Les Niles - Academia.edu (original) (raw)

Papers by Les Niles

Research paper thumbnail of Muscle fiber conduction velocity: dip analysis versus cross correlation techniques

Electromyography and clinical neurophysiology, 1991

Among the various techniques in use for computing Muscle Fiber Conduction Velocities (MFCV) the C... more Among the various techniques in use for computing Muscle Fiber Conduction Velocities (MFCV) the Cross Correlation (CCT) and the Dip Analysis (DAT) Techniques are the most similar to each other. The CCT has been applied to intramuscular and surface EMG ...

Research paper thumbnail of Training methods for a connectionist model of consonant-vowel syllable recognition

IEEE International Conference on Neural Networks, 1988

A description is given of several CV (consonant-vowel) syllable recognition experiments using neu... more A description is given of several CV (consonant-vowel) syllable recognition experiments using neural network learning and retrieval paradigms. Previously (1988), the authors trained both one- and two-speaker systems to classify phonemes from the set {b d g}×{a i u} at a rate over 90%. However, when more phonemes are added to the system, additional training techniques are necessary to maintain

Research paper thumbnail of Multipage Document Images on the Internet

While client/server document imaging systems have matured considerably, fully satisfactory mechan... more While client/server document imaging systems have matured considerably, fully satisfactory mechanisms for distributing and providing interactive access to document images over the World-Wide Web have not yet emerged. The interface functionality of most scanned document viewers and browsers is primitive compared to what is available for revisable-form electronic documents. Common image viewers provide only scrolling within a page, change of magnification and jumping to the next/previous page. By contrast, electronic document browsers often provide content-based operations such as string search with highlighting of search hits, up-down-next-previous navigation through logical structure trees, and hypertext links from indexes and tables of contents to body text. Recently, there have been a number of efforts aimed at enlivening imaged documents by providing more contentbased interfaces. Examples include Adobe Capture, Dienst, Xerox's DocuWeb, and the UC Berkeley multivalent documen...

Research paper thumbnail of Error-correcting training for phoneme spotting

[Proceedings] ICASSP-92: 1992 IEEE International Conference on Acoustics, Speech, and Signal Processing, 1992

Error-correcting training is extended to spotting algorithms. An error measure for spotting tasks... more Error-correcting training is extended to spotting algorithms. An error measure for spotting tasks is defined, and the derivatives of that error with respect to the parameters of the spotter are derived. The derivatives are computed by modifying the backward pass of the forward-backward algorithm. The training algorithm is applicable to any hidden Markov model (HMM)-based classification task in which a

Research paper thumbnail of Neural networks, maximum mutual information training, and maximum likelihood training (speech recognition)

International Conference on Acoustics, Speech, and Signal Processing

A Gaussian-model classifier trained by maximum mutual information estimation (MMIE) is compared t... more A Gaussian-model classifier trained by maximum mutual information estimation (MMIE) is compared to one trained by maximum-likelihood estimation (MLE) and to an artificial neural network (ANN) on several classification tasks. Similarity of MMIE and ANN results for uniformly distributed data confirm that the ANN is better than the MLE in some cases due to the ANNs use of an error-correcting

Research paper thumbnail of A comparison of three feature vector clustering procedures in a speech recognition paradigm

ICASSP '83. IEEE International Conference on Acoustics, Speech, and Signal Processing

This paper describes the design, implementation and results of the image-based ego-motion estimat... more This paper describes the design, implementation and results of the image-based ego-motion estimation algorithm. As a source data the images captured from the bike platform are used. The device is supposed to be a part of a mobile mapping system prototype. Firstly the feature detection and matching is carried out providing the set of characteristic points in all images in the sequence. The 5-point solution based on the Gröbner basis is used to solve for essential matrices and to reject outliers. Least-square relative pose model fitting is accomplished using quaternion-based bundle adjustment. In the next step the modified Horn formula is used to recover bike trajectory up to the absolute orientation. Within this step the scene structure recovery is provided in the form of a point cloud. Finally ground control information is used to obtain data geo-referencing and the accuracy analysis. Obtained results provide satisfying robustness and accuracy. However some improvements and development scenarios are suggested.

