A V/UVSPEECHDETECTIONBASEDON CHARACTERIZATIONOF BACKGROUND NOISE

Thepaper presents anadaptive system forVoiced/Unvoiced (V/UV)speech detection inthepresence ofbackground noise. Genetic algorithms wereusedtoselect thefeatures that offer thebestV/UVdetection according tothe output ofabackground NoiseClassifier (NC)andaSignaltoNoiseRatio Estimation (SNRE)system. Thesystemwasimplemented andthetests performed using the TIMITspeech corpus andits phonetic classification. The results werecompared withanon-adaptive classification system andtheV/UVdetectors adopted bythree importantspeech coding standards: LPC10,ITU-TG.723.1 and ETSIAMR.Inall cases theadaptive V/UVclassifier outperformed thetraditional solutions.