Google announced Project Euphonia on May 7, 2019, a research effort to make automatic speech recognition work for people whose speech is hard for standard systems to understand. The target conditions include neurological impairments such as ALS, stroke, multiple sclerosis, traumatic brain injury, and Parkinson’s disease, which can cause dysarthria and other atypical speech patterns.
The technical approach converts recorded voice samples into spectrograms and trains models to recognize the speaker’s particular patterns, in many cases building personalized recognition models for an individual rather than relying on a one-size-fits-all system. Google partnered with the ALS Therapy Development Institute and the ALS Residence Initiative to gather recordings, and the launch featured Google researcher Dimitri Kanevsky, who is deaf, and Steve Saling, who has lived with ALS for years and controls devices with non-speech sounds. The team later expanded data collection through the Speech Accessibility Project, a cross-industry effort hosted at the University of Illinois.
Why business readers should care: Euphonia is a reminder that mainstream speech recognition is trained on mainstream voices, and that the long tail - accents, disabilities, atypical speech - is left behind unless someone deliberately collects data for it. Closing those gaps is both an accessibility obligation and a market most vendors ignore.