Research paper thumbnail of Combining hidden Markov model and neural network classifiers

International Conference on Acoustics, Speech, and Signal Processing

S8.2 COMBINING HIDDEN MARKOV MODEL AND NEURAL NETWORK CLASSIFIERS Les T. Niks and Harvey F. Silve... more S8.2 COMBINING HIDDEN MARKOV MODEL AND NEURAL NETWORK CLASSIFIERS Les T. Niks and Harvey F. Silverman LEMS, Division of Engineering, Brown University, Providence, RI 02912 ABSTRACT ... MK}) = £>gPr{y„|,V(c„} = 5>c», n n gmmi = £ jbg XcAn) _ ]og^ J2 At ...

Research paper thumbnail of Architecture for a real-time LPC-based feature measurement integrated circuit

ICASSP '84. IEEE International Conference on Acoustics, Speech, and Signal Processing

An architecture for an integrated circuit to perform LPC-based feature measurement in real time h... more An architecture for an integrated circuit to perform LPC-based feature measurement in real time has been developed for speech recognition applications. The integrated circuit architecture is suitable for both isolated word recognition, in which the pattern matching occurs after the end of the utterance, and connected word recognition, where the pattern matching proceeds in synchrony with the speech input. A major feature of this architecture is the presence of stored program control which implements the LPC-based feature extraction algorithm on a single set of computational resources. Preliminary timing analysis indicates that a portion of real time remains unused. Thus, in addition to performing standard LPC-based feature analysis in real time, through program modification and memory addition, the architecture is expected to support more advanced concepts in speech recognition such as vector quantization. To aid in program development, software tools which include an assembler and a simulator and run on the UNIX* operating system have been developed. The projected chip complexity is approximately 20,000 transistors of random logic, 40,000 bits of ROM, and 2,500 bits of RAM.

Research paper thumbnail of Multimodal browsing of images in Web documents

SPIE Proceedings, 1999

In this paper, we describe a system for performing browsing and retrieval on a collection of web ... more In this paper, we describe a system for performing browsing and retrieval on a collection of web images and associated text on an HTML page. Browsing is combined with retrieval to help a user locate interesting portions of the corpus, without the need to formulate a query well matched to the corpus. Multi-modal information, in the form of text surrounding

Research paper thumbnail of How limited training data can allow a neural network to outperform an 'optimal' statistical classifier

International Conference on Acoustics, Speech, and Signal Processing

Research paper thumbnail of A connectionist model for consonant-vowel syllable recognition

ICASSP-88., International Conference on Acoustics, Speech, and Signal Processing

The authors describe preliminary CV (consonant-vowel syllable) recognition experiments using neur... more The authors describe preliminary CV (consonant-vowel syllable) recognition experiments using neural network learning and retrieval paradigms. They have trained both one and two speaker systems and report on the results of both speaker dependent and speaker independent testing. The one-speaker systems performed at 94 percent correct classifying the three voiced stop consonants learned in 3 different vowel contexts using 40ms

Research paper thumbnail of <title>Multipage document images on the Internet</title>

Multimedia Computing and Networking 1996, 1996

While client/server document imaging systems have matured considerably, fully satisfactory mechan... more While client/server document imaging systems have matured considerably, fully satisfactory mechanisms for distributing and providing interactive access to document images over the World-Wide Web have not yet emerged. The interface functionality of most scanned document viewers and browsers is primitive compared to what is available for revisable-form electronic documents. Common image viewers provide only scrolling within a page, change of magnification and jumping to the next/previous page. By contrast, electronic document browsers often provide content-based operations such as string search with highlighting of search hits, up-down-next-previous navigation through logical structure trees, and hypertext links from indexes and tables of contents to body text. Recently, there have been a number of efforts aimed at enlivening imaged documents by providing more content-based interfaces. Examples include Adobe Capture, Dienst, Xerox's DocuWeb, and the UC Berkeley multivalent document browser. This paper reviews some of the methods currently used for transmitting and browsing page images of documents on the Internet and presents a design for adding some desirable features to future document image browsers.

