Software can detect when video-callers are using sign language.
People with speech difficulties relating to Parkinson’s disease, however, often withdraw from conversations, providing the spur for a US-based research team to improve speech recognition software.
In a recent study, 211 patients read specially prepared texts or responded spontaneously to questions about everyday tasks. Over 151 hours, 74,467 answers were recorded and manually transcribed.
Notes detailing patterns were added to 30 responses per participant. Enhancing standardised software with the results of the study reduced the word error rate from 36.3% to 23.7%.
The data from the study are available, allowing other researchers, charitable organisations and companies to improve their own speech-to-text software.
A good companion to “hearing” glasses.
