagement. If validation on your specific device or with your branded voices is required, we run that as a structured paid proof of concept with dedicated engineering support.

Case-by-case based on device fleet size and use case, typically annual per-device licensing for connected environments, with alternative structures for fully offline deployments. Contact sales to scope your project.