A V/UV Speech Detection based on Characterization of Background Noise

The paper presents an adaptive system for voiced/unvoiced (V/UV) speech detection in the presence of background noise. Genetic algorithms were used to select the features that offer the best V/UV detection according to the output of a background noise classifier (NC) and a signal to noise ratio estimation (SNRE) system. The system was implemented and the tests performed using the TIMIT speech corpus and its phonetic classification. The results were compared with a non-adaptive classification system and the V/UV detectors adopted by three important speech coding standards: LPC10, ITU-T G.723.1 and ETSI AMR. In all cases the adaptive V/UV classifier outperformed the traditional solutions.