We study a discrete contextual stochastic (CS) model for complex and variant patterns like handwritten Chinese characters. Three fundamental problems of using CS models for character recognition are discussed, and several practical techniques for solving these problems are investigated. A formulation for discriminative training of CS model parameters is also introduced and its practical usage investigated. To illustrate the characteristics of the various algorithms, comparative experiments are performed on a recognition task with a vocabulary consisting of 50 pairs of highly similar handwritten Chinese characters. The experimental results confirm the effectiveness of the discriminative training for improving recognition performance.

Two statistical language models have been investigated on their effectiveness in upgrading the accuracy of a Chinese character recognizer. The baseline model is one of lexical analytic nature which segments a sequence of character images according to the maximum matching of words with consideration of word binding forces. A model of bigram statistics of word-classes is then investigated and compared against the baseline model in terms of recognition rate improvement on the image recognizer. On the average, the baseline language model improves the recognition rate by about 7% while the bigram statistics model upgrades it by about 10%

The theory of contextual stochastic modeling of structurally complex and variant patterns like handwritten Chinese characters is presented. The key lies in capturing the contextual information of the strokes in the character image which is segmented into regions systematically. A model for a character captures the region-region and region-pixel stochastic relationship. An unknown image is recognized by computing the pseudo likelihood of identifying its pixels to regions of each model. Based on this theory, an off-line, large vocabulary, hand-written character recognizer has been constructed.

A handwritten Chinese character off-line recognizer based on contextual vector quantization (CVQ) of every pixel of an unknown character image has been constructed. Each template character is represented by a codebook. When an unknown image is matched against a template character, each pixel of the image is quantized according to the associated codebook by considering not just the feature vector observed at each pixel, but those observed at its neighbors and their quantization as well. Structural information such as stroke counts observed at each pixel are captured to form a cellular feature vector. Supporting a vocabulary of 4616 simplified Chinese characters and alphanumeric and punctuation symbols, the writer-independent recognizer has an average recognition rate of 77.2 percent. Three statistical language models for postprocessing have been studied for their effectiveness in upgrading the recognition rate of the system. Among them, the CVQ-based language model is the most effective one upgrading the recognition rate by 10.4 percent on the average.

An off-line hand-written Chinese character recognizer based on contextual vector quantization (CVQ) supporting a vocabulary of 4616 Chinese characters, alphanumerics and punctuation symbols has been reported. Trained with a sample for each character from each of 100 writers and tested on texts of 160000 characters written by another 200 writers, the average recognition rate is 77.2%. Two statistical language models have been investigated in this study. Their performance in terms of their capabilities in upgrading the recognition rate by 8.8% and 12.0% respectively when used as post-processors of the recognizer are reported.

A hand-printed Chinese character recognizer based on contextual vector quantization (CVQ) has been built previously. In this paper, several properties of the recognizer are discussed. The recognition of 4516 Chinese characters has a successful rate of 91.0%. The output of the recognizer is passed to a language model which when applied to recognize a passage of about 1200 characters raises the rate from 91.5% to 97.5%.

A novel ergodic multigram hidden Markov model (HMM) is introduced which models sentence production as a doubly stochastic process, in which word classes are first produced according to a first order Markov model, and then single or multi-character words are generated independently based on the word classes, without word boundary marked on the sentence. This model can be applied to languages without word boundary markers such as Chinese. With a lexicon containing syntactic classes for each word, its applications include language modeling for recognizers, and integrated word segmentation and class tagging. Pre-segmented and tagged corpus are not needed for training, and both segmentation and tagging are trained in one single model. In this paper, relevant algorithms for this model are presented, and experimental results on a Chinese news corpus are reported.

On-line quasi-Bayes adaptation of the mixture coefficients and mean vectors in semicontinuous hidden Markov model (SCHMM) is studied. The viability of the proposed algorithm is confirmed and the related practical issues are addressed in a specific application of on-line speaker adaptation using a 26-word English alphabet vocabulary.

On-line adaptation of semi-continuous (or tied mixture) hidden Markov model (SCHMM) is studied. A theoretical formulation of the segmental quasi-Bayes learning of the mixture coefficients in SCHMM for speech recognition is presented. The practical issues related to the use of this algorithm for on-line speaker adaptation are addressed. A pragmatic on-line adaptation approach to combine the long-term adaptation of the mixture coefficients and the short-term adaptation of the mean vectors of the Gaussian mixture components are also proposed. The viability of these techniques are confirmed in a series of comparative experiments using a 26-word English alphabet vocabulary.

A hand-printed Chinese character recognizer based on contextual vector quantization (CVQ) has been built. The idea of CVQ is to quantize each pixel to a codeword by considering not just the pixel itself but its neighbors and their codeword identities as well. 100 samples of each character are collected from 100 writers, among them, 92 are used for training and 8 for testing. The characters are scanned by a 300 dpi scanner, which are then noise removed, thinned, segmented and size normalized. Stroke counts and segment strengths are adopted as observation features. For a vocabulary of 470 simplified Chinese characters, a recognition rate of 97% is achieved.

