Zou Cai-rong

Modified quadratic discrimination function for non-normal distribution and its application in speech emotion recognition

The quadratic discrimination function (QDF) is a common method widely used in pattern recognition. It is optimal when the sampled features are normal with known means and covariance matrices. In this paper, a new transform is proposed and a method named modified QDF (MQDF) is deduced aim to improve the non-normality of the features during the speech emotion recognition, both prosodic and voice quality features are considered in the experiment. The results show that our method is effective and robust.

An Acoustic Echo Cancellation Technique Based on Kurtosis Subgradient Projection Algorithm

Based on the technique of projection onto convex sets (POCS), a kurtosis subgradient projection (KSP) algorithm is proposed in this paper which is applied in the scheme of the acoustic echo cancellation successfully. KSP algorithm utilizes the kurtosis of error signal after adaptive filtering as the projection area and expands this projection area to half space by using the definition of subgradient. Finally a simply projection expression can be got by projecting onto this half space. Experiments indicate that not only the convergence speed is increased but also the robustness to noise is enhanced compared to traditional method. Additionally, system performance gets promoted evaluated either by mean square error (MSE) or by echo path misalignment.

Acoustic Feedback Cancellation Based on Adaptive Projection Subgradient Method in Hearing Aids

Adaptive projection subgradient method(APSM) is used in this paper for feedback cancellation in hearing aids. APSM is derived from the technique of projection onto convex sets (POCS) and parallel subgradient projection algorithm (PSP). It utilizes the subgradient projection hyperplanes as the searching area of the relax factor in the process of relaxed projection. We applied this APSM algorithm for acoustic feedback cancellation in hearing aids. Numerical experiments demonstrate that notable improvements are achieved including speed, stability and accuracy of convergence compared to the traditional NLMS algorithm. Another exciting conclusion by further experiments is that APSM achieves excellent performance in the situation that real speech segment as input in hearing aids, and APSM is more robust for low SNR compared to other algorithms.

An Integrated Frequency Lowering Algorithm for Sensorineural Hearing Impairment

An integrated frequency lowering algorithm is proposed to increase the audibility of high frequency signal and alleviate the spectral masking associated with sensorineural hearing loss, thus increase the intelligibility of the speech. Firstly, the input signal is transformed into frequency domain by FFT algorithm. Secondly, the energy of the low frequency band and high frequency band is calculated to determine whether frequency shifting is required to increase the audibility of consonants. Thirdly, the spectral samples in each bark-scale channel are compressed towards the band center by a constant factor to reduce the effect of spectral masking. Finally, frequency envelope shaping is utilized to increase the audibility of the whole frequency band. Five subjects with moderate sensorineural hearing loss participate the test. Experimental results demonstrate that the proposed algorithm out perform the conventional frequency lowering methods.

Design and implementation of OFDM modem for low voltage narrowband powerline communication

Narrowband powerline communication has the great potential market application like Auto-Meter-Reading (AMR). This paper focuses on the design and implementation of narrowband powerline communication module with the OFDM technology. Firstly, it introduces the OFDM solution for low voltage powerline communication module design, then it describes the implementation of the sync module, Viterbi code translation module and FFT module based on FPGA platform in details. Finally, it provides the actual test results for the communication module. It shows, the communication data rate of the narrowband powerline communication is greatly improved with OFDM technology.

A new method of multichannel loudness compensation for hearing aids

Loudness compensation is one of the most important functions of hearing aids. Because hearing loss is commonly more severe for high-frequency signal than low-frequency signal, the gain of the loudness compensation should be modified with the frequency of the signal. In this paper, a new method for multichannel loudness compensation based on human auditory is proposed. Then design algorithm for near perfect reconstruction oversampled filter banks is presented, and the parameters of the sub-band filters are designed for senile hearing loss. Simulation and experimentation results prove the validity of this method.

A parameter estimation method for multilevel Gibbs random fields

In this paper, we present a parameter estimation algorithm for multilevel Gibbs-Markov random fields (GMRF). It effectively improves the linear estimation method proposed by Derin and Eillott, especially for small size images.