Study probes ASR model for dysarthric speech effects

Researchers have conducted a layer-wise probing analysis on a transformer Automatic Speech Recognition (ASR) model using Mandarin dysarthric speech. The study found that phoneme boundary information remains weak across all layers for dysarthric speech, while phoneme identity becomes more discernible in upper layers. Recognition difficulty is encoded in the deepest layers, and lexical tone is a persistent challenge. The findings suggest that disordered speech impacts higher-level representations more significantly than low-level acoustic features, guiding the development of parameter-efficient fine-tuning methods. AI

IMPACT Provides insights into how ASR models process dysarthric speech, potentially improving performance for low-resource languages.

RANK_REASON Academic paper detailing a layer-wise probing analysis of an ASR model. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Darwin Jelestin Muthu
• Gotit.pub
• Hugging Face
• Lora
• ScienceCast
• Standard Chinese

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →