---
language: en
pipeline_tag: automatic-speech-recognition
tags:
  - high-vibes
  - parakeet
  - onnx
  - quantized
  - int4
  - mobile
license: other
---

# High Vibes Voice Models

This repository hosts voice model artifacts for the High Vibes app. The
primary target is on-device speech-to-text with a mobile-friendly download
size.

## Included Artifacts

### Parakeet CTC 0.6B (Quantized)
- Model ID: `parakeet-ctc-0.6b-quant`
- Format: ONNX with external data
- Quantization: MatMulNBits int4 + fp16 conversion for remaining weights
- Intended usage: mobile speech-to-text (offline)

## Source Model

The quantized artifact is derived from `nvidia/parakeet-ctc-0.6b` via
NVIDIA NeMo export tooling and ONNX Runtime quantization.

## Manifests

Platform manifests live here:
- `manifests/ios.json`
- `manifests/android.json`

## Notes

These artifacts are intended for use inside High Vibes. They are not
provided as training checkpoints.