Time-domain based Temporal Processing with Application of

speech proceesing, speech recognition, TRAP, feature extraction

Time-domain based Temporal Processing with Application of Orthogonal Transformations

In the paper, novel approach that efficiently extracts the temporal information of speech has been proposed. This algorithm is fully employed in time-domain, and the preprocessing blocks are well justified by psychoacoustic studies. The achieved results show the different properties of proposed algorithm compared to the traditional approach. The algorithm is advantageous in terms of possible modifications and computational inexpensiveness. Then, in our experiments, we have focused on different representation of time trajectories. Classical methods that are efficient in conventional feature extraction approaches showed not to be suitable to approximate temporal trajectories of speech. However, the application of some orthogonal transformations, such as discrete Fourier transform or discrete cosine transform, on top of previously derived temporal trajectories outperforms classification in original domain. In addition, these transformed features are very efficient to reduce the dimensionality of data. %in data reduction.