The objective of this chapter is to verify the identity of the claimed learner by extracting the prosodic features of the speech signal. TIMIT Acoustic-Phonetic Continuous Speech Corpus is used for learner verification using prosodic and articulation features such as energy, pitch, and formants. The prosodic feature includes pitch (F0), and articulation feature includes formants (F1-F7). From this database, for this project in the training phase, 200 learners were used and in the testing phase 160 learners were used. The pitch and formants were extracted using linear predictive analysis. The first seven formants were used for verification purpose. The feature set consists of eight features. The features are fed into the Guassian mixture model. In the Gaussian mixture model, parameters are estimated from the training and testing data using the iterative expectation-maximization. Log likelihood score is computed using these parameters, and then these scores are normalized to make decisions. The decision is made based on the threshold.
Linear predictive analysis of speech using non-Toeplitz autocorrelation matrices.