Audio context recognition in variable mobile environments from short segments using speaker and language recognizers

The problem of context recognition from mobile audio data is considered. We consider ten different audio contexts (such as car, bus, office and outdoors) prevalent in daily life situa-tions. We choose mel-frequency cepstral coefficient (MFCC) parametrization and present an extensive comparison of six dif-ferent classifiers: k-nearest neighbor (kNN), vector quantiza-tion (VQ), Gaussian mixture model trained with both maxi-mum likelihood (GMM-ML) and maximum mutual information (GMM-MMI) criteria, GMM supervector support vector ma-chine (GMM-SVM) and, finally, SVM with generalized linear discriminant sequence (GLDS-SVM). After all parameter opti-mizations, GMM-MMI and and VQ classifiers perform the best with 52.01 %, and 50.34 % context identification rates, respec-tively, using 3-second data records. Our analysis reveals further that none of the six classifiers is superior to each other when class-, user- or phone-specific accuracies are considered. 1