Downloads · 30 days
31
14% of all-time downloads
KasuleTrevor/cdli-qwen3-asr-lg-atypical-stage3-1p7b-typical-base-const1e4
cdli-qwen3-asr-lg-atypical-stage3-1p7b-typical-base-const1e4 is a automatic speech recognition model from KasuleTrevor. Use it when you need speech turned into text.
This repository contains a Qwen3-ASR checkpoint fine-tuned for Luganda atypical speech from the CDLI Luganda non-standard speech dataset.
Downloads · 30 days
31
14% of all-time downloads
All-time downloads
227
Public
Parameters
2B
8.2 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors4.1 GB · 100%
From the Hugging Face model README
This repository contains a Qwen3-ASR checkpoint fine-tuned for Luganda atypical speech from the CDLI Luganda non-standard speech dataset.
KasuleTrevor/cdli-qwen3-asr-lg-typical-1p7b-base-finetunecdli/ugandan_luganda_nonstandard_speech_v1.0lg)qwen3-asr-luganda-atypical-cleaned-finetune.ipynb1e-4constant_with_warmup324250Notebook outputs show that validation checkpoint comparison selected
checkpoint-2000 by normalized corpus WER.
| Checkpoint | Validation WER | Validation CER | Eval loss |
|---|---|---|---|
checkpoint-2000 | 0.674941 | 0.365374 | 0.561479 |
A later exploratory notebook cell evaluated checkpoint-1250 after the upload.
That post-hoc test result should not be interpreted as the uploaded best
checkpoint.
The notebook contains test metrics for the later post-hoc checkpoint-1250
evaluation:
| Checkpoint | Test WER | Test CER | Avg WER capped | Avg CER capped |
|---|---|---|---|---|
checkpoint-1250 | 0.750161 | 0.492839 | 0.558925 | 0.327015 |
These per-speaker aggregates come from the notebook-displayed grouped analysis
for the recovered post-hoc checkpoint-1250 test artifact.
| Speaker | Severity | Disorder | Etiology | N | Mean WER | Mean CER | Median WER | Median CER |
|---|---|---|---|---|---|---|---|---|
UG057 | Severe (frequent breakdowns) | Voice disorder | Acquired hearing impairment | 105 | 0.7087 | 0.4463 | 0.8286 | 0.4842 |
UG042 | Severe (frequent breakdowns) | Stuttering (Disfluency Disorders) | Developmental | 85 | 0.6138 | 0.2924 | 0.6667 | 0.2381 |
UG058 | Severe (frequent breakdowns) | Stuttering (Disfluency Disorders) | Developmental | 125 | 0.5332 | 0.3034 | 0.5714 | 0.2571 |
UG052 | Moderate (requires effort to understand) | Stuttering (Disfluency Disorders) | Developmental | 79 | 0.5293 | 0.2623 | 0.5000 | 0.2097 |
UG021 | Moderate (requires effort to understand) | Articulation Disorders | Structural presence of akloglosia, simply tong... | 91 | 0.5230 | 0.2498 | 0.5517 | 0.1929 |
UG022 | Mild (easily understood with minimal effort) | Articulation Disorders | Developmental | 117 | 0.4837 | 0.2311 | 0.5000 | 0.1642 |
UG036 | Moderate (requires effort to understand) | Dysarthria | Cerebral Palsy | 177 | 0.4481 | 0.2001 | 0.4375 | 0.1429 |
UG014 | Mild (easily understood with minimal effort) | Stuttering (Disfluency Disorders) | Idiopathic | 149 | 0.3728 | 0.1271 | 0.3333 | 0.0645 |
UG001 | Mild (easily understood with minimal effort) | Dysarthria | Cerebral palsy - cerebral malaria | 99 | 0.3573 | 0.1542 | 0.3636 | 0.1399 |
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.