Showing papers in "Digital Signal Processing in 2000"

TL;DR: The major elements of MIT Lincoln Laboratory's Gaussian mixture model (GMM)-based speaker verification system used successfully in several NIST Speaker Recognition Evaluations (SREs) are described.

TL;DR: The test normalization method is extended to use knowledge of the handset type, and the world, cohort, and zero normalization techniques are explained.

TL;DR: This article summarizes the 1999 NIST Speaker Recognition Evaluation, which discussed the overall research objectives, the three task definitions, the development and evaluation data sets, the specified performance measures and their manner of presentation, the overall quality of the results.

TL;DR: Logistic regression is introduced as a particular case of the naive Bayesian classifier and is applied to the fusion of all submitted data of the 1-speaker recognition task, part of the NIST'99 campaign.

TL;DR: This paper discusses the application of STFD to high-resolution direction finding and focuses on both the role and the effect of crossterms in angle estimation when multiple time-frequency points are incorporated.

TL;DR: The speaker verification performance of human listeners was compared to that of computer algorithms/systems and human performance in general seemed relatively robust to degradation.

TL;DR: Two approaches to detecting and tracking speakers in multispeaker audio using an adapted Gaussian mixture model, universal background model (GMM-UBM) speaker detection system as the core speaker recognition engine and an external segmentational algorithm based on blind clustering are described.

TL;DR: The performance of a recently proposed model-based space–time adaptive processing detection method is considered here and compared with several candidate algorithms.

TL;DR: A novel method for designing filters that are capable of normalizing the variability introduced by different telephone handsets is introduced and the effectiveness of the proposed channel normalizing filter in improving speaker verification performance in mismatched conditions is demonstrated.

TL;DR: A new approach for the classification of SAR targets is presented here, which combines maximally decimated directional filter banks with higher-order neural networks (HONNs) and is effective in enhancing the discrimination power of the HONN inputs, leading to significantly improved performance.

TL;DR: This paper provides an overview of approaches and issues in modeling the performance of automatic target recognition systems and the difficulty of characterizing and representing in a computational model the inherent variability of target signatures and the backgrounds in which they are embedded.

TL;DR: This work presents a multirecognizer architecture designed to cope with this issue in the framework of Automatic Speaker Recognition, based on various individual recognizers that exploits different classes of information conveyed by the speech signal.

TL;DR: SRI's speaker tracking and detection system is designed for tracking switchboard conversations and uses a two-speaker and silence hidden Markov model (HMM) with a minimum state duration constraint and Gaussian mixture model (GMM) state distributions adapted from a single gender- and handset-independent imposter model distribution.

TL;DR: “Cheating” tests using NIST-supplied keys lead to some improvements in channel normalization, and the roles that speaker segmentation and segment selection play in these tasks are illuminated.

TL;DR: This paper focuses on two important locality issues in detecting or tracking speakers in a telephone conversation, for which the speaker change frequency is usually high and channel estimation needs sufficiently long but homogeneous segments.

TL;DR: Two different speaker modeling approaches, based on multilayer perceptrons (MLPs) and on Gaussian mixture models (GMMs) are studied, and the MLP-based segmental systems have performance comparable to that of the global MLPs, and in the mismatched train-test conditions slightly better results are obtained with the segmental MLP system.

TL;DR: A geometry optimization technique is presented that permits accurate DOA estimation of arbitrarily correlated sources and is an important factor in the GSS algorithm and subsequent array geometry design.

TL;DR: This paper reviews how prior knowledge can be included at different normalization levels and the benefit of including knowledge of the target speaker gender and, eventually, of the test segment handset type, with experiments on the Switchboard corpus.

TL;DR: The main features of these systems are the implementation of a Maximum A Posteriori approach for speaker model estimation, an utterance-length dependent z-normalization scheme for test segment scoring, a speaker/world mixture model for addressing the two-speaker detection task, and the use of a change point detection method for speaker tracking.

TL;DR: It is demonstrated that discrimination of natural background and man-made objects using low resolution synthetic aperture radar imagery is possible using multiscale autoregressive (MAR), multiscales autore progressive moving average (MARMA) models, and singular value decomposition (SVD) methods.

TL;DR: This article presents the RIMO/ELISA speaker verification system which has been used in the 1999 NIST speaker recognition evaluation using a new technique for analyzing speech signals called time?frequency principal component(TFPC) analysis.