cdli/whisper-large-v3_finetuned_ugandan_luganda_waxal_7_standard_speech_v1.0

This is a fine-tuned version of openai/whisper-large-v3 for Luganda standard speech. It is part of CDLI's effort to make speech technology work for people whose speech is underserved by mainstream ASR systems.

All CDLI models and datasets can be found on CDLI's HuggingFace page.

Dataset

The model has been fine-tuned using google/WaxalNLP. The lug_asr subset of the dataset was used.

Training

The train split was used for training, and the dev split for selecting the best checkpoint.

This Whisper model was fine-tuned and is decoded using the Swahili (sw) language setting — out of all languages Whisper supports, the one most similar to Luganda.

All model parameters (encoder, decoder, and output projection) were fine-tuned, with SpecAugment enabled.

Evaluation

This model was evaluated on the test split of the dataset. Utterances longer than 30 seconds were excluded:

• Examples evaluated: 503
• Speakers: 194

For decoding we ran Whisper with language=sw, task=transcribe, greedy search (num_beams=1, do_sample=False).

We report two complementary word error rate (WER) metrics, both computed on text normalized with Whisper's BasicTextNormalizer:

• Standard (corpus-level) WER — the usual error rate, pooling all reference words and edit errors across the entire test set.
• Per-utterance averaged WER — WER computed separately for each utterance, each capped at 1.0, then averaged across utterances.

The per-utterance averaged WER bounds each utterance to [0, 1] and weights all utterances equally, so it reflects typical performance without a few catastrophic utterances dominating — but it is not a true error rate and isn't directly comparable to other published WER, hence we report the standard, corpus-level WER as well.

Results

Overall Results

Model	Standard WER	Per-utterance averaged WER
This model	0.13	0.13