Speech detection method for VDR based on formant-consonance characteristic and SVM

Speech detection for voyage data recorder(VDR) has some special requires.After the analysis of speech generation process and the character of all kinds of sounds in steer house on ships,a method based on formant-consonance characteristic and support vector machines(SVM) was developed.First formant frequency was obtained by linear prediction analysis(LPA),and the number of consonance and standard deviation of consonance frequency intervals were obtained from residual error of LPA.These were the main characters for speech detection.The SVM was chosen to classify the speech and non-speech.We test the method in experiments by the speech and non-speech signals collected in steer house,and high accuracy rate of detection was obtained.