---
language:
- en
library_name: mlx
tags:
- asr
- speech-recognition
- apple-silicon
- mlx
- parakeet
- tdt
- sonic-speech
license: cc-by-4.0
base_model: nvidia/parakeet-tdt-0.6b-v2
pipeline_tag: automatic-speech-recognition
---

# Parakeet TDT 0.6B V2 (MLX, BF16)

NVIDIA Parakeet-TDT 0.6B V2 converted to MLX SafeTensors format for
Apple Silicon inference. This is the **reference BF16 checkpoint** — see
quantized variants for reduced memory:

- [`sonic-speech/parakeet-tdt-0.6b-v2-int8`](https://huggingface.co/sonic-speech/parakeet-tdt-0.6b-v2-int8) — Encoder INT8 / Decoder BF16 (recommended)

## Performance (M3 Max, 64GB)

| Metric | Value |
|--------|-------|
| WER (LibriSpeech test-clean) | 1.67% |
| RTFx | 73x realtime |
| Peak memory | ~3GB |
| Parameters | 627M |
| Format | BF16 SafeTensors |

## Usage

```python
from parakeet import from_pretrained

model = from_pretrained("sonic-speech/parakeet-tdt-0.6b-v2")
result = model.transcribe("audio.wav")
print(result.text)
```

## Origin

Weights converted from [`nvidia/parakeet-tdt-0.6b-v2`](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v2)
via the [mlx-community](https://huggingface.co/mlx-community) conversion pipeline.
Hosted by [Sonic Speech](https://sonicspeech.io) for the Sonic voice AI project.

## License

CC-BY-4.0 (following NVIDIA's original license)