Christophe Micheyl on Small and Big Data Challenges for Hearing Aids
That is where machine learning and big data come in. Can we learn from other patients to better fit a new patient? Are users that complain of the same problem likely to need the same parameters? How many different ways can people's hearing differ? Are their common solutions.
I'm very happy to introduce Christophe Micheyl to the Hearing Seminar. Christophe has wide-ranging interests, and I think it will be interesting to see how he can apply machine learning and lots of data to the hearing-aid problem.
Who: Christophe Micheyl (Starkey Research)
What: Small and big data challenges for hearing aids
When: Friday, December 5, 2014 at 11AM
Where: CCRMA Seminar Room, Top Floor of the Knoll
Why: Because there is more to hearing than volume
This talk will be interesting to people interested in perception, connecting humans to sound-processing algorithms, and machine learning. Bryan Pardo talked about a related issue earlier in the year, translating human requests into sound production specifications. This is the same idea, but for the general population, as we are likely to suffer hearing loss at some point.
An estimated 30 million people over the age of 12 in the US have hearing loss [1]. Hearing disabilities can have immediate manifestations, such as communication difficulties, as well as longer-term consequences, such as increased social isolation. Although hearing-aid technology has evolved substantially during the last 30 years, significant challenges remain, both at the diagnosis stage (how can we better characterize hearing disabilities and actual benefits of audio-processing algorithms in everyday life?) and at the remediation stage (how can we devise algorithms that better alleviate the listening difficulties experienced by hearing-impaired listeners?). In this talk, I will illustrate some of these challenges, and point out promising areas for cross-disciplinary investigations at the interface between audiology, psychology (psychometrics and cognitive science), audio signal processing, statistics, and machine learning.
[1] http://www.nidcd.nih.gov/health/statistics/pages/quick.aspx