P.L. De Leon

Speaker Model Clustering for Efficient Speaker Identification in Large Population Applications

In large population speaker identification (SI) systems, likelihood computations between an unknown speakers feature vectors and the registered speaker models can be very time-consuming and impose a bottleneck. For applications requiring fast SI, this is a recognized problem and improvements in efficiency would be beneficial. In this paper, we propose a method whereby GMM-based speaker models are clustered using a simple k-means algorithm. Then, during the test stage, only a small proportion of speaker models in selected clusters are used in the likelihood computations resulting in a significant speed-up with little to no loss in accuracy. In general, as the number of selected clusters is reduced, the identification accuracy decreases; however, this loss can be controlled through proper tradeoff. The proposed method may also be combined with other test stage speed-up techniques resulting in even greater speed-up gains without additional sacrifices in accuracy.

Investigating the Option of Removing the Antialiasing Filter From Digital Relays

Digital relays traditionally employ sampling rates of less than 100 samples/cycle. In order to avoid aliasing due to fault transients, these relays employ an analog antialiasing filter before critical-sampling (Nyquist rate) the input waveforms coming from instrument transformers. In many applications of electrical engineering, oversampling (greater than the Nyquist rate) has long been used to simplify the requirements of an antialiasing filter with a sharp cutoff; in some cases, the filter can even be eliminated. This paper investigates this option for a digital relay. The performance of a traditional digital relay is compared with a method that uses oversampling without using an antialiasing filter. By processing a comprehensive array of fault waveforms from Electromagnetic Transients Program simulations, a suitable oversampling rate is suggested. A comparison of phasor estimates using the traditional relay and the proposed method is made for different operating and fault conditions. The results suggest that oversampling can eliminate the antialiasing filter traditionally employed in digital relays.

On the inversion of Mel-frequency cepstral coefficients for speech enhancement applications

The use of Mel-frequency cepstral coefficients (MFCCs) is well established in the fields of speech processing, particularly for speaker modeling within a Gaussian mixture model (GMM) speaker recognition system. The use of GMMs for speech enhancement applications has only recently been proposed in the literature; the concept of direct inversion of the MFCCs, however, has not been studied. In this paper we present a means to invert MFCCs for use in speech enhancement applications. Results for cepstral inversion is evaluated on the TIMIT speech corpus using perceptual evaluation of speech quality (PESQ).

Low Bit-Rate Speech Coding Through Quantization of Mel-Frequency Cepstral Coefficients

In this paper, we propose a low bit-rate speech codec based on vector quantization (VQ) of the mel-frequency cepstral coefficients (MFCCs). We begin by showing that if a high-resolution mel-frequency cepstrum (MFC) is computed, good-quality speech reconstruction is possible from the MFCCs despite the lack of phase information. By evaluating the contribution toward speech quality that individual MFCCs make and applying appropriate quantization, our results show that the MFCC-based codec exceeds the state-of-the-art MELPe codec across the entire range of 600-2400 bps, when evaluated with the perceptual evaluation of speech quality (PESQ) (ITU-T recommendation P.862). The main advantage of the proposed codec is in distributed speech recognition (DSR) since the MFCCs can be directly applied thus eliminating additional decode and feature extract stages; furthermore, the proposed codec better preserves the fidelity of MFCCs and better word accuracy rates as compared to CELP and MELPe codecs.

Reducing Speaker Model Search Space in Speaker Identification

For large population speaker identification (SID) systems, likelihood computations between an unknown speakers test feature set and speaker models can be very time-consuming and detrimental to applications where fast SID is required. In this paper, we propose a method whereby speaker models are clustered during the training stage. Then during the testing stage, only those clusters which are likely to contain high-likelihood speaker models are searched. The proposed method reduces the speaker model space which directly results in faster SID. Although there maybe a slight loss in identification accuracy depending on the number of clusters searched, this loss can be controlled by trading off speed and accuracy.

