This repository contains a Qwen3-ASR checkpoint fine-tuned for Luganda atypical speech from the CDLI Luganda non-standard speech dataset, using the SpecAugment-enabled training notebook.

Training Setup

Base model: KasuleTrevor/cdli-qwen3-asr-lg-typical-1p7b-base-finetune
Dataset: cdli/ugandan_luganda_nonstandard_speech_v1.0
Language: Luganda (lg)
Training source notebook: qwen3-asr-luganda-atypical-cleaned-finetune-spec.ipynb
Learning rate: 5e-5
Scheduler: constant_with_warmup
Epochs: 3
Batch size: 2
Gradient accumulation: 4
Save steps: 250
SpecAugment: enabled

Checkpoint Selection

Notebook outputs show that validation checkpoint comparison selected checkpoint-500 by normalized corpus WER.

Checkpoint	Validation WER	Validation CER	Eval loss
`checkpoint-500`	`0.620471`	`0.311040`	`0.460929`

Test Results

Checkpoint	Test WER	Test CER	Avg WER capped	Avg CER capped
`checkpoint-500`	`0.536493`	`0.259594`	`0.457453`	`0.205291`

Per-Speaker Breakdown

These per-speaker aggregates come from the notebook-displayed grouped analysis for checkpoint-500.

Speaker	Severity	Disorder	Etiology	N	Mean WER	Mean CER	Median WER	Median CER
`UG057`	Severe (frequent breakdowns)	Voice disorder	Acquired hearing impairment	105	`0.7070`	`0.4520`	`0.8000`	`0.3913`
`UG042`	Severe (frequent breakdowns)	Stuttering (Disfluency Disorders)	Developmental	85	`0.6037`	`0.2352`	`0.6364`	`0.2034`
`UG058`	Severe (frequent breakdowns)	Stuttering (Disfluency Disorders)	Developmental	125	`0.5296`	`0.2842`	`0.5294`	`0.2439`
`UG052`	Moderate (requires effort to understand)	Stuttering (Disfluency Disorders)	Developmental	79	`0.4982`	`0.2493`	`0.5000`	`0.1698`
`UG022`	Mild (easily understood with minimal effort)	Articulation Disorders	Developmental	117	`0.4287`	`0.1647`	`0.3750`	`0.0962`
`UG021`	Moderate (requires effort to understand)	Articulation Disorders	Structural presence of akloglosia, simply tong...	91	`0.4251`	`0.1453`	`0.4400`	`0.1266`
`UG036`	Moderate (requires effort to understand)	Dysarthria	Cerebral Palsy	177	`0.3885`	`0.1525`	`0.3636`	`0.1129`
`UG001`	Mild (easily understood with minimal effort)	Dysarthria	Cerebral palsy - cerebral malaria	99	`0.3476`	`0.1357`	`0.3000`	`0.1020`
`UG014`	Mild (easily understood with minimal effort)	Stuttering (Disfluency Disorders)	Idiopathic	149	`0.3133`	`0.1023`	`0.2727`	`0.0500`

Evaluation Notes

The model card is reconstructed from executed notebook outputs and extracted local summaries. Full per-utterance prediction files were referenced in notebook outputs as remote /workspace/... artifact paths, but were not embedded in the notebook JSON available locally.

KasuleTrevor/cdli-qwen3-asr-lg-atypical-stage3-1p7b-typical-base-spec

Qwen3-ASR Luganda Atypical Stage 3 1.7B SpecAugment

Training Setup

Checkpoint Selection

Test Results

Per-Speaker Breakdown

Evaluation Notes

KasuleTrevor/
cdli-qwen3-asr-lg-atypical-stage3-1p7b-typical-base-spec