A new government ID tool — comedy. Researchers have a laughter biometric recognition algorithm
Now it is your laughter that can give you away. A Nigerian research team say they have been able to sort out people through the biometric digital signature of each person’s laugh.
A trio of scientists from the University of Lagos in Lagos, Nigeria, wrote a prototype algorithm that is 90 percent accurate in matching laughter to a person. Using the software with another biometric approach boosted accuracy by 5.06 percent.
The researchers have written a paper cached by InderScience Publishers describing how they used statistical analysis of audible frequencies unique to each laugh. A Gaussian Mixture Model, which has been deployed to extract information from speech data, was 65 percent accurate.
“Mel frequency cepstral coefficients (MFCC) features were extracted, and a Kruskal-Wallis tests performed on each coefficient,” according to a summary of the work.
A dynamic-average Mel frequency cepstral coefficients was created by the team from “typical MFCC features for system training using (a) Gaussian Mixture model (GMM) and support a support vector machine” algorithm.
Combining the Gaussian model with their prototype algorithm boosted accuracy by 5 percent.