Speech detection using Real Adaboost in car environments
Tetsuya Takiguchi, Hiroyoshi Matsuda, Yasuo Ariki · The Journal of the Acoustical Society of America · 2006
In real noisy environments, a speech detection algorithm plays an especially important role for noise reduction, speech recognition, and so on. In this paper, a speech/nonspeech detection algorithm using Real Adaboost is described, which can achieve extremely high detection rates. Boosting is a technique of combining a set weak classifiers to form one high-performance prediction rule, and Real Adaboost [R. E. Schapire and Y. Singer, Mach Learn. 37, 3, 297–336, (1999)] is an adaptive boosting algorithm in which the rule for combining the weak classifiers adapts to the problem and is able to yield extremely efficient classifiers. The Real Adaboost algorithm is investigated for speech/nonspeech detection problem. The proposed method shows an increasing speech detection rate in car environments, comparable with that of the detector based on GMM (Gaussian mixture model), where the detection accuracy rate was 98% for the proposed method and 92% for GMM in a car at highway speed. The results of the experiments clarified the effectiveness of the proposed method.