Whisper intelligibility enhancement based on noise robust feature and SVM

A machine learning based speech enhancement method is proposed to improve the intelligibility of whispered speech. A binary mask estimated by a two-class support vector machine （SVM） classifier is used to synthesize the enhanced whisper. A novel noise robust feature called Gammatone feature cosine coefficients （GFCCs） extracted by an auditory periphery model is derived and used for the binary mask estimation. The intelligibility performance of the proposed method is evaluated and compared with the traditional speech enhancement methods. Objective and subjective evaluation results indicate that the proposed method can effectively improve the intelligibility of whispered speech which is contaminated by noise. Compared with the power subtract algorithm and the log-MMSE algorithm, both of which do not improve the intelligibility in lower signal-to-noise ratio （SNR） environments, the proposed method has good performance in improving the intelligibility of noisy whisper. Additionally, the intelligibility of the enhanced whispered speech using the proposed method also outperforms that of the corresponding unprocessed noisy whispered speech.

IMPROVING VOICE ACTIVITY DETECTION VIA WEIGHTING LIKELIHOOD AND DIMENSION REDUCTION

The performance of the traditional Voice Activity Detection （VAD） algorithms declines sharply in lower Signal-to-Noise Ratio （SNR） environments. In this paper, a feature weighting likelihood method is proposed for noise-robust VAD. The contribution of dynamic features to likelihood score can be increased via the method, which improves consequently the noise robustness of VAD. Divergence based dimension reduction method is proposed for saving computation, which reduces these feature dimensions with smaller divergence value at the cost of degrading the performance a little. Experimental results on Aurora Ⅱ database show that the detection performance in noise environments can remarkably be improved by the proposed method when the model trained in clean data is used to detect speech endpoints. Using weighting likelihood on the dimension-reduced features obtains comparable, even better, performance compared to original full-dimensional feature.

Intelligent Seismic-Acoustic System for Identifying the Location of the Areas of an Expected Earthquake

A brief overview of the state-of-the-art in the field of earthquake study and forecasting is presented in this paper. We analyze the principles of the methods of determining the coordinates of earthquake focuses by means of ground seismic stations. We demonstrate that those methods cannot be used in the system for monitoring of the beginning of the earthquake preparation process (in the network of RNM ASP stations). As we know, the earthquake preparation process is accompanied by spreading noisy seismic-acoustic signals. Theoretically, the system for monitoring of the beginning of the earthquake preparation process is based on the technologies for seismic-acoustic signal processing-Robust Noise Monitoring (RNM). Noise characteristics determined by RNM technologies indicate the beginning of anomalous seismic processes (ASP) and, consequently, the possibility of ASP monitoring. Considering that the seismic-acoustic signal can be represented as the sum of the useful signal and noise, we present the technologies for determining noise characteristics. It is demonstrated in the paper that a change in the estimate of the cross-correlation function between the useful signal and the noise, noise variance and the value of noise correlation determine the beginning of ASP. One RNM ASP station determines the beginning of ASP within a radius of about 500 km. Determining the location of an expected earthquake requires a network of RNM ASP stations. We analyze the results of noise technology-based monitoring of anomalous seismic processes performed from July 2010 to June 2015 on nine seismic-acoustic stations built at the head of 10 m, 200 m, 300 m and 1400 - 5000 m deep wells. Based on the results of the experimental data obtained in the period covering over three years, an intelligent system has been built, which allows for identifying the location of the zone of an earthquake, using the combinations of time of change in the estimate of the correlation function between the useful signal and the noise of the seismic-acoustic information received from different stations 10 - 20 hours before the earthquake. In the long term, the system can be used by seismologists as a tool for determining the location of the zone of an expected earthquake.

Experimentally realizing efficient quantum control with reinforcement learning

We experimentally investigate deep reinforcement learning(DRL)as an artificial intelligence approach to control a quantum system.We verify that DRL explores fast and robust digital quantum controls with operation time analytically hinted by shortcuts to adiabaticity.In particular,the protocol’s robustness against both over-rotations and off-resonance errors can still be achieved simultaneously without any priori input.For the thorough comparison,we choose the task as single-qubit flipping,in which various analytical methods are well-developed as the benchmark,ensuring their feasibility in the quantum system as well.Consequently,a gate operation is demonstrated on a trapped^(171) Yb^(+)ion,significantly outperforming analytical pulses in the gate time and energy cost with hybrid robustness,as well as the fidelity after repetitive operations under time-varying stochastic errors.Our experiments reveal a framework of computer-inspired quantum control,which can be extended to other complicated tasks without loss of generality.

Experimental realization of nonadiabatic holonomic single‐qubit quantum gates with two dark paths in a trapped ion

For circuit-based quantum computation,experimental implementation of a universal set of quantum logic gates with high-fidelity and strong robustness is essential and central.Quantum gates induced by geometric phases,which depend only on global properties of the evolution paths,have built-in noise-resilience features.Here,we propose and experimentally demonstrate nonadiabatic holonomic single-qubit quantum gates on two dark paths in a trapped ^(171)γδ^(+)ion based on four-level systems with resonant drives.We confirm the implementation with measured gate fidelity through both quantum process tomography and randomized benchmarking methods.Meanwhile,we find that nontrivial holonomic two-qubit quantum gates can also be realized within current experimental technologies.Compared with previous implementations,our experiments share both the advantages of fast nonadiabatic evolution and robustness against systematic errors.Therefore,our experiments confirm a promising method for fast and robust holonomic quantum computation.

Robust Object Tracking via Information Theoretic Measures

Object tracking is a very important topic in the field of computer vision.Many sophisticated appearance models have been proposed.Among them,the trackers based on holistic appearance information provide a compact notion of the tracked object and thus are robust to appearance variations under a small amount of noise.However,in practice,the tracked objects are often corrupted by complex noises(e.g.,partial occlusions,illumination variations)so that the original appearance-based trackers become less effective.This paper presents a correntropy-based robust holistic tracking algorithm to deal with various noises.Then,a half-quadratic algorithm is carefully employed to minimize the correntropy-based objective function.Based on the proposed information theoretic algorithm,we design a simple and effective template update scheme for object tracking.Experimental results on publicly available videos demonstrate that the proposed tracker outperforms other popular tracking algorithms.