In order for speech recognizers to deal with increased task perplexity, speaker variation, and environment variation, improved speech recognition is critical. Steady progress has been made along these three dimensions at Carnegie Mellon. In this paper, we review the SPHINX-II speech recognition system and summarize our recent efforts on improved speech recognition. This research was sponsored by the Defense Advanced Research Projects Agency and monitored by the Space and Naval Warfare Systems Command under Contract N00039-91-C-0158, ARPA Order No. 7239. The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of the U.S. Government. Keywords: Speech recognition, hidden Markov models, SPHINX-II 1. INTRODUCTION At Carnegie Mellon, wehave made significant progress in large-vocabulary speaker-independent continuous speech recognition during the past years [1, 2, 3]. SP...

A recent trend in automatic speech recognition systems is the use of continuous mixture-density hidden Markov models (HMMs). Despite the good recognition performance that these systems achieve on average in large vocabulary applications, there is a large variability in performance across speakers. Performance degrades dramatically when the user is radically different from the training population. A popular technique that can improve the performance and robustness of a speech recognition system is adapting speech models to the speaker, and more generally to the channel and the task. In continuous mixture-density HMMs the number of component densities is typically very large, and it may not be feasible to acquire a sufficient amount of adaptation data for robust maximum-likelihood estimates. To solve this problem, we propose a constrained estimation technique for Gaussian mixture densities. The algorithm is evaluated on the large-vocabulary Wall Street Journal corpus for both ...

In this paper we report high phone accuracies on three corpora: WSJ0, BREF and TIMIT. The main characteristics of the phone recognizer are: high dimensional feature vector (48), context- and genderdependent phone models with duration distribution, continuous density HMM with Gaussian mixtures, and n-gram probabilities for the phonotatic constraints. These models are trained on speech data that have either phonetic or orthographic transcriptions using maximum likelihood and maximum a posteriori estimation techniques. On the WSJ0 corpus with a 46 phone set we obtain phone accuraciesof 72.4% and 74.4% using 500 and 1600 CD phone units, respectively. Accuracy on BREF with 35 phones is as high as 78.7% with only 428 CD phone units. On TIMIT using the 61 phone symbols and only 500 CD phone units, we obtain a phoneaccuracyof 67.2% which correspond to 73.4% when the recognizer output is mapped to the commonly used 39 phone set. Making reference to our work on large vocabularyCSR, we show that ...

In large-vocabulary speech recognition, the decoder often encounters triphones that are not covered in the training data. These unseen triphones are usually represented by corresponding diphones or context independent monophones. We propose to use decision-tree based senones to generate needed senonic baseforms for unseen triphones. A decision tree is built for each individual Markov state of each phone, and the leaves of the trees constitute the senone codebook. To find the senone a Markov state of any triphone is associated with, we traverse the corresponding tree until we reach a leaf node, where a senone is represented. We used the DARPA 5,000-word speaker-independent Wall Street Journal dictation task to evaluate the proposed method. The word error rate was reduced by 11% when unseen triphones were modeled by the decision-tree based senones. When there were at least 5 unseen triphones in each test utterance, the error rate could be reduced by more than 20%. This research was spons...

An algorithm is proposed that achieves a good trade-off between modeling resolution and robustness by using a new, general scheme for tying of mixture components in continuous mixture-density hidden Markov model (HMM)-based speech recognizers. The sets of HMM states that share the same mixture components are determined automatically using agglomerative clustering techniques. Experimental results on ARPA's Wall-Street Journal corpus show that this scheme reduces errors by 25% over typical tied-mixture systems. New fast algorithms for computing Gaussian likelihoods--the most time-consuming aspect of continuous-density HMM systems--are also presented. These new algorithms significantly reduce the number of Gaussian densities that are evaluated with little or no impact on speech recognition accuracy. Corresponding Author: Vassilios Digalakis Address: Electronic and Computer Engineering Department Technical University of Crete, Kounoupidiana Chania, 73100 GREECE Phone: +30-821...

The main goal of this work is to develop an alternative methodology for acoustic-- phonetic modelling of speech sounds. The approach utilizes a segment--based framework to capture the dynamical behavior and statistical dependencies of the acoustic attributes used to represent the speech waveform. Temporal behavior is modelled explicitly by creating dynamic tracks of the acoustic attributes used to represent the waveform, and by estimating the spatio--temporal correlation structure of the resulting errors. The tracks serve as templates from which synthetic segments of the acoustic attributes are generated. Scoring of an hypothesized phonetic segment is then based on the error between the measured acoustic attributes and the synthetic segments generated for each phonetic model.

The Multi-State Time Delay Neural Network (MS-TDNN) integrates a nonlinear time alignment procedure (DTW) and the highaccuracy phoneme spotting capabilities of a TDNN into a connectionist speech recognition system with word-level classification and error backpropagation. We present an MS-TDNN for recognizing continuously spelled letters, a task characterized by a small but highly confusable vocabulary. Our MS-TDNN achieves 98.5/92.0% word accuracy on speaker dependent/independent tasks, outperforming previously reported results on the same databases. We propose training techniques aimed at improving sentence level performance, including free alignment across word boundaries, word duration modeling and error backpropagation on the sentence rather than the word level. Architectures integrating submodules specialized on a subset of speakers achieved further improvements. 1 INTRODUCTION The recognition of spelled strings of letters is essential for all applications involving proper names,...

This paper provides an overview of the state-of-the-art in laboratory speaker-independent, large vocabulary continuous speech recognition (LVCSR) systems with a view towards adapting such technology to the requirements of real-world applications. While in speech recognition the principal concern is to transcribe the speech signal as a sequence of words, the same core technology can be applied to domains other than dictation. The main topics addressed are acoustic-phonetic modeling, lexical representation, language modeling, decoding and model adaptation. After a brief summary of experimental results some directions towards usable systems are given. In moving from laboratory systems towards real-world applications, different constraints arise which influence the system design. The application imposes limitations on computational resources, constraints on signal capture, requirements for noise and channel compensation, and rejection capability. The difficulties and costs of adapting existing technology to new languages and application need to be assessed. Near term applications for LVCSR technology are likely to grow in somewhat limited domains such as spoken language systems for information retrieval, and limited domain dictation. Perspectives on some unresolved problems are given, indicating areas for future research

Most modern automatic speech recognition systems make use of acoustic models based on hidden Markov models. To obtain reasonable recognition performance within a large vocabulary framework, the acoustic models usually include a pronunciation model, together with complex parameter tying schemes. In many cases the pronunciation model operates on a phoneme level and is derived independently of the underlying models. In contrast, this work is aimed at improving pronunciation modelling on a sub-phone level in a combined framework. The modelling of pronunciation variation is assumed to be of special importance for recognition of spontaneous speech.

The benefits of a speech recognition machine would be many, resulting in the improvement of the quality of life for people. The design of a speech recognition system can be divided into two parts, commonly known as the front-end and back-end. The front-end deals with the conversion of the analog speech signal into features for classification. This thesis investigates optimal feature-sets for speech recognition. The objectives for an optimal feature-set are improved recognition performance, noise robustness, talker insensitivity and efficiency. Three problems that make it difficult to find optimal features are: 1) the amount of resources (time and computations) required to evaluate the performance of a feature-set, 2) the size of the feature space, and 3) the dependence of features upon some words in t...