A theoretical framework for Bayesian adaptive training of the parameters of a discrete hidden Markov model (DHMM) and of a semi-continuous HMM (SCHMM) with Gaussian mixture state observation densities is presented. In addition to formulating the forward-backward MAP (maximum a posteriori) and the segmental MAP algorithms for estimating the above HMM parameters, a computationally efficient segmental quasi-Bayes algorithm for estimating the state-specific mixture coefficients in SCHMM is developed. For estimating the parameters of the prior densities, a new empirical Bayes method based on the moment estimates is also proposed. The MAP algorithms and the prior parameter specification are directly applicable to training speaker adaptive HMMs. Practical issues related to the use of the proposed techniques for HMM-based speaker adaptation are studied. The proposed MAP algorithms are shown to be effective especially in the cases in which the training or adaptation data are limited.>

A theoretical framework for Bayesian adaptive learning of semi-continuous HMM parameters is presented. Formulations of MAP estimation of SCHMM parameters are developed. An empirical Bayes method to estimate the hyperparameters of prior densities based on the moment estimate is proposed. Practical issues related to the use of the proposed technique for speaker adaptation application are studied. Effects of various adaptation schemes are examined and their viability is confirmed in a series of comparative experiments using a 26-word English alphabet vocabulary. The proposed method is applicable to other problems in HMM training for speech recognition such as sequential training, context adaptation and parameter smoothing.>

This paper presents an artificial neural network (ANN) for speaker-independent isolated word speech recognition. The network consists of three subnets in concatenation. The static information within one frame of speech signal is processed in the probabilistic mapping subnet that converts an input vector of acoustic features into a probability vector whose components are estimated probabilities of the feature vector belonging to the phonetic classes that constitute the words in the vocabulary. The dynamics capturing subnet computes the first-order cross correlation between the components of the probability vectors to serve as the discriminative feature derived from the interframe temporal information of the speech signal. These dynamic features are passed for decision-making to the classification subnet, which is a multilayer perceptron (MLP). The architecture of these three subnets are described, and the associated adaptive learning algorithms are derived. The recognition results for a subset of the DARPA TIMIT speech database are reported. The correct recognition rate of the proposed ANN system is 95.5%, whereas that of the best of continuous hidden Markov model (HMM)-based systems is only 91.0%.>

A static model (SM) in the form of a single vector is proposed to represent the temporal properties of a sequence of speech feature vectors. In contrast to a hidden Markov model which captures the conditional probabilities of state transitions of consecutive observations x/sup to //sub t/ and x/sup to //sub t+1/ over time, an SM captures their average joint probabilities of belonging to a pair of phonetic classes omega /sub i/ and omega /sub j/ without any Markovian assumption. SM is tested with isolated words derived from the TIMIT database as well as artificially created words. The vocabulary is a subset of TIMIT consisting of 21 words derived from the two 'sa' sentences spoken by 420 speakers. The artificial vocabulary of 10 words is designed to study the limitations of SM. Experimental results indicate that apart from a rather mild limitation of SM in handling a certain type of vocabulary, SM actually performs better than baselined continuous hidden Markov models (CHMM) in terms of recognition rate as far as isolated word recognition is concerned, and it takes only 60% of the time needed by CHMM in recognition.>

Based on the assumption that most probability densities in real life can be approximated by a mixture of Gaussian densities, we propose here a three-layer adaptive network with each neuron in the lower hidden layer representing a Gaussian basis function (covariance matrix equal to [Formula: see text] where I is a unit matrix) to estimate various probability densities and serve as a Bayes classifier. The width of the basis function [Formula: see text] may be the same for all neurons in this layer or it may vary from one neuron to another. This paper investigates the effectiveness of the network for both cases and presents a localized learning algorithm to adjust the network parameters. The network was trained with artificial data derived from known mixtures of memoryless Gaussian sources as well as exponential and Gamma densities. The performance of the network as a pattern density estimator was measured in terms of the relative difference between the target probability density function (p.d.f.) which generates the training and testing data and the network output representing the estimation. Samples from two mixtures corresponding to two classes were used to test the network capability as a classifier by comparing its error rate against that of a Bayes classifier. Both one- and two-dimensional cases were explored. The successfulness of the network depended on how well the target p.d.f.’s were represented by the training samples, the number of hidden neurons employed in the network and how thoroughly the network was trained. It was also found that allowing each basis function to have an independent width had a predominant effect on the network performance.

A new algorithm for constructing entropy reduction based decision tree classifier in two steps for large reference-class sets is proposed. The d-dimensional feature space is first mapped onto a line thus allowing a dynamic choice of features and then the resultant linear space is partitioned into two sections while minimizing the average system entropy. Classes of each section, again considered as a collection of references classes in a d-dimensional feature space, can be further split in a similar manner should the collection still be considered excessively large, thus forming a binary decision tree of nodes with overlapping members. The advantage of using such a classifier is that the need to match a test feature vector against all the references exhaustively is avoided. We demonstrate in this paper that discrete syllable recognition with dynamic programming equipped with such a classifier can reduce the recognition time by a factor of 40 to 100. The recognition speed is one third to one half of that using hidden Markov models (HMM), while the recognition rate is somewhat higher. The theory is considerably simpler than that of HMM but the decision tree can occupy a lot of memory space.

By means of capturing an abrupt change in the energy, zero crossing rate or oscillating rate of a speech time series, a technique has been developed to segment Putonghua syllables into unvoiced/voiced (UV/V) sections with boundaries in close agreement with those hand picked by examining the time series visually. For verification purposes, unvoiced consonants isolated by this method were classified according to their manners and places of articulation with high successful recognition rates.