Enrique Monte

Nonlinear prediction with neural nets in ADPCM

In the last years there has been a growing interest for nonlinear speech models. Several works have been published revealing the better performance of nonlinear techniques, but little attention has been dedicated to the implementation of the nonlinear model into real applications. This work is focused on the study of the behaviour of a nonlinear predictive model based on neural nets, in a speech waveform coder. Our novel scheme obtains an improvement in SEGSNR between 1 and 2 dB for an adaptive quantization ranging from 2 to 5 bits.

A Comparative Study Between Linear and Nonlinear Speech Prediction

This paper is focused on nonlinear prediction coding, which consists on the prediction of a speech sample based on a nonlinear combination of previous samples. It is known that in the generation of the glottal pulse, the wave equation does not behave linearly [2], [10], and we model these effects by means of a nonlinear prediction of speech based on a parametric neural network model. This work is centred on the neural net weights quantization and on the compression gain.

Application of the LVQ Neural Method to a Stellar Catalogue

We present the application of the Learning Vectorial Quantification Neural Method to the study of a stellar catalogue containing more than 12000 stars. One of the first results is the appearance of three statistically well defined groups of stars that fulfill some kinematic properties corresponding to the Thin Disk, Thick Disk and Halo components of the Milky Way.

A Speech Recognition System that Integrates Neural Nets and HMM

In this paper we present a speech recognition system based on neural networks and on Hidden Markov Models. This system makes use of the discriminating properties of the multilayer perceptron, the properties of the fonotropical maps of Kohonen for clustering data and the properties for dealing with sequentiallity of the Hidden Markov Models (HMM). We also present preliminary results.

Efficient nonlinear prediction in ADPCM

In the last years there has been a growing interest for nonlinear speech models. Several works have been published revealing the better performance of nonlinear techniques, but little attention has been dedicated to the implementation of the nonlinear model into real applications. This work is focused on the study of the behaviour of a nonlinear predictive model based on neural nets, in a speech waveform coder. Our novel scheme obtains an improvement in SEGSNR between 1 and 2 dB for an adaptive quantization ranging from 2 to 5 bits.

Feature Selection, Ranking of Each Feature and Classification for the Diagnosis of Community Acquired Legionella Pneumonia

Diagnosis of community acquired legionella pneumonia (CALP) is currently performed by means of laboratory techniques which may delay diagnosis several hours. To determine whether ANN can categorize CALP and non-legionella community-acquired pneumonia (NLCAP) and be standard for use by clinicians, we prospectively studied 203 patients with community-acquired pneumonia (CAP) diagnosed by laboratory tests. Twenty one clinical and analytical variables were recorded to train a neural net with two classes (LCAP or NLCAP class). In this paper we deal with the problem of diagnosis, feature selection, and ranking of the features as a function of their classification importance, and the design of a classifier the criteria of maximizing the ROC (Receiving operating characteristics) area, which gives a good trade-off between true positives and false negatives. In order to guarantee the validity of the statistics; the train-validation-test databases were rotated by the jackknife technique, and a multistarting procedure was done in order to make the system insensitive to local maxima.

Non Parametric Coding of Speech by Means of a MLP with Hints

This paper presents a non parametric compression system which makes use of the fact that a MLP has an internal representation of the data in the hidden layer. The system that we present makes a compression by using 4 or 8 times less units in the hidden layer than in the input. In order to improve the performance of the system we decided to use hints at the output of the system [3], these hints proved to be of use for improving the performance of the system. Several kind of hints were studied, and the results are compared with a system without hints. We also considered other aspects related with the implementation and learning in neural nets with a high number of weights

Phoneme Recognition by Means of Predictive Neural Networks

In this paper we present a phoneme recognition system based on predictive neural networks. Both feed-forward and recurrent neural networks are used for the prediction of observation vectors of speech frames. Preliminary experiments are conducted to study the discriminative quality of the prediction error as distortion measure and other similarity measures based on the Gaussian and Rayleigh distributions. The average prediction error of the neural networks is interpreted as a new feature generated by the neural net through nonlinear feature transformation. The proposed system is evaluated on a continuous speech phoneme recognition task. The recognition results that we obtain with the proposed neural network based system are compared with results obtained by a continuous density HMM system.

Analysis of Industrial Economics by means of Neural Nets

This work was has been done in order to see the feasibility of prediction of the performance of firms in industrial economics, using two different methods. One was the a multilayered perceptron and the other the a linear prediction filter. A data base of industrial corporations of Spain was used for the experiments. The predicted variables that were the benefits of the corporations and sales. The information used for the prediction was related to the internal variables of each corporation, the relative position in the sector, and additional macroeconomic data. Also some statistical tests were done in order to ascertain the reliability of the results. It was found that the results with the neural net based predictor were statistically more reliable than linear prediction in the sense that results were more accurate with a better confidence margin.

RAMSES: A Spanish Demisyllable Based Continuous Speech Recognition System

A continuous speech recognition system (called RAMSES) has been built based on the demisyllable as phonetic unit and tools from connected speech recognition. Speech is parameterized by band-pass lifted LPC-cepstra and demisyllables are represented by hidden Markov models (HMM). In this paper, the application of this system to recognize integer numbers from zero to one thousand is described. The paper contains a general overview of the system, an outline of the grammar inference, a description of the HMM training procedure and an assessment on the recognition performance in a speaker independent experiment.