Research paper thumbnail of EMG interference pattern power spectrum analysis in neuro-muscular disorders

Electromyography and clinical neurophysiology

One hundred sixty six subjects, who had clinically proven Neuromuscular (NM) diagnoses and 37 nor... more One hundred sixty six subjects, who had clinically proven Neuromuscular (NM) diagnoses and 37 normal controls, had their biceps muscle EMGs, recorded in sustained isometric maximum voluntary contraction (MVC). The EMG signals were repeatedly transformed into power spectra (PS). The PS were then analysed by various statistical methods. The statistical analyses showed an overall significant difference in power between the sexes but only in the Dysschwannian Neuropathy group and no differences due to age; a very highly significant fatigue trend that manifests, to a different extent, in all frequency bands; the groups were significantly different from one another in both total power and in band-specific power, and to differ in their responses to fatigue. These analyses showed that based on PS alone discriminant analysis can separate the cases into only two significantly different groups: normal controls could not be separated from neuropathies, but both were significantly different from...

Research paper thumbnail of Improved techniques for measuring muscle fiber conduction velocity

Research paper thumbnail of Muscle fiber conduction velocity and mean power spectrum frequency in neuromuscular disorders and in fatigue

Muscle & Nerve, 1992

This study investigated the relation of muscle fiber conduction velocity (MFCV) to difference pow... more This study investigated the relation of muscle fiber conduction velocity (MFCV) to difference power spectrum mean frequency (MF), their fatigue trends, and differences between their values and their fatigue trends in various neuromuscular disorders. Electromyographic interference pattern was recorded inside the biceps in continuous isometric maximal voluntary contractions. Each subject was encouraged to pull for as long as possible. Fatigue was calculated as percent of time to complete inability to sustain contraction. The MFCV was computed by cross-correlation. The MF was computed by differencing, windowing, FFT, squaring of coefficient, and repeat averaging. There were 33 healthy, 86 polyneuropathic, 28 myasthenic, 13 myotonic, and 32 myopathic patients. Both MFCV and MF changed significantly with fatigue-the MFCV linearly, while the MF in a markedly nonlinear fashion. Both were found to be insensitive to the end stages of muscle fatigue-the MFCV did not change its slope toward complete fatigue, and the MF did not change at all beyond the 40% fatigue point. A statistically sound fatigue regression equation was derived for each, and a nonlinear equation was found to best describe their relationship. Neither MFCV nor its fatigue changes were found to be significantly different across the neuromuscular disorders. The MF, however, was found to be significantly different in some neuromuscular disorders in both its average values and fatigue trends. This study showed, in contrast to the literature, a nonlinear relationship between MFCV and MF. It also shows that neither the MFCV nor the MF had reasonable diagnostic power on its own; however, the MF was very promising to serve as an adjunct to other variables.

Research paper thumbnail of Hidden Markov model/neural network training techniques for connected alphadigit speech recognition

A neural network formulation for an HMM (hidden Markov model) is presented, and training using ma... more A neural network formulation for an HMM (hidden Markov model) is presented, and training using maximum likelihood, maximum mutual information, minimum mean-squared-error (MMSE), and unconstrained MMSE is described. Recognition results are presented for the variously trained models evaluated on a speaker-independent, connected alphadigit speech recognition task. It is concluded that viewing neural networks as HMMs provides a framework for building

Research paper thumbnail of Timit phoneme recognition using an HMM-derived recurrent neural network

… European Conference on Speech Communication and …, 1991

Model TIMIT Labels ih2 ih, ix ah2 ah, ax, axh u2 uw, ux ao2 ao, aa er2 er, axr 12 1, el m2 m, em ... more Model TIMIT Labels ih2 ih, ix ah2 ah, ax, axh u2 uw, ux ao2 ao, aa er2 er, axr 12 1, el m2 m, em n2 n, nx, en ng2 ng, eng sh2 sh, zh hh2 hh, hv sil2 q, bel, del, gel, pel, tei, kel, epi, pau, sil Table 1. Composite models, formed by combining models for two or more TIMIT phones, "sil" ...

