---
language: tw
license: apache-2.0
tags:
- kenlm
- asr
- twi
- ghananlp
datasets:
- ghananlpcommunity/pristine-twi
---

# Twi KenLM 5-gram Language Model (Quantized (Small))

This is a 5-gram KenLM language model trained on a massive Twi corpus of **278 Million tokens** and a vocabulary of **381,218 unique words**.

It is specifically designed to be used with **Wav2Vec2** or **MMS** (Massively Multilingual Speech) models for Speech-to-Text (STT) tasks to improve transcription accuracy and word correction.

## Model Details
- **Order:** 5-gram
- **Type:** Quantized (Small)
- **RAM Usage:** ~1.3GB
- **Dataset:** GhanaNLP Pristine Twi (999k rows)

## Usage
This model can be integrated into Hugging Face `Wav2Vec2ProcessorWithLM`.

```python
from transformers import Wav2Vec2ProcessorWithLM
# Load your processor and point to this .bin file
```

---
Created with love by **Mich-Seth Owusu** for the **Ghana NLP Community**.