carmen benitez - Academia.edu (original) (raw)
Papers by carmen benitez
Scientific Reports
The main objective of this work is to show that Shannon Entropy (SE) calculated on continuous sei... more The main objective of this work is to show that Shannon Entropy (SE) calculated on continuous seismic signals can be used in a volcanic eruption monitoring system. We analysed three years of volcanic activity of Volcán de Colima, México, recorded between January 2015 and May 2017. This period includes two large explosions, with pyroclastic and lava flows, and intense activity of less energetic explosion, culminating with a period of quiescence. In order to confirm the success of our results, we used images of the Visual Monitoring system of Colima Volcano Observatory. Another of the objectives of this work is to show how the decrease in SE values can be used to track minor explosive activity, helping Machine Learning algorithms to work more efficiently in the complex problem of distinguishing the explosion signals in the seismograms. We show that the two big eruptions selected were forecasted successfully (6 and 2 days respectively) using the decay of SE. We conclude that SE could b...
In this work we demonstrate that Shannon Entropy (SE) calculated on continuous seismic signals ca... more In this work we demonstrate that Shannon Entropy (SE) calculated on continuous seismic signals can be used efficiently in a volcanic monitoring system. We analysed three years of volcanic activity of Volcán de Colima, México, recorded between January 2015 and May 2017. This period includes two large explosions, with pyroclastic and lava flows, and intense activity of less energetic explosion, culminating with a period of quiescence. In order to confirm the success of our results, we used images of the Visual Monitoring system of Colima Volcano Observatory. Another of the objectives of this work is to show how the decrease in the SE values can be used to track minor explosive activity, helping Machine Learning algorithms to work more efficiently in the complex problem of distinguishing the explosion signals in the seismograms. We demonstrated the two big eruptions selected were forecasted successfully (6 and 2 days respectively) using the decay of the SE. We conclude that the SE coul...
In this work we demonstrate that Shannon Entropy (SE) calculated on continuous seismic signals ca... more In this work we demonstrate that Shannon Entropy (SE) calculated on continuous seismic signals can be used efficiently in a volcanic monitoring system. We analysed three years of volcanic activity of Volcán de Colima, México, recorded between January 2015 and May 2017. This period includes two large explosions, with pyroclastic and lava flows, and intense activity of less energetic explosion, culminating with a period of quiescence. In order to confirm the success of our results, we used images of the Visual Monitoring system of Colima Volcano Observatory. Another of the objectives of this work is to show how the decrease in the SE values can be used to track minor explosive activity, helping Machine Learning algorithms to work more efficiently in the complex problem of distinguishing the explosion signals in the seismograms. We demonstrated the two big eruptions selected were forecasted successfully (6 and 2 days respectively) using the decay of the SE. We conclude that the SE coul...
The search for pre-eruptive observables that can be used for short-term volcanic early warning re... more The search for pre-eruptive observables that can be used for short-term volcanic early warning remains a scientific challenge. Pre-eruptive patterns in seismic data are usually identified by analyzing seismic catalogues (e.g., the number and types of recorded seismic events), the evolution of seismic energy, or changes in the tensional state of the volcanic medium as a consequence of changes in the volume of the volcano. However, although successful volcanic predictions have been achieved, there is still no generally valid model suitable for a large range of eruptive scenarios. In this study, we evaluate the potential successful use of Shannon entropy as short-term volcanic eruption forecasting extracted from seismic signals at five well studied volcanoes (Etna, Mount St. Helens, Kilauea, Augustine, and Bezymianny). We identified temporal patterns that can be used as short-term eruptive precursors. We quantified how the Shannon entropy drops several hours before the eruptions analyz...