Research paper thumbnail of Muscle fiber conduction velocity: dip analysis versus cross correlation techniques

Electromyography and clinical neurophysiology, 1991

Among the various techniques in use for computing Muscle Fiber Conduction Velocities (MFCV) the C... more Among the various techniques in use for computing Muscle Fiber Conduction Velocities (MFCV) the Cross Correlation (CCT) and the Dip Analysis (DAT) Techniques are the most similar to each other. The CCT has been applied to intramuscular and surface EMG ...

Research paper thumbnail of Training methods for a connectionist model of consonant-vowel syllable recognition

IEEE International Conference on Neural Networks, 1988

A description is given of several CV (consonant-vowel) syllable recognition experiments using neu... more A description is given of several CV (consonant-vowel) syllable recognition experiments using neural network learning and retrieval paradigms. Previously (1988), the authors trained both one- and two-speaker systems to classify phonemes from the set {b d g}×{a i u} at a rate over 90%. However, when more phonemes are added to the system, additional training techniques are necessary to maintain

Research paper thumbnail of Multipage Document Images on the Internet

While client/server document imaging systems have matured considerably, fully satisfactory mechan... more While client/server document imaging systems have matured considerably, fully satisfactory mechanisms for distributing and providing interactive access to document images over the World-Wide Web have not yet emerged. The interface functionality of most scanned document viewers and browsers is primitive compared to what is available for revisable-form electronic documents. Common image viewers provide only scrolling within a page, change of magnification and jumping to the next/previous page. By contrast, electronic document browsers often provide content-based operations such as string search with highlighting of search hits, up-down-next-previous navigation through logical structure trees, and hypertext links from indexes and tables of contents to body text. Recently, there have been a number of efforts aimed at enlivening imaged documents by providing more contentbased interfaces. Examples include Adobe Capture, Dienst, Xerox's DocuWeb, and the UC Berkeley multivalent documen...

Research paper thumbnail of Error-correcting training for phoneme spotting

[Proceedings] ICASSP-92: 1992 IEEE International Conference on Acoustics, Speech, and Signal Processing, 1992

Error-correcting training is extended to spotting algorithms. An error measure for spotting tasks... more Error-correcting training is extended to spotting algorithms. An error measure for spotting tasks is defined, and the derivatives of that error with respect to the parameters of the spotter are derived. The derivatives are computed by modifying the backward pass of the forward-backward algorithm. The training algorithm is applicable to any hidden Markov model (HMM)-based classification task in which a

Research paper thumbnail of Neural networks, maximum mutual information training, and maximum likelihood training (speech recognition)

International Conference on Acoustics, Speech, and Signal Processing

A Gaussian-model classifier trained by maximum mutual information estimation (MMIE) is compared t... more A Gaussian-model classifier trained by maximum mutual information estimation (MMIE) is compared to one trained by maximum-likelihood estimation (MLE) and to an artificial neural network (ANN) on several classification tasks. Similarity of MMIE and ANN results for uniformly distributed data confirm that the ANN is better than the MLE in some cases due to the ANNs use of an error-correcting

Research paper thumbnail of A comparison of three feature vector clustering procedures in a speech recognition paradigm

ICASSP '83. IEEE International Conference on Acoustics, Speech, and Signal Processing

This paper describes the design, implementation and results of the image-based ego-motion estimat... more This paper describes the design, implementation and results of the image-based ego-motion estimation algorithm. As a source data the images captured from the bike platform are used. The device is supposed to be a part of a mobile mapping system prototype. Firstly the feature detection and matching is carried out providing the set of characteristic points in all images in the sequence. The 5-point solution based on the Gröbner basis is used to solve for essential matrices and to reject outliers. Least-square relative pose model fitting is accomplished using quaternion-based bundle adjustment. In the next step the modified Horn formula is used to recover bike trajectory up to the absolute orientation. Within this step the scene structure recovery is provided in the form of a point cloud. Finally ground control information is used to obtain data geo-referencing and the accuracy analysis. Obtained results provide satisfying robustness and accuracy. However some improvements and development scenarios are suggested.

