---
license: apache-2.0

library_name: transformers
tags:
- speech
- audio
- wav2vec2
- automatic-speech-recognition
pipeline_tag: automatic-speech-recognition
---

# OmniASR-CTC-300M-v2 GGUF

GGUF quantizations of **OmniASR-CTC-300M-v2**, a multilingual CTC-based Automatic Speech Recognition (ASR) model supporting over **1,600 languages**.

This repository provides pre-quantized GGUF variants for efficient CPU and edge deployment with **CrispASR** and other GGUF-compatible runtimes.

## Original Model

For Hugging Face Transformers format and usage, see the original [model repository](https://huggingface.co/KevinKibe/omniASR-CTC-300M-v2):


## Available Quantizations

| Quantization | Description                          |
| ------------ | ------------------------------------ |
| `q2_k`       | 2-bit K-quant                        |
| `q3_k`       | 3-bit K-quant                        |
| `q4_0`       | 4-bit quantization                   |
| `q4_1`       | 4-bit quantization                   |
| `q4_k`       | 4-bit K-quant                        |
| `q5_0`       | 5-bit quantization                   |
| `q5_1`       | 5-bit quantization                   |
| `q5_k`       | 5-bit K-quant                        |
| `q6_k`       | 6-bit K-quant                        |
| `q8_0`       | 8-bit quantization                   |

## Usage with CrispASR

### Installation

Follow the CrispASR installation guide:

https://github.com/CrispStrobe/CrispASR#install--build

### Example

```bash
./build/bin/crispasr \
  --backend omniasr \
  --vad \
  -m omniasr-ctc-300m-v2-q8_0.gguf \
  -f audio.wav
```


## Quantization Documentation

For details on GGUF quantization formats and instructions for generating additional quantized variants, see:

https://github.com/CrispStrobe/CrispASR/blob/main/docs/quantize.md

## License

Apache 2.0