Application of Kohonen self-organizing maps to improve the performance of objective methods for speech quality assessment

A new proposal to improve the performance and effectiveness of objective methods for speech quality assessment is presented. Such proposal uses the Kohonen self-organizing maps (KSOM), also known as Kohonen networks, which increase the efficiency of the mapping process from objective to subjective measures. The validation of this new approach is performed using the Objective Speech Quality Measure (MOQV). A performance analysis of the algorithm allows the comparison with traditional techniques that use third-order polynomials or other monotonic functions. This analysis is used as a base to infer the scope to be assigned to this new mapping technique, in order to extend its application to already existing and future algorithms.