Using a bidirectional dc-dc converter along with low-voltage energy storage for the high-voltage dc bus and traction motor drives has been a prominent option for hybrid electric and fuel cell vehicles. This paper will describe the significance of energy management power converters and their circuit topology options for efficiency, size, and cost considerations. Whether isolated or nonisolated, soft switching techniques have been widely used in high-power bidirectional dc-dc converters. Through some design examples, the component selection and circuit design optimization are discussed, and their efficiency evaluation results are also given. Major difficulties of developing a high-power bidirectional dc-dc converter are found in lack of high-power passive components and lack of multiphase dc-dc controllers. More development work needs to be done in these areas

We propose a new method for adaptively removing noise and interference from a signal. In this method, isolation of the signal and interference components is facilitated by a concentration process applied to the short time Fourier transform (STFT). Unwanted components are removed from the concentrated surface, and a clean signal is estimated by integration. The concentrated STFT is a linear representation, free of cross terms and having the property that signal and interference components are easily recognized because their distributions are more concentrated in time and frequency. We demonstrate the advantages of the proposed method over conventional methods.

We develop new algorithms which address the problems of accurate frequency estimation and the comparison of the frequency content of two different spectra. We apply these algorithms to speech to estimate formants accurately from very little data and to estimate the frequency differences of individual formants of similar vowel instances. The first algorithm is a simple method for the construction of a very high-resolution, high-gain spectrum. This phase-reparameterized spectrum is based on a deterministic signal model and uses both Fourier phase and magnitude and accounts for all of the energy in the Fourier spectrum. To compare the formants of similar vowels, we introduce a spectral correlation algorithm based on an outer product rather than the inner product of conventional correlation.

We present an algorithm, based on the EM algorithm, which simultaneously estimates both physical scale and vowel identities from a segment of speech. The validity of the algorithm depends on the scale hypothesis that the variation of the formant frequencies for a given vowel is mainly influenced by the physical size of the speaker. This is both a new application and a new confirmation of a hypothesis that is often accepted without proof.

A correlation based on the mixed partial derivatives of the phase of the short time Fourier transform (STFT) is developed and applied to speech. This correlation function recovers the unambiguous pitch fundamental, removing most of the effects of the vocal tract resonance. It is therefore well suited for efficient and estimation of pitch and as a voicing detector.

We present experimental results that show that the scale-factor relating the formant frequencies of different speakers increases with decreasing values of formant frequency. Based on these results, we experimentally obtain a frequency warping function aimed at separating speaker dependencies from the inherent characterization of the sound. We find that the frequency warping function is similar to the Mel scale, and we believe that this is the first time that a Mel-like scale has been obtained using only speech. Our results and methods may therefore explain, from a speech point of view, the Mel scale, which was obtained historically from hearing based experiments.

At the beginning of the 21st century, fuel cells appear poised to meet the power needs of a variety of applications. Fuel cells are electrochemical devices that convert chemical energy to electricity and thermal energy. Fuel cell systems are available to meet the needs of applications ranging from portable electronics to utility power plants. In addition to the fuel cell stack itself, a fuel cell system includes a fuel processor and subsystems to manage air, water thermal energy, and power. The overall system is efficient at full and part-load, scaleable to a wide range of sizes, environmentally friendly, and potentially competitive with conventional technology in first cost. Promising applications for fuel cells include portable power, transportation, building cogeneration, and distributed power for utilities. For portable power a fuel cell coupled with a fuel container can offer a higher energy storage density and more convenience than conventional battery systems. In transportation applications, fuel cells offer higher efficiency and better part-load performance than conventional engines. In stationary power applications, low emissions permit fuel cells to be located in high power density areas where they can supplement the existing utility grid. Furthermore, fuel cell systems can be directly connected to a building to provide both power and heat with cogeneration efficiencies as high as 80%.

We consider the properties of the short time Fourier transform as a representation of the distribution of the analytic signal in time and frequency and use the properties of the surface to remap the surface and reconstruct the analytic representations of signal components. In the remapping, the phase gradient of the surface is used to redistribute the signal on the surface, resting in a more focused image of multicomponent FM signals. In the reconstruction of analytic signal components the STFT surface is summed, resulting in a direct inversion process, which does not involve the use of inverse transforms.

Computationally efficient speech extraction algorithms have significant potential economic benefit, by automating an extremely tedious manual process. Previously, algorithms which discriminate between speech and one specific other signal type have been developed, and often fail when the specific non-speech signal is replaced by a different signal type. Moreover, several such signal specific discriminators have been combined in order to tackle the general speech vs. non-speech discrimination problem, with predictable negative results. When the number of discriminating features is large, compression methods such as principal components have been applied to reduce dimension, even though information may be lost in the process. In this paper, graphical tools are applied to determine a set of features which produce excellent speech vs. nonspeech clustering. This cluster structure provides the basis for a general speech vs. non-speech discriminator, which significantly outperforms the TALKATIVE speech extraction algorithm.

We show that there are many qualitatively different equations, each with few parameters, that fit the experimentally obtained Mel scale. We investigate the often made remark that there are two regions to the Mel scale, the first region (

In this paper, we study the scale transform of the spectral-envelope of speech utterances by different speakers. This study is motivated by the hypothesis that the formant frequencies between different speakers are approximately related by a scaling constant for a given vowel. The scale transform has the fundamental property that the magnitude of the scale-transform of a function X(f) and its scaled version /spl radic//spl alpha/X(/spl alpha/f) are same. The methods presented here are useful in reducing variations in acoustic features. We show that the F-ratio tests indicate better separability of vowels by using scale-transform based features than mel-transform based features. The data used in the comparison of the different features consist of 200 utterances of four vowels that are extracted from the TIMIT database.

