theoracleguy/parakeet-tdt-0.6b-v3
MLX-compatible version of nvidia/parakeet-tdt-0.6b-v3, prepared for use with OpenVox and other Apple Silicon speech-to-text workflows.
Parakeet TDT 0.6B V3 is NVIDIA's multilingual automatic speech recognition model based on a FastConformer encoder and Token-and-Duration Transducer (TDT) decoder.
It supports transcription across multiple European languages and is designed for fast, accurate speech recognition.
For architecture details, benchmarks, training information, limitations, and intended use, see the original NVIDIA model card.
OpenVox
This model is distributed through the theoracleguy Hugging Face repository for use with OpenVox, a local AI voice platform for Mac, iPad, and Windows.
OpenVox uses local AI models for supported speech workflows, allowing transcription and other processing to run directly on your device.
This repository can also be used independently with compatible MLX tools such as mlx-audio and parakeet-mlx.
Use with MLX
mlx-audio
Install:
pip install -U mlx-audio
Run transcription:
python -m mlx_audio.stt.generate \
--model theoracleguy/parakeet-tdt-0.6b-v3 \
--audio audio.wav \
--output somewhere
parakeet-mlx
Install:
pip install -U parakeet-mlx
Run:
parakeet-mlx audio.wav \
--model theoracleguy/parakeet-tdt-0.6b-v3
Supported Languages
The model supports:
- English
- Spanish
- French
- German
- Bulgarian
- Croatian
- Czech
- Danish
- Dutch
- Estonian
- Finnish
- Greek
- Hungarian
- Italian
- Latvian
- Lithuanian
- Maltese
- Polish
- Portuguese
- Romanian
- Slovak
- Slovenian
- Swedish
- Russian
- Ukrainian
For language-specific performance and limitations, refer to the original NVIDIA model card.
Original Model
This repository is based on:
The original model was converted to MLX format using the community conversion workflow originally documented here.
License & Attribution
The original Parakeet TDT 0.6B V3 model is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.
This repository contains a format conversion of the original model weights. The original license and applicable attribution requirements continue to apply.
Original model © NVIDIA Corporation.
MLX inference is supported through tools including mlx-audio and parakeet-mlx.
This repository is maintained by The Oracle Guy for use with OpenVox and the wider MLX community.