MMS-300M-Ugandan — Ugandan Languages ASR

Fine-tuned for automatic speech recognition on Ugandan languages.

Languages

Japadhola, English, Luganda, Runyakore, Acholi

Training Details

Property	Value
Base model	facebook/mms-300m
Training approach	Adapter-based fine-tuning (~2M trainable params), base model frozen
Train samples	5545
Test samples	158
Final WER	100.0%
Trained on	2026-05-21 19:00
GPUs	4× GPU with DDP

Supported Languages

• Japadhola (Luo/Nilotic)
• English
• Luganda (Bantu)
• Runyakore (Bantu)
• Acholi (Luo/Nilotic)

Usage

from transformers import pipeline
pipe = pipeline("automatic-speech-recognition", model="Bateesa/mms-300m-ugandan")
result = pipe("your_audio.wav")
print(result["text"])