Modelpublic

KasuleTrevor/qwen3-asr-en-atypical-promptless-specaug-encoder

sourceHugging Faceupdated 1mo agoView on Hugging Face

0likes12downloads

Model Card

Qwen3-ASR Atypical English — Encoder Adaptation + SpecAugment

This repository contains the validation-selected encoder checkpoint from the promptless Qwen3-ASR atypical-English architectural ablation.

Experiment

—Base model: Qwen/Qwen3-ASR-1.7B
—Dataset: cdli/ugandan_english_nonstandard_speech_v1.0
—Condition: encoder
—Prompt: none
—SpecAugment: enabled, train-only
—LR: 5e-05
—Scheduler: cosine
—Seed: 42
—Batch / grad accumulation: 4 / 4
—Effective batch: 16

SpecAugment

Setting	Value
Probability	0.5
Time masks	2
Time width	40
Frequency masks	2
Frequency width	12

Trainable parameters

Metric	Value
Total	2,038,052,480
Trainable	314,328,704
Trainable %	15.4230%

Validation checkpoint selection

Selection was performed on the validation split only using mean normalized per-utterance WER/CER capped at 1.0.

Metric	Value
Selected checkpoint	`checkpoint-1500`
Mean capped WER	31.33%
Mean capped CER	22.29%
Corpus WER	30.88%
Corpus CER	21.17%

Final test results

Primary reporting uses normalized per-utterance WER/CER, capped at 1.0 per utterance and then averaged. Normalized corpus metrics are retained as secondary diagnostics.

Metric	Result
Mean capped utterance WER	22.59%
Mean capped utterance CER	14.30%
Normalized corpus WER	23.54%
Normalized corpus CER	14.62%
Relative primary WER reduction vs base	32.45%

Results by severity

Group	n	Speakers	Mean capped WER	Mean capped CER	Corpus WER	Corpus CER
Mild (easily understood with minimal effort)	334	3	20.21%	11.76%	23.12%	13.09%
Moderate (requires effort to understand)	340	3	23.23%	14.32%	22.37%	13.79%
Severe (frequent breakdowns)	339	3	24.30%	16.78%	25.05%	17.11%

Results by disorder

Group	n	Speakers	Mean capped WER	Mean capped CER	Corpus WER	Corpus CER
Articulation Disorders	176	2	21.69%	11.72%	22.79%	12.19%
Dysarthria	253	2	23.67%	14.80%	23.79%	14.86%
Stuttering (Disfluency Disorders)	488	4	21.54%	14.20%	22.23%	14.50%
Voice disorder	96	1	26.72%	18.18%	28.67%	19.60%

Reproducibility

The exact corrected trainer is archived in this repository under:

code/qwen3_asr_sft_fixed_ablation.py

The companion private results repository is:

KasuleTrevor/qwen3-asr-en-atypical-promptless-specaug-results

No raw participant audio is redistributed.