Sangki Kang

Frequency-Domain Double-Talk Detection Based on the Gaussian Mixture Model

In this letter, we propose a novel frequency-domain approach to double-talk detection (DTD) based on the Gaussian mixture model (GMM). In contrast to a previous approach based on a simple and heuristic decision rule utilizing time-domain cross-correlations, GMM is applied to a set of feature vectors extracted from the frequency-domain cross-correlation coefficients. Performance of the proposed approach is evaluated through objective tests under various environments, and better results are obtained as compared to the time-domain method.

Fast communication: An enhanced post-filter for improving the performance of an acoustic echo canceller in mobile application

In this paper, we consider an enhanced post-filter system to improve the performance of an acoustic echo canceller especially in mobile hands-free application. The proposed system includes noise reduction, linear prediction inverse filtering, pitch inverse filtering, and center clipping. The residual echo is first whitened via a linear prediction inverse filter followed by a pitch inverse filter during no near-end talker period. With the aid of these inverse filters, we can remove speech characteristics from the echo and obtain whitened residual echo in result. Then noise reduction technique based on spectral analysis is applied to reduce the power of the residual echo as well as ambient noise. Finally, the level of the residual echo is further attenuated by a center clipper placed in the send path of an echo canceller. For the hands-free application in a moving car, the proposed system attenuated interferences more than 18dB at a constant speed of 30 and 80km/h.

Detection of Basal Cell Carcinoma Based on Gaussian Prototype Fitting of Confocal Raman Spectra

Confocal Raman spectroscopy is known to have strong potential for providing noninvasive dermatological diagnosis of skin cancer. According to the previous work, various well known methods including maximum a posteriori probability classifier (MAP), linear classifier using minimum squared error (MSE) and multi layer perceptron networks classifier (MLP) showed competitive results for basal cell carcinoma (BCC) detection. The experimental results are hard to interpret, however, since the classifiers uses global features obtained by principal component analysis (PCA). In this paper, we propose a method that can identify which regions of the spectra are discriminating for BCC detection. For the purpose, 5 and 7 Gaussian prototypes were built located on the typical peak position of BCC and normal (NOR) tissue spectra respectively. Every spectrum is approximated by a linear combination of the Gaussian prototypes. Decision tree is then applied to identify which prototypes are important for the detection of BCC. Among 12 prototypes, 5 discriminating prototypes were selected and the associated weights were used as an input feature vector. According to the experiments involving 216 confocal Raman spectra, support vector machines (SVM) gave 97.4% sensitivity, which confirms that the peak regions corresponding to the selected features are significant for BCC detection and the proposed fitting method is effective.

Improved Global Soft Decision Using Smoothed Global Likelihood Ratio for Speech Enhancement

In this letter, we propose an improved global soft decision for noisy speech enhancement. From an investigation of statistical model-based speech enhancement, it is discovered that a global soft decision has a fundamental drawback at the speech tail regions of speech signals. For that reason, we propose a new solution based on a smoothed likelihood ratio for the global soft decision. Performances of the proposed method are evaluated by subjective tests under various environments and show better results compared with the our previous work.

Residual echo reduction based on MMSE estimator in acoustic echo canceller

In this paper, a residual echo cancellation method is proposed that uses an estimation of the minimum mean-square error (MMSE) based on a statistical model of a speech signal and an echo signal. After the suppression of the echo signal based on the adaptive filter, residual echo is further reduced by the proposed MMSE estimator and the results are compared with the conventional Wiener filter based method.