Research paper thumbnail of Combining hidden Markov model and neural network classifiers

International Conference on Acoustics, Speech, and Signal Processing

S8.2 COMBINING HIDDEN MARKOV MODEL AND NEURAL NETWORK CLASSIFIERS Les T. Niks and Harvey F. Silve... more S8.2 COMBINING HIDDEN MARKOV MODEL AND NEURAL NETWORK CLASSIFIERS Les T. Niks and Harvey F. Silverman LEMS, Division of Engineering, Brown University, Providence, RI 02912 ABSTRACT ... MK}) = £>gPr{y„|,V(c„} = 5>c», n n gmmi = £ jbg XcAn) _ ]og^ J2 At ...

Research paper thumbnail of Architecture for a real-time LPC-based feature measurement integrated circuit

ICASSP '84. IEEE International Conference on Acoustics, Speech, and Signal Processing

An architecture for an integrated circuit to perform LPC-based feature measurement in real time h... more An architecture for an integrated circuit to perform LPC-based feature measurement in real time has been developed for speech recognition applications. The integrated circuit architecture is suitable for both isolated word recognition, in which the pattern matching occurs after the end of the utterance, and connected word recognition, where the pattern matching proceeds in synchrony with the speech input. A major feature of this architecture is the presence of stored program control which implements the LPC-based feature extraction algorithm on a single set of computational resources. Preliminary timing analysis indicates that a portion of real time remains unused. Thus, in addition to performing standard LPC-based feature analysis in real time, through program modification and memory addition, the architecture is expected to support more advanced concepts in speech recognition such as vector quantization. To aid in program development, software tools which include an assembler and a simulator and run on the UNIX* operating system have been developed. The projected chip complexity is approximately 20,000 transistors of random logic, 40,000 bits of ROM, and 2,500 bits of RAM.

Research paper thumbnail of Multimodal browsing of images in Web documents

SPIE Proceedings, 1999

In this paper, we describe a system for performing browsing and retrieval on a collection of web ... more In this paper, we describe a system for performing browsing and retrieval on a collection of web images and associated text on an HTML page. Browsing is combined with retrieval to help a user locate interesting portions of the corpus, without the need to formulate a query well matched to the corpus. Multi-modal information, in the form of text surrounding

Research paper thumbnail of How limited training data can allow a neural network to outperform an 'optimal' statistical classifier

International Conference on Acoustics, Speech, and Signal Processing

Research paper thumbnail of A connectionist model for consonant-vowel syllable recognition

ICASSP-88., International Conference on Acoustics, Speech, and Signal Processing

The authors describe preliminary CV (consonant-vowel syllable) recognition experiments using neur... more The authors describe preliminary CV (consonant-vowel syllable) recognition experiments using neural network learning and retrieval paradigms. They have trained both one and two speaker systems and report on the results of both speaker dependent and speaker independent testing. The one-speaker systems performed at 94 percent correct classifying the three voiced stop consonants learned in 3 different vowel contexts using 40ms

Research paper thumbnail of <title>Multipage document images on the Internet</title>

Multimedia Computing and Networking 1996, 1996

While client/server document imaging systems have matured considerably, fully satisfactory mechan... more While client/server document imaging systems have matured considerably, fully satisfactory mechanisms for distributing and providing interactive access to document images over the World-Wide Web have not yet emerged. The interface functionality of most scanned document viewers and browsers is primitive compared to what is available for revisable-form electronic documents. Common image viewers provide only scrolling within a page, change of magnification and jumping to the next/previous page. By contrast, electronic document browsers often provide content-based operations such as string search with highlighting of search hits, up-down-next-previous navigation through logical structure trees, and hypertext links from indexes and tables of contents to body text. Recently, there have been a number of efforts aimed at enlivening imaged documents by providing more content-based interfaces. Examples include Adobe Capture, Dienst, Xerox's DocuWeb, and the UC Berkeley multivalent document browser. This paper reviews some of the methods currently used for transmitting and browsing page images of documents on the Internet and presents a design for adding some desirable features to future document image browsers.

