---
license: cc-by-4.0
base_model: nvidia/canary-qwen-2.5b
library_name: voixful
tags: [automatic-speech-recognition, apple-core-ai, on-device, palette4, voixful, privovoice]
---

# voixful-canary-qwen-2.5b

Apple **Core AI** (Palette4-quantized) conversion of [`nvidia/canary-qwen-2.5b`](https://huggingface.co/nvidia/canary-qwen-2.5b) — a SALM (Conformer + Qwen3) with the best punctuation/casing, for on-device speech-to-text on Apple Silicon. Used by the [Voixful](https://github.com/AustinJiangH/voixful) engine and [PrivoVoice](https://github.com/AustinJiangH/PrivoVoice).

## License & attribution
Weights derive from `nvidia/canary-qwen-2.5b` (**CC-BY-4.0**, © NVIDIA), redistributed under CC-BY-4.0 with attribution. The Voixful conversion pipeline is Apache-2.0.