Zoran Saric

Hands-free voice communication with TV

This paper presents a system for full-duplex hands free voice communication integrated with TV technology. The system provides comfortable conversation by utilization of microphone array and advanced voice processing algorithms, even with simultaneous TV usage. Signal processing includes superdirective beamformer steered by direction-finding module, postprocessing module, acoustic echo canceller, stationary noise reduction module and automatic gain control. All processing is realized in real-time on DSP based platform. As communication channel GSM or VoIP can be used.

Acoustic Source Localization in Wireless Sensor Network

In this paper, we consider the problem of acoustic source localization in a wireless sensor network based on different measured signal quantities, such as the received signal strength (RSS), the angle of arrival (AOA) and the time of arrival (TOA). For each of these quantities, an appropriate weighted least squares criterion function is developed to be used for sound source localization. The weights of each criterion function take into account the decrease in the signal-to-noise ratio (SNR) with distance from the source. In addition, RSS localization algorithm proposed in this paper provides improvement of the localization accuracy for low SNR. Finally, separate criterion functions for RSS, TOA and AOA are used together to obtain minimal localization error and maximal reliability of the acoustic source localization. Simulation analysis confirms improved performance of the proposed localization algorithm.

Adaptive microphone array for unknown desired speaker’s transfer function

The main drawback of minimum variance distortionless response (MVDR) beamformer is the cancellation of the desired speech signal and its degradation in multi-path wave propagation environment. To make the adaptive algorithm robust against room reverberation and to prevent desired signal cancellation an estimation of unknown desired speakers transfer function was proposed. The estimation is based on the signal and the interference covariance matrices. The estimated transfer function is then applied to the MVDR beamformer. The proposed algorithm was tested on a simulated room with reverberation. The results showed better quality of the restored speech compared to some typical adaptive algorithms.

Augmenting a small parallel text with morpho-syntactic language resources for Serbian-English statistical machine translation

In this work, we examine the quality of several statistical machine translation systems constructed on a small amount of parallel Serbian-English text. The main bilingual parallel corpus consists of about 3k sentences and 20k running words from an unrestricted domain. The translation systems are built on the full corpus as well as on a reduced corpus containing only 200 parallel sentences. A small set of about 350 short phrases from the web is used as additional bilingual knowledge. In addition, we investigate the use of monolingual morpho-syntactic knowledge i.e. base forms and POS tags.

Adaptive microphone array based on pause detection

In this paper the performance of generalized sidelobe canceller (GSC) in room with reverberation is analyzed. GSC is inefficient in interference suppression when there is any correlation between interference and desired signal. The reason for this is unwanted cancellation of desired signal. It was shown that cancellation of the desired signal is proportional to the correlation between the direct wave of desired signal and its reflections. There is no cancellation if the GSC weightings are estimated in pauses of desired signal. Using these facts, a new algorithm based on pause detection was proposed. Simulations prove the advantage of proposed algorithm.

Application of the maximum signal to interference ratio criterion to the adaptive microphone array

The minimum variance (MV) criterion is widely used for the weight vector estimation of the adaptive microphone array. The drawback of this criterion is the cancellation of the desired speech signal and its degradation in multipath wave propagation environment. Applying the maximum signal to interference ratio (MSIR) instead of the MV criterion has two benefits. The first one is the high suppression of the interferences and the second one the desired speech enhancement. The proposed MSIR criterion is applied to the new generalized eigenvalue based beamformer (GEVBF). Its superiority is experimentally proved by simulating a room with reverberation.

A New Post-filter Algorithm Combined with Two-step Adaptive Beamformer

The optimal microphone array, in the sense of minimum mean square errors (MMSE), includes two processing blocks: the minimum variance distortionless response (MVDR) beamformer and the single-channel Wiener filter, which acts as post-filter. In this paper, we propose a new post-filter algorithm based on assumptions that both the noise power attenuation factor (NPAF) and signal power attenuation factor (SPAF) are time invariant in the reverberant room. The algorithm recursively estimates both factors from available measurements and uses them in estimation of the post-filter parameters. Additionally, to overcome the problem of the poor performance of the MVDR beamformer in reverberant conditions, we propose the usage of the two-step (TS) MVDR algorithm. This algorithm improves the robustness of the beamformer and its ability to suppress the interferences using an estimate of the desired speaker transfer function. Although TS MVDR beamformer and proposed post-filter can work separately, or combined with other algorithms, the best performance is obtained when they work together. The performance of the proposed combination of new post-filter algorithm and TS MVDR beamformer is tested in a simulated reverberant room and compared with similar algorithms, which gave rather good results.

Hands-free voice communication platform integrated with TV

This paper presents a system for full-duplex hands free voice communication integrated with TV technology. The system provides comfort conversation by utilization of microphone array and advanced voice processing algorithms, even with simultaneous TV usage. As communication channel GSM or VoIP can be used.

Time‐frequency detection of stridence in fricatives and affricates

As wheezes in abnormal breath sounds observed in patients with obstructive pulmonary diseases, the stridence in voice is manifested as excessively sharp, conspicuous, usually habitual hiss that is especially distinct with whispering. This paper reviews the articulator and acoustics features of stridence in unvoiced fricatives and affricates, and presents an algorithm for detection of stridence. Detection of stridence was based on: time‐frequency representation by FFT power spectra, time‐frequency representation by AR‐Burg power spectra, and power trajectory of signals in characteristic frequency bands. Many features are extracted from this analysis, as: the local power spectra maximum to average surrounding power ratio, the correlation coefficient between spectral and power maxima in signal, the spectral power slope in selected frequency band, spectral entropy in selected frequency band, and phoneme duration. The extracted set of features is input to the nonlinear classificator that decides about stridenc...

Sequential speech segmentation based on the spectral ARMA transition measure

Sequential segmentation algorithms based on the AR model tend to produce false alarms or to omit the change for sequences that corresponds to the ARMA model. In this paper a new sequential segmentation algorithm based on the ARMA model is presented. The ARMA model is estimated over the relatively short sequence, which has called for the implementation of the estimation algorithm with appropriately initialized starting values. The proposed algorithm adopts the MGLR concept of the sliding reference and test windows, which allow the process of decision making to be separated from the evaluation of the discrimination function. This has enabled the new triangular decision rule to be proposed; this is based on the expected shape of the discrimination function at the time of the model change. Two possible discrimination functions have been suggested. One of them is optimal in the statistical sense; the other has the better asymptotic behavior. Natural speech signal segmentation is also discussed, and an appropriate pitch-synchronous signal prearrangement has been suggested. This not only enhances the segmentation algorithm but also increases its speed, as the time can be increased by a step equal to the pitch period. The segmentation algorithm is verified on test signals as well as on the natural speech signal. The experimental results also include a comparison of the sequential AR and ARMA model-based segmentation.