| # Third-party notices - Voz |
|
|
| The Voz models derive from a third-party speech recogniser. Its license applies to the |
| component named below; nothing in the Desert Ant Labs Source-Available License overrides it. |
|
|
| ## Base model |
|
|
| ### Parakeet TDT 0.6B v3 - NVIDIA |
| - **Source:** https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3 |
| - **License:** [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/legalcode.en) |
| - **Use in Voz:** the acoustic encoder, decoder and token vocabulary in this repository are |
| derived from this model. |
|
|
| ## CC BY 4.0 attribution and modification notice |
|
|
| CC BY 4.0 requires attribution to the creator and a statement of what was changed. The |
| weights here have been: |
|
|
| - converted from NeMo/PyTorch to compiled Core ML (`.mlmodelc`), |
| - compressed by 6-bit palettization, |
| - restructured into the encoder, frontend and decoder graphs this runtime dispatches, |
| with the token embedding table extracted to a sidecar file. |
|
|
| Weight values are otherwise unchanged from the base model. |
|
|
| Rights granted by CC BY 4.0 in the base model are unaffected by this repository. Every |
| recipient of that material receives an offer directly from NVIDIA under CC BY 4.0, and |
| nothing here restricts it. The Desert Ant Labs Source-Available License covers the Desert |
| Ant Labs contributions: the conversion, compression, packaging, and the Swift runtime that |
| drives these graphs. |
|
|
| ## Evaluation data |
|
|
| - **FLEURS** - Google - **CC BY 4.0**. Used only to measure this export for the per-language |
| table in the model card. Dataset: https://huggingface.co/datasets/google/fleurs |
| No FLEURS audio or text is redistributed here. |
|
|
| ## Runtime frameworks |
|
|
| Voz links only Apple system frameworks, including Core ML and AVFoundation. No third-party |
| runtime library is bundled. |
|
|
