As the amount of available audiovisual data in digital formats is increasing, automatic scene change detection becomes paramount. Many studies have been proposed to treat it recently. Nevertheless, none of these techniques consider the audio signals under a low SNR noisy environment. In this paper, a hierarchical scene change-detection scheme is adopted to detect the scene change automatically under a low SNR noisy environment. The proposed algorithm contains three steps: A statistical model-based audio activity detection scheme that employs the likelihood ratio test is used to segment the audio signal into pure noise segments and noisy audio segments in the first step. In the second step, a noisy robustness feature is adopted to construct a K-Nearest Neighbor (KNN) based classifier and segments the noisy audio segments into speech and music segments. The feature we proposed is called the likelihood ratio crossing rate (LRCR) derived from the likelihood ratio waveform that is obtained in the first step. In last step, a novel speaker change detection based on modified Bayesian information criterion (BIC) is performed to detect the speaker change points for each speech segment. A series of tests were conducted showing the advantage of the proposed scheme. Furthermore, it also shows the robustness under a low SNR noisy environment.

A speaker-independent isolated word recognizer is proposed. It is obtained by concatenating a Bayesian neural network and a Hopfield time-alignment network. In this system, the Bayesian network outputs the a posteriori probability for each speech frame, and the Hopfield network is then concatenated for time warping. A proposed splitting Learning Vector Quantization (LVQ) algorithm derived from the LBG clustering algorithm and the Kohonen LVQ algorithm is first used to train the Bayesian network. The LVQ2 algorithm is subsequently adopted as a final refinement step. A continuous mixture of Gaussian densities for each frame and multi-templates for each word are employed to characterize each word pattern. Experimental evaluation of this system with four templates/word and five mixtures/frame, using 53 speakers (28 males, 25 females) and isolated words (10 digits and 30 city names) databases, gave average recognition accuracies of 97.3%, for the speaker-trained mode and 95.7% for the speaker-independent mode, respectively. Comparisons with K-means and DTW algorithms show that the integration of the splitting LVQ and LVQ2 algorithms makes this system well suited to speaker-independent isolated word recognition. A cookbook approach for the determination of parameters in the Hopfield time-alignment network is also described.