A KD-Tree Based Non-Intrusive Speech Quality Evaluation for Telephony Systems

A new single-ended method for non-intrusive speech quality assessment for telephony applications is proposed and its performance evaluated. The method computes objective auditory distances between perception-based parametric vectors representing portions of the speech signal, whose quality is to be evaluated, and matching reference vectors extracted from a pre-formulated speech reference book or books. The reference books are constructed by efficiently partitioning a large number of perception-based parametric speech vectors extracted from a database of clean speech signals, using an optimised KD-tree data structure. The measured auditory distances are then mapped into objective listening quality scores. Reported evaluation results show that the method oifers sufficiently accurate and low-complexity speech quality evaluation method with potential for real time applications.