There are many situations in which it is desirable to be able to distinguish spontaneous speech and speech which is non-spontaneous. Examples of situations in which this problem may arise include forensic evidence situations, sorting voice-mail responses from voice-mail menus, and automatic segmentation of spontaneous responses from prepared questions. The latter situation can occur if it is desired to create a database of spontaneous data from data which consists of spontaneous discourse responding to prepared prompts. This paper outlines and compares three methods for automatically classifying spontaneous and non-spontaneous speech and presents the experimental results comparing the performance of the methods. All three methods are based on an analysis of the probability distributions of prosodic features extracted from the speech signal. The first method uses an expansion of the probability distribution in terms of the statistical moments. The second method is an application of a modified Hellinger's method applied to histograms of signal amplitude and other speech features. The third method is based on a measure of the non-Gaussianity of the data.

The spectral phase is a quantity which is normally discarded in analyzing signals. The concept of a complex time-frequency representation is presented in which the rows are narrow bandpass filters and the columns are broadband Fourier spectra. Methods are developed which exploit the spectral phase of the surface to recover the FM modulating function of an FM modulated tone and an FM modulated multi-component harmonic signal.

We present improvements over the original scale-cepstrum proposed by Umesh, Cohen, Marinovic and Nelson (see IEEE Trans. on Speech and Audio Processing, 1996). The scale-cepstrum was motivated by a desire to normalize the first-order effects of differences in vocal-tract lengths for a given vowel. Our subsequent work has shown that a more appropriate frequency-warping than the log-warping used in in the original scale-cepstrum is necessary to account for the frequency dependency of the scale-factor. Using this more appropriate frequency-warping and a modified method of computing the scale-cepstrum we have obtained improved features that provide better separability between vowels than before, and are also robust to noise.

A new method for generating speech spectrograms is presented. This algorithm is based on an autocorrelation function whose parameters are chosen provide processing gain and formant resolution, while minimizing pitch artifacts in the spectrum. Crisp formants are produced, and the power ratio of the formants can be adjusted by pre-filtering the data. The autocorrelation process is functionally equivalent to a time-smoothed, windowed Wigner distribution. The process is an improvement over the normal FFT implementation since it requires much less data to resolve the speech formants, and it is an improvement over the un-smoothed Wigner distribution since the cross-terms normally associated with the Wigner distribution are greatly attenuated by the smoothing operation.

We have proposed the use of scale-cepstral coefficients as features in speech recognition. We have developed a corresponding frequency-warping function, such that, in the warped domain the formant envelopes of different speakers are approximately translated versions of one and another for any given vowel. These methods were motivated by a desire to achieve speaker-normalization. In this paper, we point out very interesting parallels of the various steps in computing the scale-cepstrum, with those observed in computing features based on physiological models of the auditory system or psychoacoustic experiments. It may therefore be useful to have a better understanding of the need for the various signal-processing steps which may result in the development of more robust recognizers.

An automatic technique which attempts to determine and cluster topical information in text and transcribed speech is described. The method uses an electronic dictionary which is recursively closed. The method identifies each word of a given excerpt of text by its root form in the dictionary and a salience is affixed. The salience from each individual word is then propagated to those terms which define it, which terms may themselves be propagated. Thus, each word in the excerpt spawns a directed graph of dictionary words and related weights. A combining function is then applied to the union of these graphs which enhances any mutual information. The N words with the largest combined scores are identified as either the topic or words pertaining to the topic. These top N words and their scores can then be used in cross-message correlations as a means to cluster topics.

We propose a computationally efficient method for estimation of frequency of a single complex sinusoid at low SNR. This method is motivated by the cross-power spectrum method of Nelson (1993) and the weighted phase averager (WPA) methods of Tretter (1985), Kay (1988), and Lovell et al. (1991). We demonstrate that by a simple preprocessing, we can extend the threshold SNR of the WPA significantly. Further, unlike the WPA, the proposed method can be easily extended for estimation of frequencies of multiple sinusoids that are well-separated in frequency. We also derive the variance of the proposed estimator and provide simulation results comparing the proposed method with the WPA.

Most machine speech analysis and processing is based on a warped spectral representation. The intent of the paper is to present a method by which proper warped representations can be computed efficiently. In the case of log-warping functions, the methods of the paper produce a wavelet-like transform as a linear convolution of a single log-warped wavelet basis element and a log-warped representation of the signal. The resulting doubly warped transform is referred to in the paper as a Mellin-wavelet transform. The majority of the paper is devoted to deriving design parameters for implementation of the transform, with speech as the primary application.

This paper describes a new algorithm, the NP algorithm, for detecting speech signals of varying quality and gain levels. NP operates in the frequency domain and renders speech/no-speech decisions based on a signal-to-noise ratio (SNR) derived from a sorted power spectrum. In addition to the SNR estimates, a spectral whitening process and an estimate of the variance in the ratio of the signal power to total energy are also used to identify and reject signals that are stationary or nearly stationary The key features of this algorithm are: the detection is based on a single FFT; decisions are independent of the signal gain; the process has 3 dB/octave processing gain from the transform; and frequency domain processing permits the exploitation of the signal structure.