Isolated Word Speech Recognition Research Articles

Some performance benchmarks for isolated work speech recognition systems

The performance of isolated word speech recognition system has steadily improved over time as we learn more about how to represent the significant events in speech, and how to capture these events via appropriate analysis procedures and training algorithms. In particular, algorithms based on both template matching (via dynamic time warping (DTW) procedures) and hidden Markov models (HMMs) have been developed which yield high accuracy on several standard vocabularies, including the 10 digits (zero to nine) and the set of 26 letters of the English alphabet (A-Z). Results are given showing currently attainable performance of a laboratory system for both template-based (DTW) and HMM-based recognizers, operating in both speaker trained and speaker independent modes, on the digits and the alphabet vocabularies using telephone recordings. We show that the average error rates of these systems, on standard vocabularies, are significantly lower than those reported several years back on the exact same databases, thereby reflecting the progress which has been made in all aspects of the speech recognition process.

Artificial intelligence in speech understanding: Two applications at C.R.I.N.

For the past fifteen years the “Pattern Recognition and Artificial Intelligence” research group at thecrin has been involved in various projects on isolated-word and continuous speech recognition. It is now clear that the recognition of natural speech necessitates making optimum use of all available knowledge sources. Present artificial intelligence techniques, especially knowledge-based systems, can be very helpful within this framework. In this paper we explain how in our projects these techniques are taken into account: expert systems for acoustic-phonetic decoding and phonological interpretation, multi-knowledge sources cooperation for actual man-machine dialogue implementation. The basic ideas are illustrated with short examples.

Speaker independen telephone speech recognition and reference pattern generation.

This paper describes the speaker independent isolated word speech recognition method developed for telephone speech response systems. To recognize speech, input utterances are first frequency analyzed by 19 channel BPFs. The frame cycle used is 8ms. Then the analyzed data undergo logarithmic conversion, normalization of voice chords sound source characteristics by least squares approximation line and time normalization by linear companding to 32 frames. The speech patterns thus obtained undergo pattern matching with multiple reference patterns generated separately for male and female speakers in advance. In applying this recognition method, it is necessary to optimize the reference patterns so that the speech can be correctly recognized in spite of the difference of formant frequencies, the differences in individual speaker's habits, the variations of phonetic positions, non-vocalization, and slight segmentation errors. To evaluate the performance of this recognition method, voices of about 2, 000 persons were recorded through long distance telephone lines. A 16 Japanese words vocabulary was used. A total of 256 male and female reference patterns were generated using the training voice data of about 570 persons. The speech recognition accuracy of this method in recognizing non-training voice data was 97.8 %.

A study of ambient and amplitude correction in isolated utterance recognition

Experiments demonstrating the importance of ambient and amplitude correction in an isolated word speech recognition system are reported. Whenever a speaker produces the speech input in a nonanechoic environment, such as in in ordinary office, the method of ambient correction helps to neutralize the effect of any change in ambient which would otherwise adversely influence the performance of the recognizer. On the other hand, the goal of amplitude correction is to compensate for amplitude level fluctuations encountered in any normal speech input. The experiments compare the performances with and without the corrections in the context of an experimental speech recognition system developed at IBM. A database consisting of the speech collected from three male and one female speaker is used for this study. Performance is measured by examining both the error rate and the time taken by the decoder. Significant improvement in performance is observed on both counts when the corrections are applied.

Isolated-word speech recognition using multisection vector quantization codebooks

A new approach to isolated-word speech recognition using vector quantization (VQ) is examined. In this approach, words are recognized by means of sequences of VQ codebooks, called multisection codebooks. A separate multisection codebook is designed for each word in the recognition vocabulary by dividing the word into equal-length sections and designing a standard VQ codebook for each section. Unknown words are classified by dividing them into corresponding sections, encoding them with the multisection codebooks, and finding the multisection codebook that yields the smallest average distortion. For speaker-independent recognition of the digits, this approach achieved a recognition accuracy of 98 percent. In addition, the approach achieved greater than 99 percent accuracy for speaker-dependent recognition of the digits with only one distortion computation per input frame per vocabulary word. The approach is described, detailed experimental results are presented and discussed, and computational requirements are analyzed.

On improvement of performance of isolated word recognition for degraded speech

We investigate the performance of an isolated word speech recognition (IWSR) system for degraded speech. We propose a recognition scheme which adapts itself to mild degradations in speech and improves the reliability of recognition significantly. The scheme does not use a priori information regarding the nature and extent of noise. We suggest techniques which adaptively discriminate between noisy and noise-free parameters by using a selective weighting procedure in the final distance calculation. A new measure of performance is adopted to compare several recognition schemes using small data sets. Our scheme lends itself to greater flexibility in handling degradations in speech input than do the existing recognition schemes.

