New Spanish speech corpus targets ASR for neurological conditions

Researchers have introduced S-DiverSe, a new corpus of Spanish speech data specifically designed to improve automatic speech recognition (ASR) for individuals with neurological conditions. The dataset comprises 3.2 hours of in-the-wild recordings from 22 Spanish speakers affected by amyotrophic lateral sclerosis, Parkinson's disease, and stroke. S-DiverSe includes 444 transcribed audio segments with metadata on speaker characteristics and condition, aiming to facilitate ASR development and evaluation for this challenging speech domain. Initial experiments suggest that text post-processing methods are more effective than fine-tuning for adapting ASR models to this specialized data. AI

IMPACT This dataset could lead to more accessible speech recognition technology for individuals with neurological conditions.

RANK_REASON The item describes a new academic paper and dataset release. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• amyotrophic lateral sclerosis
• Fernando López
• Hugging Face
• Parkinson's disease
• S-DiverSe
• Spanish Diverse Speech
• speech recognition
• stroke

• paper
• other

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