Annals of Geophysics, 2016
In this manuscript we present the new friendly seismic tomography software based on joint inversi... more In this manuscript we present the new friendly seismic tomography software based on joint inversion of active and passive seismic sources called PARTOS (Passive Active Ray TOmography Software). This code has been developed on the base of two well-known widely used tomographic algorithms (LOTOS and ATOM-3D), providing a robust set of algorithms. The dataset used to set and test the program has been provided by TOMO-ETNA experiment. TOMO-ETNA database is a large, high-quality dataset that includes active and passive seismic sources recorded during a period of 4 months in 2014. We performed a series of synthetic tests in order to estimate the resolution and robustness of the solutions. Real data inversion has been carried out using 3 different subsets: i) active data; ii) passive data; and iii) joint dataset. Active database is composed by a total of 16,950 air-gun shots during 1 month and passive database includes 452 local and regional earthquakes recorded during 4 months. This large...
The detection of the arrival time of seismic waves or picking is of great importance in many seis... more The detection of the arrival time of seismic waves or picking is of great importance in many seismology applications. Traditionally, picking has been carried out by human operators. This process is not systematic and relies completely on the expertise and judgment of the analysts. The limitations of manual picking and the increasing amount of data daily stored in the seismic networks worldwide distributed and in active seismic experiments lead to the development of automatic picking algorithms. Current conventional algorithms work with single signals, such as the "short-term average over long-term average" (STA/LTA) algorithm, autoregressive methods or the recently developed "Adaptive Multiband Picking Algorithm" (AMPA). This work proposes a correlation-based picking algorithm, whose main advantage is the fact of using the information of a set of signals, improving the signal to noise ratio and therefore the picking accuracy. The main advantage of this approach is that the algorithm does not require to set up sophisticated parameters, in contrast to other automatic algorithms.
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2022
Understanding how deep hierarchical models build their knowledge is a key issue in the usage of a... more Understanding how deep hierarchical models build their knowledge is a key issue in the usage of artificial intelligence to interpret the reality behind data. Depending on the discipline and models used, such knowledge may be represented in ways that are more or less intelligible for humans, limiting further improvements on the performance of the existing models. In order to delve into the characterization and modeling of volcano-seismic signals, this article emphasizes the idea of deciphering what and how recurrent neural networks (RNNs) model, and how this knowledge can be used to improve data interpretation. The key to accomplishing these objectives is both analyzing the hidden state dynamics associated with their hidden units as well as pruning/trimming based on the specialization of neurons. In this article, we process, analyze, and visualize the hidden states activation maps of two RNN architectures when managing different types of volcano-seismic events. As a result, the class-dependent discriminative behavior of most active neurons is analyzed, thereby increasing the comprehension of the detection and classification tasks. A representative dataset from the deception island volcano (Antarctica), containing volcano-tectonic earthquakes, long period events, volcanic tremors, and hybrid events, is used to train the models. Experimental analysis shows how neural activity and its associated specialization skills change depending on the architecture chosen and the type of event analyzed.
Annals of Geophysics, 2016
This work describes the automatic picking of the P-phase arrivals of the 3*106 seismic registers ... more This work describes the automatic picking of the P-phase arrivals of the 3*106 seismic registers originated during the TOMO-ETNA experiment. Air-gun shots produced by the vessel “Sarmiento de Gamboa” and contemporary passive seismicity occurring in the island are recorded by a dense network of stations deployed for the experiment. In such scenario, automatic processing is needed given: (i) the enormous amount of data, (ii) the low Signal-to-Noise ratio of many of the available registers and, (iii) the accuracy needed for the velocity tomography resulting from the experiment. A preliminary processing is performed with the records obtained from all stations. Raw data formats from the different types of stations are unified, eliminating defective records and reducing noise through filtering in the band of interest for the phase picking. The Advanced Multiband Picking Algorithm (AMPA) is then used to process the big database obtained and determine the travel times of the seismic phases....
Expert Systems with Applications, 2013
Process mining techniques have been used to analyze event logs from information systems in order ... more Process mining techniques have been used to analyze event logs from information systems in order to derive useful patterns. However, in the big data era, real-life event logs are huge, unstructured, and complex so that traditional process mining techniques have difficulties in the analysis of big logs. To reduce the complexity during the analysis, trace clustering can be used to group similar traces together and to mine more structured and simpler process models for each of the clusters locally. However, a high dimensionality of the feature space in which all the traces are presented poses different problems to trace clustering. In this paper, we study the effect of applying dimensionality reduction (preprocessing) techniques on the performance of trace clustering. In our experimental study we use three popular feature transformation techniques; singular value decomposition (SVD), random projection (RP), and principal components analysis (PCA), and the state-of-the art trace clustering in process mining. The experimental results on the dataset constructed from a real event log recorded from patient treatment processes in a Dutch hospital show that dimensionality reduction can improve trace clustering performance with respect to the computation time and average fitness of the mined local process models.
