Discriminative kernel-based phoneme sequence recognition

Pattern Recognition, 2007

This paper present a principled SVM based speaker verification system. We propose a new framework and a new sequence kernel that can make use of any Mercer kernel at the frame level. An extension of the sequence kernel based on the Max operator is also proposed. The new system is compared to state-of-the-art GMM and other SVM based systems found in the literature on the Banca and Polyvar databases. The new system outperforms, most of the time, the other systems, statistically significantly. Finally, the new proposed framework clarifies previous SVM based systems and suggests interesting future research directions.

2008

Speaker independent feature extraction is a critical problem in speech recognition. Oriented principal component analysis (OPCA) is a potential solution that can find a subspace robust against noise of the data set. The objective of this paper is to find a speaker-independent subspace by generalizing OPCA in two steps: First, we find a nonlinear subspace with the help of a kernel trick, which we refer to as kernel OPCA. Second, we generalize OPCA to problems with more than two phonemes, which leads to oriented discriminant analysis (ODA). In addition, we equip ODA with the kernel trick again, which we refer to as kernel ODA. The models are tested on the CMU ARCTIC speech database. Our results indicate that our proposed kernel methods can outperform linear OPCA and linear ODA at finding a speaker-independent phoneme space.

2005

In this paper we use kernel-based Fisher Discriminants (KFD) for classification by integrating this method in a HMM-based speech recognition system. We translate the outputs of the KFD-classifier into conditional probabilities and use them as production probabilities of a HMM-based decoder for speech recognition. To obtain a good performance also in terms of computational complexity the Recursive Least Squares Algorithm (RLS-Algorithm) is enforced. We train and test the described hybrid structure on the Resource Management Corpus (RM1).

2005

We propose a new paradigm for aligning a phoneme sequence of a speech utterance with its acoustical signal counterpart. In contrast to common HMM-based approaches, our method employs a discriminative learning procedure in which the learning phase is tightly coupled with the alignment task at hand. The alignment function we devise is based on mapping the input acousticsymbolic representations of the speech utterance along with the target alignment into an abstract vector space. We suggest a specific mapping into the abstract vector-space which utilizes standard speech features (e.g. spectral distances) as well as confidence outputs of a framewise phoneme classifier. Building on techniques used for large margin methods for predicting whole sequences, our alignment function distills to a classifier in the abstract vector-space which separates correct alignments from incorrect ones. We describe a simple iterative algorithm for learning the alignment function and discuss its formal properties. Experiments with the TIMIT corpus show that our method outperforms the current state-of-the-art approaches.

Lecture Notes in Computer Science, 2004

We investigate techniques for acoustic modeling in automatic recognition of context-independent phoneme strings from the TIMIT database. The baseline phoneme recognizer is based on TempoRAl Pat- terns (TRAP). This recognizer is simplifled to shorten processing times and reduce computational requirements. More states per phoneme and bi-gram language models are incorporated into the system and evaluated. The question of insu-cient amount

2009

The robustness of phoneme classification to white Gaussian noise and pink noise in the acoustic waveform domain is investigated using support vector machines. We focus on the problem of designing kernels which are tuned to the physical properties of speech. For comparison, results are reported for the PLP representation of speech using standard kernels. We show that major improvements can be achieved by incorporating the properties of speech into kernels. Furthermore, the high-dimensional acoustic waveforms exhibit more robust behavior to additive noise. Finally, we investigate a combination of the PLP and acoustic waveform representations which attains better classification than either of the individual representations over a range of noise levels.

2003

String kernels have been introduced recently in an attempt to apply support vector machine (svm) classifiers to variable-length sequential data from a discrete alphabet. They have been used in the areas of text classification and bioinformatics, where notable results have been obtained. In the present paper string kernels are applied to a Spanish digit recognition task and their performance is compared to that of discrete hidden markov models (dhmm). It is found that string kernels produce comparable results and may offer an alternative discriminative approach for certain speech recognition tasks.

Linear Discriminative Analysis techniques have been used in pattern recognition to map feature vectors to achieve op- timal classification. Kernel Discriminative Analysis(KDA) seeks to introduce non-linearity in this approach by map- ping the features to a non-linear space before applying LDA analysis. The formulation is expressed as an eigen- value problem resolution. Using a different kernel, one can cover a wide class of nonlinearities. In this paper, we de- scribe this technique and present an application to a speech recognition problem. We give classification results for a connected digit recognition task and analyze some existing problems.

1997

In this paper, we present results for the Minimum Classi cation Error MCE 1 framework for discriminative training applied to tasks in continuous phoneme recognition. The results obtained using MCE are compared with results for Maximum Likelihood Estimation MLE. We examine the ability of MCE to attain high recognition performance with a small number of parameters. Phoneme-level and string-level MCE loss functions were used as the optimization criteria for a Prototype-Based Minimum Error Classi er PBMEC 2 and an HMM 3 . The former was optimized using Generalized Probabilistic Descent, the latter was optimized using an approximated second order method, the Quickprop algorithm. Two databases were used in this evaluation: 1 the ATR 5240 isolated word datasets for 6 speakers, in both speaker-dependent and multi-speaker mode; 2 the TIMIT database. For both databases, MCE training yielded striking gains in performance and classi er compactness compared to MLE baselines. For instance, through MCE training, performance similar to that of the Maximum Likelihood Successive State Splitting algorithm ML-SSS 4 could be obtained with 20 times fewer parameters.

IEEE Transactions on Audio, Speech, and Language Processing, 2000

This work proposes methods for combining cepstral and acoustic waveform representations for a front-end of support vector machine (SVM) based speech recognition systems that are robust to additive noise. The key issue of kernel design and noise adaptation for the acoustic waveform representation is addressed first. Cepstral and acoustic waveform representations are then compared on a phoneme classification task. Experiments show that the cepstral features achieve very good performance in low noise conditions, but suffer severe performance degradation already at moderate noise levels. Classification in the acoustic waveform domain, on the other hand, is less accurate in low noise but exhibits a more robust behavior in high noise conditions. A combination of the cepstral and acoustic waveform representations achieves better classification performance than either of the individual representations over the entire range of noise levels tested, down to −18dB SNR.