Research paper thumbnail of EMG interference pattern power spectrum analysis in neuro-muscular disorders

Electromyography and clinical neurophysiology

One hundred sixty six subjects, who had clinically proven Neuromuscular (NM) diagnoses and 37 nor... more One hundred sixty six subjects, who had clinically proven Neuromuscular (NM) diagnoses and 37 normal controls, had their biceps muscle EMGs, recorded in sustained isometric maximum voluntary contraction (MVC). The EMG signals were repeatedly transformed into power spectra (PS). The PS were then analysed by various statistical methods. The statistical analyses showed an overall significant difference in power between the sexes but only in the Dysschwannian Neuropathy group and no differences due to age; a very highly significant fatigue trend that manifests, to a different extent, in all frequency bands; the groups were significantly different from one another in both total power and in band-specific power, and to differ in their responses to fatigue. These analyses showed that based on PS alone discriminant analysis can separate the cases into only two significantly different groups: normal controls could not be separated from neuropathies, but both were significantly different from...

Research paper thumbnail of Improved techniques for measuring muscle fiber conduction velocity

Research paper thumbnail of Muscle fiber conduction velocity and mean power spectrum frequency in neuromuscular disorders and in fatigue

Muscle & Nerve, 1992

This study investigated the relation of muscle fiber conduction velocity (MFCV) to difference pow... more This study investigated the relation of muscle fiber conduction velocity (MFCV) to difference power spectrum mean frequency (MF), their fatigue trends, and differences between their values and their fatigue trends in various neuromuscular disorders. Electromyographic interference pattern was recorded inside the biceps in continuous isometric maximal voluntary contractions. Each subject was encouraged to pull for as long as possible. Fatigue was calculated as percent of time to complete inability to sustain contraction. The MFCV was computed by cross-correlation. The MF was computed by differencing, windowing, FFT, squaring of coefficient, and repeat averaging. There were 33 healthy, 86 polyneuropathic, 28 myasthenic, 13 myotonic, and 32 myopathic patients. Both MFCV and MF changed significantly with fatigue-the MFCV linearly, while the MF in a markedly nonlinear fashion. Both were found to be insensitive to the end stages of muscle fatigue-the MFCV did not change its slope toward complete fatigue, and the MF did not change at all beyond the 40% fatigue point. A statistically sound fatigue regression equation was derived for each, and a nonlinear equation was found to best describe their relationship. Neither MFCV nor its fatigue changes were found to be significantly different across the neuromuscular disorders. The MF, however, was found to be significantly different in some neuromuscular disorders in both its average values and fatigue trends. This study showed, in contrast to the literature, a nonlinear relationship between MFCV and MF. It also shows that neither the MFCV nor the MF had reasonable diagnostic power on its own; however, the MF was very promising to serve as an adjunct to other variables.

Research paper thumbnail of Hidden Markov model/neural network training techniques for connected alphadigit speech recognition

A neural network formulation for an HMM (hidden Markov model) is presented, and training using ma... more A neural network formulation for an HMM (hidden Markov model) is presented, and training using maximum likelihood, maximum mutual information, minimum mean-squared-error (MMSE), and unconstrained MMSE is described. Recognition results are presented for the variously trained models evaluated on a speaker-independent, connected alphadigit speech recognition task. It is concluded that viewing neural networks as HMMs provides a framework for building

Research paper thumbnail of Timit phoneme recognition using an HMM-derived recurrent neural network

… European Conference on Speech Communication and …, 1991

Model TIMIT Labels ih2 ih, ix ah2 ah, ax, axh u2 uw, ux ao2 ao, aa er2 er, axr 12 1, el m2 m, em ... more Model TIMIT Labels ih2 ih, ix ah2 ah, ax, axh u2 uw, ux ao2 ao, aa er2 er, axr 12 1, el m2 m, em n2 n, nx, en ng2 ng, eng sh2 sh, zh hh2 hh, hv sil2 q, bel, del, gel, pel, tei, kel, epi, pau, sil Table 1. Composite models, formed by combining models for two or more TIMIT phones, "sil" ...