IEEE Transactions on Geoscience and Remote Sensing, 2022
2018 26th European Signal Processing Conference (EUSIPCO), 2018
Whilst recent advances in the field of artificial neural networks could be applied to monitor vol... more Whilst recent advances in the field of artificial neural networks could be applied to monitor volcanoes, its direct application remains a challenge given the complex geodynamics involved and the size of available datasets. However, Bayesian Neural Networks (BNNs) are probabilistic models that could classify and provide uncertainty estimation for transient seismic sources, even under data scarcity conditions. This research focuses on practical applications of BNNs to classify volcano-seismic signals using two variational learning approaches: Bayes by back-prop and Monte-Carlo dropout. We evaluate classification performance on seven classes of isolated events registered at “Volcán de Fuego”, Colima. Experimental results show an overall improvement for Monte-Carlo dropout approximation when compared to Bayes by backprop. Being at the intersection of Bayesian learning and geophysics, we demonstrate that BNNs provide uncertainty estimations when internal volcano-seismic sources change, w...
Computers & Geosciences, 2020
This is a PDF file of an article that has undergone enhancements after acceptance, such as the ad... more This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Electronics Letters, 2015
A prototype of phonetics learning application based on neurofeedback is presented. The learner... more A prototype of phonetics learning application based on neurofeedback is presented. The learner's auditive discrimination of chosen phonemes is evaluated using his EEG response to auditory contrasts in an oddball paradigm experiment. When auditory contrasts distinction takes place, the well-known mismatch negativity (MMN) potential registered presents a negative amplitude while zero or positive amplitudes indicate that the participant is not able to discriminate among phonemes. MMN can be processed or directly presented as user feedback, depending on the target-user profile (age, special needs, learning abilities etc.). The paradigm of application presented has many potential usages as phonetics learning tool given its capacity to automatically (i) evaluate the phonetic discrimination providing feedback, (ii) adapt to the individual needs/challenges of the user based on that feedback and (iii) keep track of the improvements achieved.
Journal of Volcanology and Geothermal Research, 2014
Automatic recognition of volcano-seismic events is becoming one of the most demanded features in ... more Automatic recognition of volcano-seismic events is becoming one of the most demanded features in the early warning area at continuous monitoring facilities. While human-driven cataloguing is time-consuming and often an unreliable task, an appropriate machine framework allows expert technicians
IEEE Transactions on Geoscience and Remote Sensing, 2019
Over the past few years, deep learning has emerged as an important tool in the fields of volcano ... more Over the past few years, deep learning has emerged as an important tool in the fields of volcano and earthquake seismology. However, these methods have been applied without performing thorough analyses of the associated uncertainties. Here we propose a solution to enhance volcano-seismic monitoring systems, through probabilistic Bayesian Deep Learning; we implement and demonstrate a workflow for waveform classification, rapid quantification of the associated uncertainty, and link these uncertainties to changes in volcanic unrest. Specifically, we introduce Bayesian Neural Networks (BNNs) to perform event identification, classification, and their estimate uncertainty on data gathered at two active volcanoes, Mount St. Helens, USA, and Bezymianny, Russia. We demonstrate how BNNs achieve excellent performance (92.08 %) in discriminating both the type of event and its origin when the two datasets are merged together and no additional training information is provided. Finally, we demonstrate that the data representations learned by the BNNs are transferable across different eruptive periods. We also find that the estimated uncertainty is related to changes in the state of unrest at the volcanoes, and propose that it could be used to gauge whether the learned models may be exported to other eruptive scenarios.
