Upload README.md with huggingface_hub

README.md

@@ -1,5 +1,5 @@
 ---
-language: sw
+language: lg
 library_name: transformers
 pipeline_tag: automatic-speech-recognition
 tags:
@@ -12,27 +12,36 @@ metrics:
 - wer
 ---
 
-This is a fine-tuned version of [**`openai/whisper-large-v3`**](https://huggingface.co/openai/whisper-large-v3) for `Ugandan 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.
+This is a fine-tuned version of [**`openai/whisper-large-v3`**](https://huggingface.co/openai/whisper-large-v3) for `Luganda` non-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**](https://huggingface.co/cdli).
 
-## Evaluation
+## Dataset
+
+The model has been fine-tuned using [`google/WaxalNLP`](https://huggingface.co/datasets/google/WaxalNLP), a dataset of speech samples of people living with impaired speech across a range of impairment **severity levels** and **etiologies**. The `lug_asr` subset of the dataset was used.
 
-This model was evaluated on the **`test`** split of [`google/WaxalNLP`](https://huggingface.co/datasets/google/WaxalNLP), a speech dataset for African Languages. The `lug_asr` subset of the dataset was used for fine-tuning and evaluation. Examples longer than 30 seconds were excluded.
-For decoding we ran Whisper with `language=sw`, `task=transcribe`, greedy search (`num_beams=1`, `do_sample=False`). 
+## Training
 
-Note: We fine-tuned the Whisper model for the Swahili (sw) language as out of all languages supported by Whisper, this is the language most similar to Luganda.
+The *train* split was used for training, and the *dev* split for selecting the best checkpoint. **Note:** 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
 
-### Metrics
+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. 
+  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.
 
@@ -41,6 +50,7 @@ The per-utterance averaged WER bounds each utterance to
 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