Community-Supported Shared Infrastructure in Support of Speech Accessibility.
Mark Hasegawa-Johnson, Xiuwen Zheng, Heejin Kim and 22 others
PMID 39325951WHAT IT FOUND
A new speech dataset built from 211 people with Parkinson's disease helps computers understand their speech better than existing systems.
The dataset is available for research, but the authors did not test how well this improves daily communication tools for patients.
Key findings
01The Speech Accessibility Project (SAP) dataset contains speech from 211 people with Parkinson's disease, which is roughly ten times larger than previous datasets for this purpose.
02A speech recognition system trained on this new Parkinson's dataset made fewer errors when recognizing speech from other Parkinson's patients compared to a system trained on general speech.
03Participants were recruited based on self-reported Parkinson's diagnosis and screened by speech-language pathologists for speech or voice impairment greater than normal limits.
STILL TO COME
How it was doneWhat they foundWhat it means for SLPs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
The study reports on a dataset and preliminary technical tests, not a clinical trial of a treatment's effect on patient communication or quality of life. The diagnosis of Parkinson's disease was self-reported by participants and not independently confirmed by medical records. The ASR performance was tested only on speech from people with Parkinson's disease; it is unclear how this system performs for other speech disabilities. The study does not report how well these ASR improvements translate to real-world use by patients in daily life.
Declared interests
The work was supported by a grant from the AI Accessibility Coalition, which includes major tech companies (Amazon, Apple, Google, Meta, Microsoft), and by NIH and NSF funding.
The easy way to misread this
Do not interpret the improved ASR error rates as evidence that patients will experience better communication outcomes in daily life. The study measured technical accuracy on a dataset, not patient-reported success with assistive devices.