Speaker-independent Word Recognition System Research Articles

Enhanced recognition rate of spoken Hindi paired word using probabilistic neural network approach

Probabilistic neural network (PNN) shows efficient capability in recognising low level signal patterns. PNN is a sequential arrangement of radial basis layer and a competitive transfer function layer, which picks up the highest probabilities as a final recognition result. In the present paper, wavelet transform technique has been used for attributes extraction from spoken word patterns and PNN for classification/recognition. For experimental set up, spoken Hindi (Indian official language) paired word utterances have been collected from individuals for development of database, and evaluation purpose. The proposed approach provides enhanced correct recognition rate (CRR) for spoken Hindi paired word recognition system, with considered dataset. CRR has been found to be 71.08% for speaker independent spoken Hindi paired word recognition system.

Improvements in HMM-based isolated word recognition system

A speaker independent isolated word recognition system for voice activated robots is introduced. Since such an application requires a fast and accurate recognition system, discrete hidden Markov models and vector quantisation techniques have been used. Different initialisations for VQ and HMM training are tested to obtain improvements over the other systems. The final models can be evaluated using a temporal normalisation of the HMM scores. A threshold based rejector can also be established using this temporal normalisation.

Speaker-independent word recognition method and system based upon zero-crossing rate and energy measurement of analog speech signal

First Page

A speaker-independent word-recognition system using multiple classification functions

A speaker-independent word-recognition system has been developed using multiple classification functions for separating 100 spoken words. The speech signal is first analysed and then non-uniformly time-sampled by referring to word-structure tables to construct a word pattern vector of 120 dimensions. Equivalently piece-wise quadratic classification functions are calculated based on a linear-programming algorithm using a large number of spoken-word design samples. A hardware system for real-time recognition has been built as a high-speed microprocessor complex. Using the classification functions calculated from design samples of 100 speakers, a recognition rate of 99% has been obtained for 50 unknown speakers.

Speaker-independent word recognition system for a large vocabulary based on phoneme recognition with high speed and compact hardware.

Speaker-independent word recognition system for a large vocabulary based on phoneme recognition with high speed and compact hardware.

A Study on the Ability to Automatically Recognize Telephone-Quality Speech From Large Customer Populations

To ascertain whether a speaker-independent word recognition system, using current technology, could function in normal telephone environments, it was necessary to conduct a study under such real-world conditions. Such an experiment was described by Wilpon and Rabiner (1983), in which telephone customers, speaking under ordinary telephone conditions, in Portland, Maine, were asked to speak their telephone number as a sequence of isolated digits. For each customer a maximum of four digits were obtained. The results from that study were very encouraging and led to further improvements in our recognition systems. To further study the feasibility of implementing speech recognition systems for general use over the telephone network, another field study was initiated. In this test, spoken seven-digit telephone numbers were obtained from a large number of telephone customers over a variety of transmission facilities in Baton Rouge, Louisiana. This paper presents the results of several recognition experiments performed on this database. Experiments were also carried out quantifying the robustness of template sets created in Portland, Baton Rouge, and under laboratory conditions in our Murray Hill, New Jersey, laboratory. Finally, a recognition system that incorporates syntactic information available in a seven-digit telephone is discussed. Our tests indicate a number of distinct real-world problems that must be considered when implementing a speech recognition system for widespread use. A discussion of the overall results and the implications for future research will be given.

Speaker independent word recognition system based on phoneme recognition for a large size (212 words) vocabulary.

This paper describes a speaker-independent spoken word recognition system for a large size vocabulary. Speech is analyzed by the filter bank, from whose logarithmic spectrum the 11 features are extracted every 10 ms. Using the features the speech is first segmented and the primary phoneme recognition is carried out for every segment using the Bayes decision method. After correcting errors in segmentation and phoneme recognition, secondary recognition of a part of the consonants is carried out and the phonemic sequence is determined. The word dictionary item having maximum likelihood to the sequence is chosen as the recognition output. The 75.9% score for the phoneme recognition and the 92.4% score for the word recognition are obtained for the training samples in the 212 words uttered by 10 male and 10 female speakers. For the same words uttered by 30 male and 20 female speakers different from the above speakers, the 88.1% word recognition score is obtained.

Some preliminary results on clustering for speaker‐independent word recognition

Recently, Rabiner [J. Acoust. Soc. Am. 62, S64(A) (1977)] demonstrated that clustering is a powerful tool for constructing composite reference templates for speaker independent word recognition. This talk will describe a set of specially designed clustering procedures for identifying the structure in training data for a speaker independent word recognition system. The clustering procedures identify coarse structure, fine structure, overlap among clusters and outliers, and can be used interactively or automatically on either multivariate or metric data. The techniques have been applied to a large data base consisting of four repetitions of a 39‐word vocabulary spoken by 50 male and 50 female speakers. Preliminary results show that the data is highly structured in a manner such that composite reference templates can be constructed in a natural way. Corroborating evidence of the structure is derived by means of a multidimensional scaling technique. Statistics of the analyses are given which support the hypothesis that the composite templates produced will form the basis for a reference set for a speaker independent word recognition system.

Application of an LPC distance measure to the voiced-unvoiced-silence detection problem

One of the most difficult problems in speech analysis is reliable discrimination among silence, unvoiced speech, and voiced speech which has been transmitted over a telephone line. Although several methods have been proposed for making this three-level decision, these schemes have met with only modest success. In this paper, a novel approach to the voiced-unvoiced-silence detection problem is proposed in which a spectral characterization of each of the three classes of signal is obtained during a training session, and an LPC distance measure and an energy distance are nonlinearly combined to make the final discrimination. This algorithm has been tested over conventional switched telephone lines, across a variety of speakers, and has been found to have an error rate of about 5 percent, with the majority of the errors (about <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">\frac{2}{3}</tex> ) occurring at the boundaries between signal classes. The algorithm is currently being used in a speaker-independent word recognition system.

On-line, adaptive speaker-independent word recognition system based on phonetic recognition techniques

The research reported in this paper deals with a new method of phonemic analysis of speech by statistical pattern recognition techniques and its application to the problem of Automatic Speech Recognition (ASR). An on-line, adaptive, trainable speaker-independent system is implemented using this approach. The details of the system follow: first, the beginning and end points of the speech utterance are detected. The utterance is then sent for automatic segmentation where it is segmented into the following classes: (1) voiced, (2) unvoiced, (3) transition, and (4) silence. An 11-dimensional feature vector consisting of 10 linear predictor coefficients and zero-crossing rate is extracted from these regions. For voiced and transition region, the feature extraction is done pitch synchronously and for unvoiced regions, a constant frame of 6.4 msec is used. A new phonetic unit called phoneme-pair is defined for the transition regions, while the unvoiced and voiced regions are represented using the phonemes of the IPA. Conditional probability densities for each of the phonemes and phoneme-pairs are estimated using non-parametric methods as a single polynomial in the 11-dimensional space. The classifier makes Bayes' minimum risk decision based on these probability densities. The recognition results of the ASR system are Training Set: 98.4%, Test Set: 96.0% (for speakers in the training set) and 91.0% (for speakers not in the training set). The present vocabulary of the system is 60 words and any new word can be added by entering its corresponding phonetic transcription. The adaptive and trainable characteristics of the system will also be demonstrated.