R. Wohlford

United States Department of Defense

Author Statistics

Papers

Citation

H-Index

i-10 index

Research Trends

Author Order

Document Type

Co-Authors

B. Landell

Global Aerospace (United States)

Alan L. Higgins

Ingenium (Germany)

Conghui Tan

West Bengal Electronics Industry Development Corporation Limited (India)

L. Bahler

ITT Technical Institute

A. S. House

Public Health Agency of Canada

A Smith

Annamalai University

Edwin H. Wrench

University of California, Santa Barbara

E. Neuberg

United States Department of Defense

J. Naylor

ITT Technical Institute

Thomas B. Slack

Texas Woman's University

Cooperative Institutions

ITT Technical Institute

Nokia (United States)

West Bengal Electronics Industry Development Corporation Limited (India)

Chinese University of Hong Kong

Public Health Agency of Canada

Institute For Defense Analyses

Purdue University West Lafayette

Massachusetts Institute of Technology

Annamalai University

Government Medical College

Author Statistics

Papers

Citation

H-Index

i-10 index

Research Field

A comparison of four techniques for automatic speaker recognition

R. Wohlford Edwin H. Wrench B. Landell

Four automatic speaker recognition techniques were investigated with a contain speech data base to determine their effectiveness in a text independent mode. These four techniques used the correlation of short and long term spectral averages, cepstral measurements of long term spectral averages, orthogonal linear prediction of the speech waveform, and long term average LPC reflection coefficients combined with pitch and overall power. The results of this study indicate that LPC derived parameters perform better than do those derived from cepstral and spectral data. Recognition accuracies of 95% and 93% were obtained for LPC based techniques with 13 seconds of unknown speech. The corresponding recognition accuracies for the cepstral and spectral based systems were 79% and 54% respectively.

Mel-frequency cepstrum

Linear prediction

Cepstrum

10.1109/icassp.1980.1170940

Cite

Citations (19)

Effect of vector quantization on a continuous speech recognition system

B. Landell J. Naylor R. Wohlford

Vector quantization techniques were applied to a continuous speech recognition system as a means of reducing both memory usage and computation time. The speech recognition system computes time-aligned distances between unknown speech segments and template frames. Vector quantization allowed the replacement of speech frames (vectors) with single index numbers which referenced an ordered set, or codebook, of representative frames. Two techniques for generating this codebook, clustering and covering, were examined. The covering technique provided a significant computational advantage over the clustering technique although both techniques generated codebooks which performed well in this task. Results are presented for a ten speaker, 100 word vocabulary experiment. Using speaker dependent codebooks, system performance levels were maintained while the number of distance calculations was reduced by a factor of 2 and the template storage required was reduced by a factor of 4.6. With an increase in error rate of about one third, these factors were 6.8 and 7.8 respectively.

Linde–Buzo–Gray algorithm

Word error rate

10.1109/icassp.1984.1172586

Cite

Citations (5)

Application of a sequential pattern learning system to connected speech recognition

Alice E. Smith Jeffrey N. Denenberg Thomas B. Slack Conghui Tan R. Wohlford

An Experimental Learning Element (ELE) for learning and recognizing sequential patterns is being developed as an adaptable pattern classifier of a larger learning system. Once external patterns are converted into a linear sequence of named objects, the ELE can build models that associate input object sequences with expected output state sequences. The ELE has been successfully demonstrated in learning and recognizing hand-printed characters. This paper describes the ELE and compares its performance with a Dynamic Time Wrap (DTW) based speech recognition system on the task of connected digit recognition. If permitted to continually learn the ELE reaches the same performance level as the DTW-CSR on the same quantized speech test data.

Sequence (biology)

10.1109/icassp.1985.1168282

Cite

Citations (5)

Preliminaries to the automatic recognition of speech: language identification

The Journal of the Acoustical Society of America (1975)

A. S. House E. Neuberg R. Wohlford

The goal of automatically understanding speech usually has been approached by first simplifying the task—by studying the speech of one talker, by restricting vocabulary, by considering only high-quality signals, by analyzing one language, etc. In addition, much experimental work has been characterized by an emphasis on either the speech-analysis or the pattern-recognition/statistical aspects of the problem. A new procedure is described for automatically determining the language of a connected-speech sample. The parameters extracted from the acoustic signal are a small number of gross linguistic categories that seem to be robust and reasonably independent of talker and channel, yet carry enough information to separate languages. Successful preliminary feasibility studies using manually transcribed parameters, as well as the performance of the procedures currently in use for automatic implementation of the scheme, are described. The effects of inaccuracies in labeling on the identification of a language are considered. Implications of these studies for the implementation of schemes for word identification and speech recognition are discussed.