Speech Communication, 2000
Recent research in Automatic Speech Recognition (ASR) technologies has shown the keyword spotting... more Recent research in Automatic Speech Recognition (ASR) technologies has shown the keyword spotting (KWS) systems as one of the most interesting options for accessing information using speech. KWS systems can accept spontaneous speech, which allows potential users to ask for information without learning complex protocols for the human±machine communication. One of the most relevant aspects in KWS systems is the veri®cation of keyword candidates. Utterances detected as keywords could be either`false alarms' (non-keywords or incorrectly recognized keywords) or`correct key-words'. The use of con®dence measurements allows (by additional processing of the spoken sentence) the veri®cation of the candidates and the decision as to whether each utterance must be accepted as a correctly recognized keyword or rejected as a false alarm. In this work we propose a novel method for veri®cation in those KWS systems based on phone models. Under our new approach, a phonematic speech recognizer decodes the spoken sentence in parallel with the KWS recognizer. The ®rst one produces a phone string as output while the second one generates a key-word/®ller-model string. By aligning both strings, a set of characteristics is extracted which are used to verify the putatives keyword. For that we have built two classi®ers; in the ®rst one the euclidean metric is modi®ed and adapted in a local and iterative way in order to give greater importance to the most discriminate directions between the classes. The second is a vector quantizer which was trained using adaptative technique learning. We have applied the proposed method to several KWS tasks. Experimental results presented in this paper show that the proposed veri®cation method improves the performance of the KWS systems by reducing the false alarm rate without a signi®cant increase in the rejection of correctly detected keywords.
Speech Communication, 2002
The Discriminative Feature Extraction (DFE) method provides an appropriate formalism for the desi... more The Discriminative Feature Extraction (DFE) method provides an appropriate formalism for the design of the frontend feature extraction module in pattern classification systems. In the recent years, this formalism has been successfully applied to different speech recognition problems, like classification of vowels, classification of phonemes or isolated word recognition. The DFE formalism can be applied to weight the contribution of the components in the feature vector. This variant of DFE, that we call Discriminative Feature Weighting (DFW), improves the pattern classification systems by enhancing those components more relevant for the discrimination among the different classes. This paper is dedicated to the application of the DFW formalism to Continuous Speech Recognizers (CSR) based on Hidden Markov Models (HMMs). Two different types of HMM-based speech recognizers are considered: recognizers based on Discrete-HMMs (DHMMs) (for which the acoustic evaluation is based on an Euclidean distance measure) and Semi-Continuous-HMMs (SCHMMs) (for which the acoustic evaluation is performed making use of a mixture of multivariated Gaussians). We report how the components can be weighted and how the weights can be discriminatively trained and applied to the speech recognizers. We present recognition results for several continuous speech recognition tasks. The experimental results show the utility of DFW for HMM-based continuous speech recognizers.
IEEE Geoscience and Remote Sensing Letters
Scientific Reports
The main objective of this work is to show that Shannon Entropy (SE) calculated on continuous sei... more The main objective of this work is to show that Shannon Entropy (SE) calculated on continuous seismic signals can be used in a volcanic eruption monitoring system. We analysed three years of volcanic activity of Volcán de Colima, México, recorded between January 2015 and May 2017. This period includes two large explosions, with pyroclastic and lava flows, and intense activity of less energetic explosion, culminating with a period of quiescence. In order to confirm the success of our results, we used images of the Visual Monitoring system of Colima Volcano Observatory. Another of the objectives of this work is to show how the decrease in SE values can be used to track minor explosive activity, helping Machine Learning algorithms to work more efficiently in the complex problem of distinguishing the explosion signals in the seismograms. We show that the two big eruptions selected were forecasted successfully (6 and 2 days respectively) using the decay of SE. We conclude that SE could b...
