MIPS Project Detail:
Company
MIPS Project
Voice Classification Models for Public Speaking
Project #
5205.25
|
MIPS Round
55
|
Starting Date:
Feb 2015
MIPS Project Challenge:
The goal of this MIPS project was to help VoiceVibes collect data to help train and validate its predictive “vibes” models so it can provide users with feedback on how their speaking style is received by listeners.
MIPS Role:
Through this MIPS project, researchers collected nearly 2,000 audio samples from various speakers, and had them rated by groups of listeners. Samples were rated in twenty categories called “vibes,” such as confident, boring, captivating, authentic and timid. The majority of the samples and ratings were provided to the company to enable them to train their software models. A portion of the rated samples were retained at Towson University and used in an independent validation study to determine how accurately the VoiceVibes software could predict human reactions. The VoiceVibes software scored above 90 percent accuracy for all 20 vibes when the software predictions were compared to human ratings of the same passages.
Results:
In January, 2016, VoiceVibes began beta testing its MyVoiceVibes product at local colleges and universities. Wor-Wic Community College conducted a series of pilots in sections of its basic communications course, provided detailed product feedback from instructors and students, and eventually became VoiceVibes’ first customer. The company began selling subscriptions in January, 2017, and is now used in over 20 colleges and universities across the country. As of September, 2017, VoiceVibes employed seven people.
Additional Mtech/UMD programs utilized:
UMD Law Clinic
Principal Investigator:
Ellyn
Sheffield
Associate Professor, Psychology
Project Manager:
Debra
Cancro
CEO
Technologies:
Software Development