Identification

Language identification

Carry (investment)

Sample (material)

SIGNAL (programming language)

10.1121/1.1995184

Cite

Citations (1)

Improved automatic language identification in noisy speech

International Conference on Acoustics, Speech, and Signal Processing (2003)

Fred Goodman Arnaud Martin R. Wohlford

A description is given of enhancements to an automatic language identification algorithm previously reported. The algorithm, based on linear-predictive-coding-based formant extraction, was greatly improved, reducing the error rate by more than 50%. This performance was achieved on a large (>9 h), very noisy, six-language database, using trials of less than 10 s. Experiments that improved performance are described, including tests of various distance metrics, expanded and modified parameter sets, and a new voicing statistic. Final performance results were obtained as a function of time, signal-to-noise ratio, and no-decision rate. A new rejection capability was developed to address the open-set identification problem.< >

Identification

Word error rate

Statistic

10.1109/icassp.1989.266480

Cite

Citations (38)

Improved speech recognition in noise

B. Landell R. Wohlford L. Bahler

Recent enhancements to a connected speech recognition algorithm have led to substantial improvements in recognition accuracy especially for speech in noisy environments. The performance improvements are a direct result of three features of this algorithm: elimination of a priori speech endpoint detection; adaptation of templates for additive noise; and scoring using a likelihood ratio. Experimental results are provided which illustrate the improvements in recognition performance for both within-vocabulary substitution and out-of-vocabulary rejection.

10.1109/icassp.1986.1169211

Cite

Citations (8)

Keyword recognition using template concatenation

Alan L. Higgins R. Wohlford

The error rates typically reported for connected speech recognition (CSR) and keyword recognition (KWR) differ by at least an order of magnitude. We present a simple new KWR method and use the new method to investigate the factors that are responsible for this large apparent "performance gap". The new KWR method concatenates templates to match all the input speech, and can utilize knowledge about the vocabulary and syntax of the input speech. We present performance results for systems that differ in the following factors: syntax, speaker training, and template vocabulary. The results suggest that template-concatenation KWR methods can give good results for unrestricted input speech.

Concatenation (mathematics)

Template

10.1109/icassp.1985.1168253

Cite

Citations (61)

The enhancement of wordspotting techniques

R. Wohlford A Smith M. R. Sambur

This paper presents results of research on the enhancement of speaker independent wordspotting in conversational, telephone bandwidth speech from a variety of talkers. The research involved the comparison of five LPC based parameter sets (filter coefficients, autocorrelation coefficients, cepstral coefficients, vocal tract area functions, and pseudoformants), in three levels of additive white noise, and the evaluation of two methods of noise reduction. The techniques were studied with a dynamic programming based wordspotting system. The performance of the system was measured the percent of keywords spotted with no false alarms. The results show that pseudoformants, without the use of speaker normalization or noise reduction techniques, perform as well as or better than the other parameter sets with these techniques.

Vocal tract

Normalization

Cepstrum

Mel-frequency cepstrum

10.1109/icassp.1980.1171021

Cite

Citations (5)

Event driven speech enhancement

S. Boll R. Wohlford

A composite noise suppression algorithm is described in which the current speech event under analysis dictates which enhancement technique is to be used. Large temporal and spectral variations in speech have continued to inhibit satisfactory performance of single algorithm approaches to speech enhancement. This study examines the improvements which result from matching the restoration approach to the speech event. This study considers speech enhancement for speech recognition systems. Performance was measured according to the degree of improved phonemic separability as a function of the type of preproeessing. Phonemic separation is measured using an Itakura distance between phonetically marked templates and phonetically marked unknown speech. Results show that preprocessing the unknown noisy speech at a 10 aB signal-to-noise ratio reduces the percentage of false acceptances from 75% to 25% relative to the number measured using clean speech.

10.1109/icassp.1983.1171982

Cite

Citations (4)

A new method of text-independent speaker recognition

Alan L. Higgins R. Wohlford

Text-independent speaker recognition methods have been based on measurements of long-term statistics of individual speech frames. These methods are not capable of modeling speaker-dependent speech dynamics. In this paper, we describe a new method, based on template matching, that utilizes temporal information to advantage. The template-matching method performs text-dependent recognition as a special case. Performance of the template-matching method is compared with that of similar recently-developed methods.

Template matching

Speaker diarisation

10.1109/icassp.1986.1168975

Cite

Citations (24)