Signal-dependent matching for isolated word speech recognition systems

A new approach to the design of IWSR systems is proposed in this paper. This involves a dynamic matching strategy based on the nature of the input speech segment. This is called signal-dependent matching. The computational complexity in the implementation of the proposed algorithm is significantly reduced by adopting a two stage approach in matching. In the first stage, the warping path between the test utterance and a reference utterance is determined. In the second stage, the distance between the utterances is computed along the path. There will be a slight degradation in the performance of a two stage approach as compared to the single stage approach, but this can be tolerated in view of the significant computational advantage. The performance degradation is more than compensated by the signal-dependent matching strategy in the second stage. To measure the improvement in the recognition performance, a new index of performance is defined, that reflects the characteristics of the distance matrix for a given vocabulary, rather than the characteristics of the confusion matrix. The performance of the signal-dependent matching algorithm is significantly better than the standard dynamic time warping matching algorithm for confusable as well as nonconfusable vocabulary. We also develop a signal-dependent matching algorithm, which takes into account some distortions in the input speech. As an example we offer the agorithm twice the same test utterance, once undistorted, once after a distortion. Our research until now indicates a improvement in automatic isolated word speech recognition systems while using signal-dependent parameter measuring and signal dependent matching.

On-line recognition of spoken words from a large vocabulary

It is demonstrated in this paper that a real-time, large-vocabulary, isolated-word speech recognition system can effectively be implemented using the following two-stage organization: 1. (1) conversion of the speech signal into phonemic transcriptions, 2. (2) recognition of phonemic transcriptions by advanced searching methods. A comparison of several alternatives for the first stage has indicated that the best accuracy is achieved by the learning-subspace method For the second stage the authors recommend fast string searching by redundant hash addressing combined with subsequent probabilistic analysis. The above system has been implemented in a minicomputer environment.

A speaker-independent discrete utterance recognition system, combining deterministic and probabilistic strategies

A speaker-independent isolated word speech recognition system is discussed in which an unknown utterance is described by a set of speech feature measurements and then compared with a reference set of the same measurements obtained during a training procedure with a population of speakers. To reduce significantly the number of word confusions, a segmentation procedure is used. As a result, the whole vocabulary is divided into a number of subgroups characterized by a certain phonetic structure which is represented by the sequence of four types of segments. Each of these segments reflects quite reliably certain speaker independent events in the speech signal. Thus, the subgroup may consist of the reference sets for several confusable words, on the other hand, each word is represented by its reference sets in a number of the subgroups. The system was evaluated Using a 20-word vocabulary (including ten digits). A mean recognition accuracy of about 95 percent was obtained. The tradeoff between quite precise segmentation, which leads to a very bulky system if it is speaker-independent, and rough segmentation, which in some cases does not reduce the number of confusable words too much, is also discussed.

A Vector Quantizer Combining Energy and LPC Parameters and Its Application to Isolated Word Recognition

The theory of vector quantization (VQ) of linear predictive coding (LPC) coefficients has established a wide variety of techniques for quantizing LPC spectral shape to minimize overall spectral distortion. Such vector quantizers have been widely used in the areas of speech coding and speech recognition. The conventional vector quantizer utilizes only spectral shape information and essentially disregards the energy or gain term associated with the optimal LPC fit to the signal being modeled. In this paper we present a method of incorporating LPC spectral shape and energy into the code-book entries of the vector quantizer. To do this, we postulate a distortion measure for comparing two LPC vectors that uses the weighted sum of an LPC shape distortion and a log energy distortion. Based on this combined distortion measure, we have designed and studied vector quantizers of several sizes for use in isolated word speech recognition experiments. We found that a fairly significant correlation exists between LPC shape and signal energy. Hence, an LPC shape combined with energy vector quantizer with a given distortion requires far fewer code-book entries than one in which LPC shape and energy are quantized separately. Based on isolated word recognition tests on both a 10-digit and a 129-word airlines vocabulary, we found improvements in recognition accuracy by using the VQ with both LPC shape and energy over that obtained using a VQ with LPC shape alone.