Link cache extensions for predictive routing and repair in ad hoc wireless networks

In ad hoc networks that employ source-based on-demand routing protocols, network nodes can use information contained in packet headers to populate a link cache. Given sufficient route requests and network traffic, a link cache can provide a good view of the past spatial network topology. This view can be successfully applied to making future routing decisions, provided that links are allowed to expire through a timeout mechanism, so that the cache stays fresh. We further assume that each node has the ability to determine its location, velocity, and bearing. We propose a link timeout mechanism that is based on mobility prediction, and introduce link cache extensions that incorporate temporal information regarding the future state of the network topology. By utilizing spatial and temporal information, better and more efficient routing decisions can be made. Furthermore, local repairs can be initiated preemptively in anticipation of a link break.

Parallel digital architectures for high-speed adaptive DSSS receivers

DSP-based implementations of receivers have many advantages over their analog counterparts including precise matched filtering and reconfigurability. As processing rates increase, more receiver functions are implemented digitally-the ultimate goal in this shift being all-digital receivers which sample at IF or RF. Practical limitations obviously occur when the processing rates fall behind sampling and symbol rates. We extend previous ideas for introducing parallelism into the receiver design. We describe a parallel, adaptive DSSS receiver in which individual processing units can potentially operate at rates below the symbol rate. The design is shown to have BERs equivalent to conventional designs.

Performance evaluation of the IEEE 802.11a and b WLAN physical layer on the Martian surface

The performance of IEEE 802.11a and b WLAN standards on the Martian surface is studied. The Gusev crater region and the Meridiani Planum (Hematite) region are chosen as example sites based on the mission science and mission success criteria. The radio frequency (RF) multipath environment is obtained using digital elevation maps (DEMs) from the Mars global surveyor mission, taking into account the atmosphere and other factors on the Martian surface. It is observed that IEEE 802.11a performs well in terms of packet error rates at distances up to a few hundred meters from the transmit antenna when the transmit power is 1 W and the antennas are located 1.5 in above the ground. Although the performance of IEEE 802.11b is found to be more adversely affected, its performance too can be improved significantly using a RAKE receiver. It is observed that the lower data rate modes of 802.11a show much better results in terms of bit error rates. However, both 802.11a and b appear to provide effective communications within a few hundred meters of the transmitter in the selected sites considered

Modeling the radio frequency environment of Mars for future wireless, networked rovers and sensor Webs

NASAs long-term goals for the exploration of Mars include the use of rovers and sensors, which intercommunicate through proximity wireless networks. The performance of any such wireless network depends fundamentally on the radio frequency (RF) environment. This paper presents our initial results concerning modeling the RF environment on Mars in support of determining the characteristics of potential in-situ networks. In this work, we have utilized commercial RF propagation modeling software, designed for cellular telephone system planning, together with recent topographic data for Mars to deterministically construct propagation path loss models. The software takes into account antenna heights and types, radiated power, topographic information, image data, and surface features and clutter. As we show, the resulting models provide a powerful tool for determining potential coverage patterns and data rates for mission planners.

Short-time kurtosis of speech signals with application to co-channel speech separation

Recent work into the separation of mixtures of speech signals has shown some success. One particular method is based on the assumption that scalar mixtures of speech signals have a kurtosis less than that for either source. Under this assumption, a simple gradient search algorithm is employed to maximize kurtosis thereby separating the source speech signals from the mixture. While this assumption has been observed to be generally true for long speech segments, it is quite reasonable to expect the assumption not to hold over short segments (windows) of speech. In this case, kurtosis maximization is not the appropriate strategy and the algorithm will fail to separate the signals. We examine the kurtosis of speech signals over short segments of speech, i.e., short-time kurtosis. The analysis indicates in general, how successful a kurtosis maximization strategy can be in separating speech signals from a mixture.Recent work into the separation of mixtures of speech signals has shown some success. One particular method is based on the assumption that scalar mixtures of speech signals have a kurtosis less than that for either source. Under this assumption, a simple gradient search algorithm is employed to maximize kurtosis thereby separating the source speech signals from the mixture. While this assumption has been observed to be generally true for long speech segments, it is quite reasonable to expect the assumption not to hold over short segments (windows) of speech. In this case, kurtosis maximization is not the appropriate strategy and the algorithm will fail to separate the signals. We examine the kurtosis of speech signals over short segments of speech, i.e., short-time kurtosis. The analysis indicates in general, how successful a kurtosis maximization strategy can be in separating speech signals from a mixture.