Pengfei Sun

Low-Rank and Sparsity Analysis Applied to Speech Enhancement Via Online Estimated Dictionary

In this letter, we propose an online estimated local dictionary based single-channel speech enhancement algorithm, which focuses on low-rank and sparse matrix decomposition. In the proposed algorithm, a noisy speech spectrogram can be decomposed into low-rank background noise components and an activation of the online speech dictionary, on which both low-rank and sparsity constraints are imposed. This decomposition takes the advantage of local estimated exemplars high expressiveness on speech components and also accommodates nonstationary background noise. The local dictionary can be obtained through estimating the speech presence probability (SPP) by applying expectation-maximal algorithm, in which a generalized Gamma prior for speech magnitude spectrum is used. The proposed algorithm is evaluated using signal-to-distortion ratio, and perceptual evaluation of speech quality. The results show that the proposed algorithm achieves significant improvements at various SNRs when compared to four other speech enhancement algorithms, including improved Karhunen-Loeve transform approach, SPP-based MMSE, nonnegative matrix factorization-based robust principal component analysis (RPCA), and RPCA.

Fatigue Modeling via Mammalian Auditory System for Prediction of Noise Induced Hearing Loss

Noise induced hearing loss (NIHL) remains as a severe health problem worldwide. Existing noise metrics and modeling for evaluation of NIHL are limited on prediction of gradually developing NIHL (GDHL) caused by high-level occupational noise. In this study, we proposed two auditory fatigue based models, including equal velocity level (EVL) and complex velocity level (CVL), which combine the high-cycle fatigue theory with the mammalian auditory model, to predict GDHL. The mammalian auditory model is introduced by combining the transfer function of the external-middle ear and the triple-path nonlinear (TRNL) filter to obtain velocities of basilar membrane (BM) in cochlea. The high-cycle fatigue theory is based on the assumption that GDHL can be considered as a process of long-cycle mechanical fatigue failure of organ of Corti. Furthermore, a series of chinchilla experimental data are used to validate the effectiveness of the proposed fatigue models. The regression analysis results show that both proposed fatigue models have high corrections with four hearing loss indices. It indicates that the proposed models can accurately predict hearing loss in chinchilla. Results suggest that the CVL model is more accurate compared to the EVL model on prediction of the auditory risk of exposure to hazardous occupational noise.

Measurement of field complex noise using a novel acoustic detection system

This paper represents our recent experimental measurement study of the complex noise in industrial fields, using a novel acoustic detection system and wavelet transform algorithms. Noise induced hearing loss (NIHL) continues to be one of the most prevalent occupational hazards in the United States. Number of research on NIHL showed a complex noise could produce more hearing loss than an energy-equivalent continuous or impulsive noise alone. Many workplaces in varied industries are subjected to the high level complex noise (i.e., high-level impulsive noise mixed with continuous Gaussian noise). The current noise measurement guidelines and devices (e.g., conventional sound level meters) are based on the equal energy hypothesis (EEH), which states that loss of hearing by exposure to noise is proportional to the total acoustic energy of the exposure. However, the EEH does not accurately rate the impulsive noise and the complex noise. Therefore, the conventional sound level meter may not be able to accurately assess the complex noise in industrial fields. In this project, a new waveform profile based noise measurement system has been developed for evaluation of the high level complex noise in industrial fields. The system consists of four ½ condenser microphones, and it can simultaneously detect and record four waveforms of the complex noise with high sampling rate (125 KHz). In addition, a wavelet transform based signal analysis algorithm has been modified and implemented to characterize the complex noise. Pilot field measurements have been conducted in selected local coal mining fields (e.g., wet coal preparation plant and dry coal handling plant) using the developed system. The preliminary results showed that the system successfully detected and recorded waveforms of complex noise in industrial fields. The modified algorithm can decomposed the complex noise signals and display the detailed features in the time-frequency joint domain. The key parameters of complex noise can be determined, and the hazardous complex noise in industrial fields can be identified. In addition, when measuring the equivalent A-weighted averaged sound pressure level, the developed system is comparable to a conventional sound level meter.

A waveform profile-based noise measurement system and methodology for complex noise evaluation

Noise induced hearing loss remains as one of the most prevalent occupational related health problems in the United States. Many workplaces in various industries are subjected to the high-level complex noise exposure, which is a combination of impulsive noise and continuous Gaussian noise. The current noise measurement guidelines and devices (e.g., conventional sound level meters) are limited on accurate assessment of the complex noise. In this study, a new waveform profile based noise measurement system has been developed to detect high-level complex noise in industrial workplaces. Pilot field noise measurements have been conducted using the developed system in two selected coal mining facilities (i.e., a wet coal preparation plant and a dry coal handling plant). A wavelet transform based algorithm has been applied for characterization and analysis of the complex noise measured in the fields. The results show that the developed system can effectively measure and collect the waveforms of the high-level complex noise in coal mining facilities. The proposed new metrics, together with conventional noise metrics, can more extensively assess the auditory risk of the high-level complex noise in industrial fields.

Analysis of impulse noise based on wavelet transform for military applications

In this project, we present a data acquisition and analysis study of impulse noise based on wavelet transform for military applications. Impulse noise is a type of highly transient signal widely experienced in military fields (e.g., an intense blast wave). The wavelet transform has been used to analyze signals of impact noise and vibrations, and it showed superior advantages on analysis of transient signals compared to the fast Fourier transform and the short-time Fourier transform. This study focuses on analysis of A-wave type impulse noise in the T-F domain using the continual wavelet transforms. Three different wavelets (i.e., Morlet, Mexican hat, and Meyer wavelets) were investigated and compared based on theoretical analysis and applications to experimental generated impulse noise signals. The underlying theory of continual wavelet transform was given and the temporal and spectral resolutions of different wavelets were theoretically analyzed. The results on singularity detection of the impulse noise showed the Mexican hat wavelet could better reflect the signal oscillations. Furthermore, the similarity of signals between the impulse noise and wavelets functions was investigated in time and frequency domain. The results showed the waveform of Mexican hat wavelets is more similar to the impulse noise signal than the other two wavelets. In summary, although all of three wavelets can represent detailed features of impulse noise in the T-F domain, the Mexican hat wavelets show obvious advantages over the Morlet and Meyer wavelets. The results of this study provide a possible strategy to design special wavelets for impulse noise detection and analysis.