Comparative study of nonlinear time warping techniques in isolated word speech recognition systems

In this paper, the effects of two major design choices on the performance of an isolated word speech recognition system are examined in detail. They are: 1) the choice of a warping algorithm among the Itakura asymmetric, the Sakoe and Chiba symmetric, and the Sakoe and Chiba asymmetric, and 2) the size of the warping window to reduce computation time. Two vocabularies were used: the digits (zero, one,..., nine) and a highly confusable subset of the alphabet (b, c, d, e, g, p, t, v, z). The Itakura asymmetric warping algorithm appears to be slightly better than the other two for the confusable vocabulary. We discuss the reasons why the performance of the algorithms is vocabulary dependent. Finally, for the data used in our experiments, a warping window of about 100 ms appears to be optimal.

An efficient method for testing large vocabulary automatic speech recognition systems

The performance of automatic speech recognition systems is commonly measured by using large quantities of natural speech as a benchmark. For recognizers which accept large vocabularies, or when many alternative vocabularies are of interest, this method requires an unreasonably large natural speech corpus. Synthetic speech which incorporates variations in pronunciation is an alternative in these cases. This approach was used to evaluate the performance of a speaker trained, isolated word speech recognition system [H. Murveit, M. Lowy, and R. W. Brodersen, J. Acoust. Soc. Am. Suppl. 1 69, S8 (1981)]. Each word to be recognized was modeled as a sequence of segments with each segment being variable length and having a uniform spectrum with added noise. A few tokens of natural speech were used to evaluate the parameters of the model. The model was then used to produce synthetic tokens for testing the recognizer. Good correlation was obtained between the confusion matrices for synthetic and natural speech when recognized. [Work supported in part by DARPA.]

An architecture of an MOS‐LSI speech recognition system using dynamic programming

In the past several years, a number of very accurate, isolated word speech recognition systems based on dynamic programming techniques have been designed and tested. However, as these techniques are computationally intensive, commercial systems using dynamic time warping have been costly. We have designed an architecture which exploits the capabilities of custom MOS‐LSI designs to implement a complete speech recognition system. This system would operate in real time using dynamic time warping, yet it would only require 4–5 integrated circuits for a moderate (50–200 word) vocabulary. This system is designed to be expandable so that larger vocabularies can be used by including additional IC's in parallel with the others. The integrated circuits which are required are two custom‐designed chips, a memory IC, and a low‐performance microcomputer for overall control. The custom chips include a front end processor for spectral analysis (currently a switched‐capacitor filter bank) with an endpoint detector, and an IC to implement the dynamic programming algorithm. The architecture of the custom chips has been defined so that adequate performance can be obtained from a standard MOS process. A connected speech recognition algorithm that is based on the use of the above IC's has been developed and will be described. It requires additional processing in the low‐speed microcomputer to perform a second level of dynamic programming.

A new training method for isolated word recognition

One of the most important operations in isolated word speech recognition systems is the method used to obtain the word reference templates. For speaker‐trained systems, the techniques that have been used include casual training, averaging, and use of statistical pattern recognition clustering methods. In a recent study Rabiner and Wilpon showed that the statistical techniques, when combined with the technique of averaging the autocorrelation coefficients of all tokens within the cluster, provided a reliable, robust set of reference templates. The only drawback to this method was the extensive, burdensome training required for the statistical analysis. Since the statistical training method could not be used in most practical situations, techniques were investigated for obtaining a simplified, robust training procedure which would incorporate many of the ideas of the statistical approach. Such a training method is described in this talk. The advantages of this new training procedure, over (previously used) casual training include: ease of use, robustness of the templates, minimized computation in the recognition algorithm, normalized durations of all references, and recognition accuracy comparable to previous systems with double the number of templates.

A simplified, robust training procedure for speaker trained, isolated word recognition systems

One of the most important operations in isolated word speech recognition systems is the method used to obtain the word reference templates. For speaker trained systems, the techniques that have been used include casual training, averaging, and use of statistical pattern recognition clustering methods. In a recent study, Rabiner and Wilpon showed that the statistical techniques, when combined with the technique of averaging the autocorrelation coefficients of all tokens within the cluster, provided a reliable, robust set of reference templates. The only drawback to this method was the extensive, burdensome training required for the statistical analysis. Since the statistical training method could not be used in most practical situations, techniques were investigated for obtaining a simplified, robust training procedure which would incorporate many of the ideas of the statistical approach. Such a training method is described in this paper. The advantages of this new training procedure, over (previously used) casual training include: ease of use, robustness of the templates, minimized computation in the recognition algorithm, normalized durations of all references, and recognition accuracy comparable to previous systems with double the number of templates.