K. López de Ipiña

Selection of Lexical Units for Continuous Speech Recognition of basque

The selection of appropriate Lexical Units (LUs) is an important issue in the development of Continuous Speech Recognition (CSR) systems. Words have been used classically as the recognition unit in most of them. However, proposals of non-word units are beginning to arise. Basque is an agglutinative language with some structure inside words, for which non-word morpheme like units could be an appropriate choice. In this work a statistical analysis of units obtained after morphological segmentation has been carried out. This analysis shows a potential gain of confusion rates in CSR systems, due to the growth of the set of acoustically similar and short morphemes. Thus, several proposals of Lexical Units are analysed to deal with the problem. Measures of Phonetic Perplexity and Speech Recognition rates have been computed using different sets of units and, based on these measures, a set of alternative non-word units have been selected.

Using non-word lexical units in automatic speech understanding

If the objective of a continuous automatic speech understanding system is not a speech-to-text translation, words are not strictly needed, and then the use of alternative lexical units (LUs) will bring us a new degree of freedom to improve the system performance. Consequently, we experimentally explore some methods to automatically extract a set of LUs from a Spanish training corpus and verify that the system can be improved in two ways: reducing the computational costs and increasing the recognition rates. Moreover, preliminary results point out that, even if the system target is a speech-to-text translation, using non-word units and post-processing the output to produce the corresponding word chain outperforms the word based system.

Selection of entropy based features for the analysis of the Archimedes' spiral applied to essential tremor

Biomedical systems are regulated by interacting mechanisms that operate across multiple spatial and temporal scales and produce biosignals with linear and non-linear information inside. In this sense entropy could provide a useful measure about disorder in the system, lack of information in time-series and/or irregularity of the signals. Essential tremor (ET) is the most common movement disorder, being 20 times more common than Parkinsons disease, and 50-70% of this disease cases are estimated to be genetic in origin. Archimedes spiral drawing is one of the most used standard tests for clinical diagnosis. This work, on selection of nonlinear biomarkers from drawings and handwriting, is part of a wide-ranging cross study for the diagnosis of essential tremor in BioDonostia Health Institute. Several entropy algorithms are used to generate nonlinear feayures. The automatic analysis system consists of several Machine Learning paradigms.

A Computer-Aided Decision Support System for Shoulder Pain Pathology

A musculoskeletal disorder is a condition of the musculoskeletal system which consists in that part of it is injured continuously over time. Shoulder disorders are one of the most common musculoskeletal cases attended in primary health care services. Shoulder disorders cause pain and limit the ability to perform many routine activities and affect about 15-25 % of the general population. Several clinical tests have been described to aid diagnosis of shoulder disorders. However, the current literature acknowledges a lack of concordance in clinical assessment, even among musculoskeletal specialists. In this work a Computer-Aided Decision Support (CADS) system for Shoulder Pain Pathology has been developed. The paper presents the medical method and the development of the database and the (CADS) system based on several classical classification paradigms improve by covariance estimation methods. Finally the system was evaluated by a medical specialist.

Hybrid approach for language identification oriented to multilingual speech recognition in the basque context

The development of Multilingual Large Vocabulary Continuous Speech Recognition systems involves issues as: Language Identification, Acoustic-Phonetic Decoding, Language Modelling or the development of appropriated Language Resources The interest on Multilingual Systems arouses because there are three official languages in the Basque Country (Basque, Spanish, and French), and there is much linguistic interaction among them, even if Basque has very different roots than the other two languages This paper describes the development of a Language Identification (LID) system oriented to robust Multilingual Speech Recognition for the Basque context The work presents hybrid strategies for LID, based on the selection of system elements by Support Vector Machines and Multilayer Perceptron classifiers and stochastic methods for speech recognition tasks (Hidden Markov Models and n-grams).

Morphological segmentation for speech processing in Basque

Morphological information is traditionally used to develop high quality text to speech (TTS) and automatic speech recognition (ASR) systems. The use of this information improves the naturalness and intelligibility of the TTS synthesis and provides an appropriated way to select lexical units (LU) for ASR. Basque is an agglutinative language with a complex structure inside the words and the morphological information is essential both in TTS and ASR. In this work an automatic morphological segmentation tool oriented to TTS and ASR tasks is presented.

First approach to the analysis of spontaneous activity of mice based on Permutation Entropy

Animal behavior is usually assessed by categorical classification. Here we used an alternative approach, based on the global quantification of video and pressure sensor derived signals, in order to characterize normal and pathological mouse behavior. Freely moving mice were recorded for 1h in a novel open field environment. In this preliminary study we have tested the use of permutation entropy applied to spatial information (Cartesian and polar coordinates of instantaneous position, provided by automatic video-tracking analysis) and movement-derived signal (ie total animal movement detected using piezoelectric pressure-sensors). We report that this approach could discriminate between wild type and transgenic Fmr1-KO mice, a model of neurodevelopmental disorder affecting cognitive functions and behavior in open field conditions.

Decision tree-based context dependent sublexical units for Continuous Speech Recognition of basque

This paper presents a new methodology, based on the classical decision trees, to get a suitable set of context dependent sublexical units for Basque Continuous Speech Recognition (CSR). The original method proposed by Bahl [1] was applied as the benchmark. Then two new features were added: a data massaging to emphasise the data and a fast and efficient Growing and Pruning algorithm for DT construction. In addition, the use of the new context dependent units to build word models was addressed. The benchmark Bahl approach gave recognition rates clearly outperforming those of context independent phone-like units. Finally the new methodology improves over the benchmark DT approach.

Language identification for Internet security in the Basque context

The present work describes the development of an LID system suited for handling security tasks in the Internet. The development context was the Infozazpi Internet digital radio, and the task presented substantial complexity due to the trilingual environment and the scarcity of language resources for Basque. In order to overcome previous difficulties, we propose a hybrid system based on the selection of subword units by SVMs, MLP classifiers, and discriminant analysis improved with robust regularized covariance matrix estimation methods and stochastic methods for ASR tasks (SC-HMM and n-grams). Our new subword unit proposals and the use of triphones and cross-lingual approaches considerably improve the system performance, achieving an optimal and stable LID recognition rate despite the complexity of the problem.

Matrix Covariance Estimation methods for robust Security Speech Recognition with under-resourced conditions

The long term goal of our project is the development of robust Security Speech Recognition systems are based on Automatic Speech Recognition methodologies. The development of ASR systems involves dealing with issues such as Acoustic Phonetic Decoding (APD), Language Modelling (LM) or the development of appropriated Language Resources (LR). Thus these applications are generally very language-dependent and require very large resources. This work is focused to the selection of appropriated sub-word units with under-resourced and noisy conditions oriented to security tasks. The work has been carried out with a trilingual internet radio database. Thus, in order to decrease the negative impact that the lack of resources has in this issue we apply several data optimization methodologies based on Matrix Covariance Estimation and Ontology-based approaches.