Liang Tao

Unsupervised learning of phonemes of whispered speech in a noisy environment based on convolutive non-negative matrix factorization

This paper focuses on the development of an algorithm that can be optimized for a specific acoustic environment to improve the intelligibility of whispered speech. A new convolutive non-negative matrix factorization (NMF) algorithm is proposed to extract phoneme bases from noisy whispered speech with the noise bases from prior learning; these noise bases are obtained from training using the conventional non-negative matrix factorization. The divergence function with a sparseness constraint term is selected as the objective function in the developed algorithm to obtain multiplicative update rules of the phoneme base matrix and the corresponding weight matrix. The weights of the noise bases from prior learning are also updated in the phoneme learning stage. Listening experiments were conducted to assess the intelligibility performance of speech synthesized using the proposed algorithm. The experimental results indicate that the proposed algorithm is very effective for improving the intelligibility of whispers in various noise contexts, and it outperforms conventional algorithms.

Reconstruction of whisper in Chinese by modified MELP

In the present paper, a method for the real-time conversion of whisper to normal phonated speech in Chinese through a multiple excited linear prediction (MELP) analysis-by-synthesis codec is discussed. In the analysis codec processing of whisper speech, parameters from unchanged MELP model are changed, such as the pitch and the formants which aims to reconstruct the target phonated speech using the synthesis codec. Subjective evaluation results show the effectiveness of the technique.

Novel Algorithm for Hand Vein Recognition Based on Retinex Method and SIFT Feature Analysis

Based on the Retinex method and the SIFT feature analysis, this paper presents a novel algorithm for hand vein recognition. First of all, the principle of the near-infrared hand vein image acquisition is introduced. Secondly, the Retinex method is used to normalize hand vein images, and the adaptive smoothing method is selected to estimate the illumination. Then gray cosine transform is used to enhance the discrimination of the skin and the vein in hand vein images. Thirdly, the SIFT feature analysis algorithm is used to extract the feature of hand vein. Finally, the match method of two hand vein images based on SIFT is given. A hand vein recognition system in Microsoft VC6.0 is also developed and the experimental results demonstrate the high efficiency of the proposed algorithm in runtime and correct recognition rate.

A parallel implementation of Singular Value Decomposition based on Map-Reduce and PARPACK

In the e-commerce on the Web,recommender systems become a powerful technology for extracting valuable information from its customer databases. These systems also help customers find products they want to buy from a business sites. Singular Value Decomposition(SVD) is a useful technology to speedup the recommendations with very fast online performance, requiring just a few simple arithmetic operations. Unfortunately, computing the SVD of a large scale matrix is very expensive. In this paper, we propose to parallelize the SVD algorithm to run on distributed computers. Our parallel algorithm employs a parallel ARPACK algorithm to perform parallel eigenvalue decomposition. Experimental results show that the proposed method can significantly speed up the SVD computation cost while providing comparable prediction quality.

Palmprint Recognition Using Sparse Representation of Variable Window-Width Real-Valued Gabor Feature

This paper proposed a simple but effective palmprint recognition algorithm using improved Real-valued Discrete Gabor Transform (RDGT) and Sparse Representation based Classification (SRC) method. Compared to the existing palmprint recognition methods based on the spatial texture feature of palmprint, the proposed variable window-width real-valued Gabor transform extract the palmprint feature by space-frequency analysis. Given Gauss window as the analysis window, in addition, the window-width is dynamically adjusted according to the local variance of the palmprint image when solving the coefficients of RDGT. Then test sample can be sparsely represented in an overcomplete dictionary composed by training samples. Experimental results on PolyU Palmprint Database and PolyU M_B Database demonstrate the effectiveness of our proposed method.

Robust Visual Tracking Using Oriented Gradient Convolution Networks

Convolutional networks have been successfully applied to visual tracking to extract some useful feature. However, deep networks are time-consuming to offline training and usually extract the feature from raw pixels. In this paper, we propose a two-layer convolutional network based on oriented gradient. The first layer is constructed by the convolution of the filter and an input image of oriented gradient, which is robust to the illumination variation and motion blur. Then, all of the feature maps of the simple layer are stacked to a complex feature map as the target representation. The complex feature map can encode the local structure feature which is robust to occlusion. The proposed approach is tested on nine challenging sequences in comparison with nine state-of-art trackers, and the result show that the proposed tracker achieves mean overlap rate of 0.75, which outperforms the secondary tracker by 26%.

The Voice Conversion Method Based on Sparse Convolutive Non-negative Matrix Factorization

We propose a voice conversion method based on sparse convolutive non-negative matrix factorization. The method utilizes the Itakura–Saito distance as the objective cost function, making the smaller matrix element with a smaller reconstruction error due to the property of scale invariant of the cost function. The time–frequency basis of the source and target were extracted during the training phase, and the speech is converted through time–frequency basis substitution. The effect of whisper-to-normal speech conversion experiment is also conducted. Experimental results show that the proposed voice conversion method outperforms the method based on the conventional convolutive non-negative matrix factorization and the method based on the Kullback–Leibler (K-L) cost function in the aspects of speech intelligibility.

Multirate and DFT Based Fast Parallel Algorithm for 2-D Inverse Discrete Gabor Transform

Multirate and DFT based fast parallel algorithm for the 2-D inverse discrete Gabor transform (IDGT) is presented. A 2-D synthesis filterbank is designed for the 2-D IDGT. The parallel channels in the filterbank have a unified structure and can apply the 2-D fast inverse discrete Fourier transform (IFFT) algorithm to reduce the computational load. The computational complexity of each parallel channel is very low and is independent of the oversampling rate. Thus, the proposed parallel algorithm is attractive for real time image processing.

A Contrast Enhancement Method for Fog-Degraded Images

In a foggy weather, the contrast of images is drastically degraded. This makes some applications, such as video surveillance, very sensitive to weather conditions. This chapter presents a fog-degraded image enhancement method based on a human visual system (HVS). The algorithm utilizes the HVS to segment a single fog-degraded image into the DeVries-Rose region, Weber region, low-contrast, and saturation region with three subimages. With a modified contrast limited adaptive histogram equalization (CLAHE), the contrast of the subimages will be enhanced. The defog experiments will be carried out to illustrate the efficiency of the proposed method for the fog-degraded images.

Whisper Denoising in Joint Time-Frequency Domain Based on Real-Valued Discrete Gabor Transform

Whispered speech can be effectively used for quiet and private communications over mobile phones and is also the communication means for ENT patients under a regime of voice rest. However, little progress has been made on the denoising of whispered speech in noisy environment because of its special acoustic characteristics.In this paper, we propose a whisper denoising algorithm in joint time-frequency domain based on real-valued discrete Gabor transform(RDGT). Noisy whisper is first transformed into the joint time-frequency domain by fast real-valued discrete Gabor transform. The MMSE based log-amplitude estimator is derived under speech presence uncertainty hypothesis. Clean whisper spectral is then estimated by inverse transform of RDGT. Experimental results show that the proposed algorithm is very effective in avoiding the musical residual noise and retaining weak speech components.