The 2017 Speech Industry Star Performers: Canary Speech

Canary in the Clinic: Canary Speech Helps Diagnose Maladies

In addition to Beyond Verbal, another company advancing the use of speech technology in the diagnosis of physical and mental maladies is Canary Speech, a U.S. start-up whose work with speech recognition and machine learning is being used to detect the early onsets of Parkinson’s, Alzheimer’s, and dementia. The company has also worked with sports professionals to use speech analysis to identify when a player has suffered a concussion. Other efforts have involved looking for vocal markers that could indicate depression, stress, and dyslexia.

Canary Speech’s technology examines large amounts of speech recorded from patients with particular conditions, comparing current recordings with ones made before patients were diagnosed. It takes into account not only what was said but also how it was said, with particular emphasis on the kinds of words used, their phrasing, and the overall quality of the speech. With regard to Parkinson’s, for example, a common symptom of the disease is a softening of the voice, which is detectable by Canary’s technology but often missed by the human ear.

The company hopes that by helping doctors detect the warning signs of certain conditions much earlier, it can help them begin treating patients earlier as well.

As it works toward that goal, Canary Speech—which was founded in 2016 and is based in Spanish Fork, Utah—recently began working with an unlikely and controversial source of audio data on which to train its algorithms: The company is working with a very large American healthcare and insurance provider, which it didn’t identify by name. The insurer is providing Canary with hundreds of millions of phone calls that have been collected during the past 15 years and information about the speakers’ medical histories and demographic backgrounds. The files contain people with and without diseases, and Canary is using signal processing and machine learning to identify even the most subtle of differences in the audio signals to identify conditions.

“Canary Speech should be recognized for their innovative ideas for detecting medical conditions by analyzing patients’ speech,” says Deborah Dahl, president of Conversational Technologies and chair of the World Wide Web Consortium’s Multimodal Interactions Working Group. “Tools for reliable early detection of Alzheimer’s in particular could be enormously helpful in reducing the devastating social and financial impact of this disease, estimated by the Alzheimer’s Organization to affect more than 5 million Americans. This kind of innovative thinking also has the potential to be extended to developing tools to assist in diagnosing many conditions which affect both speech and language.”