The Interspeech 2025 Speech Accessibility Project Challenge
Paper • 2507.22047 • Published
A fine-tuned variant of openai/whisper-large-v2 trained on the SAPC training set (speakers with Parkinson's disease only) using LoRA.
Compared to 17.82% WER for the openai/whisper-large-v2 baseline, this model achieves 10.46% WER (~41.3% relative improvement) on the SAPC test2 split.
- Downloads last month
- 3
Model tree for xiuwenz2/whisper-large-v2-ft-SAP-0430-pd-merged
Base model
openai/whisper-large-v2