Speaker's Age Group Classification and Recognition using Spectral Features and Gaussian Mixture Models
Abhinav Sharma, Anshu Sharma, Gopal Krishna Dwivedi, Pradeep Kumar Juneja, Adarsh Sinha · 2022
A successful age or age-group recognition can serve various fields of the society like medical, forensics, public relations, investigation and identity authentication also. In this work, Age Group Recognition of different Speakers has been performed with the help of extraction of Mel Frequency Cepstral Coefficients (MFCCs) from various speakers of both genders. Gaussian Mixture Models technique has been used for the purpose of Classification. There have been five age groups used for Male and Female speakers separately that are-less than 10 years, 11–20, 21–30, 31–40 and more than 40 years. The accuracy tests have been performed by varying the genders, number of MFCC extracted, number of iterations and number of gausses as well. Extracting more number of features delays the result calculation but may improve the chances of increased accuracy. Results show that average recognition accuracy is resulted to be highest (84.6%) for Male database, 82.2% for the female and least (71.2%) was found for the mixed database of Males and Females. Age group recognition improves when number of MFCC increases and found to be highest for 29 number of MFCC with 100 iterations and 64 components of GMM.