In this work we demonstrate that Shannon Entropy (SE) calculated on continuous seismic signals ca... more In this work we demonstrate that Shannon Entropy (SE) calculated on continuous seismic signals can be used efficiently in a volcanic monitoring system. We analysed three years of volcanic activity of Volcán de Colima, México, recorded between January 2015 and May 2017. This period includes two large explosions, with pyroclastic and lava flows, and intense activity of less energetic explosion, culminating with a period of quiescence. In order to confirm the success of our results, we used images of the Visual Monitoring system of Colima Volcano Observatory. Another of the objectives of this work is to show how the decrease in the SE values can be used to track minor explosive activity, helping Machine Learning algorithms to work more efficiently in the complex problem of distinguishing the explosion signals in the seismograms. We demonstrated the two big eruptions selected were forecasted successfully (6 and 2 days respectively) using the decay of the SE. We conclude that the SE coul...
In this work we demonstrate that Shannon Entropy (SE) calculated on continuous seismic signals ca... more In this work we demonstrate that Shannon Entropy (SE) calculated on continuous seismic signals can be used efficiently in a volcanic monitoring system. We analysed three years of volcanic activity of Volcán de Colima, México, recorded between January 2015 and May 2017. This period includes two large explosions, with pyroclastic and lava flows, and intense activity of less energetic explosion, culminating with a period of quiescence. In order to confirm the success of our results, we used images of the Visual Monitoring system of Colima Volcano Observatory. Another of the objectives of this work is to show how the decrease in the SE values can be used to track minor explosive activity, helping Machine Learning algorithms to work more efficiently in the complex problem of distinguishing the explosion signals in the seismograms. We demonstrated the two big eruptions selected were forecasted successfully (6 and 2 days respectively) using the decay of the SE. We conclude that the SE coul...
The search for pre-eruptive observables that can be used for short-term volcanic early warning re... more The search for pre-eruptive observables that can be used for short-term volcanic early warning remains a scientific challenge. Pre-eruptive patterns in seismic data are usually identified by analyzing seismic catalogues (e.g., the number and types of recorded seismic events), the evolution of seismic energy, or changes in the tensional state of the volcanic medium as a consequence of changes in the volume of the volcano. However, although successful volcanic predictions have been achieved, there is still no generally valid model suitable for a large range of eruptive scenarios. In this study, we evaluate the potential successful use of Shannon entropy as short-term volcanic eruption forecasting extracted from seismic signals at five well studied volcanoes (Etna, Mount St. Helens, Kilauea, Augustine, and Bezymianny). We identified temporal patterns that can be used as short-term eruptive precursors. We quantified how the Shannon entropy drops several hours before the eruptions analyz...
Annals of Geophysics, 2016
In this manuscript we present the new friendly seismic tomography software based on joint inversi... more In this manuscript we present the new friendly seismic tomography software based on joint inversion of active and passive seismic sources called PARTOS (Passive Active Ray TOmography Software). This code has been developed on the base of two well-known widely used tomographic algorithms (LOTOS and ATOM-3D), providing a robust set of algorithms. The dataset used to set and test the program has been provided by TOMO-ETNA experiment. TOMO-ETNA database is a large, high-quality dataset that includes active and passive seismic sources recorded during a period of 4 months in 2014. We performed a series of synthetic tests in order to estimate the resolution and robustness of the solutions. Real data inversion has been carried out using 3 different subsets: i) active data; ii) passive data; and iii) joint dataset. Active database is composed by a total of 16,950 air-gun shots during 1 month and passive database includes 452 local and regional earthquakes recorded during 4 months. This large...
The detection of the arrival time of seismic waves or picking is of great importance in many seis... more The detection of the arrival time of seismic waves or picking is of great importance in many seismology applications. Traditionally, picking has been carried out by human operators. This process is not systematic and relies completely on the expertise and judgment of the analysts. The limitations of manual picking and the increasing amount of data daily stored in the seismic networks worldwide distributed and in active seismic experiments lead to the development of automatic picking algorithms. Current conventional algorithms work with single signals, such as the "short-term average over long-term average" (STA/LTA) algorithm, autoregressive methods or the recently developed "Adaptive Multiband Picking Algorithm" (AMPA). This work proposes a correlation-based picking algorithm, whose main advantage is the fact of using the information of a set of signals, improving the signal to noise ratio and therefore the picking accuracy. The main advantage of this approach is that the algorithm does not require to set up sophisticated parameters, in contrast to other automatic algorithms.
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2022
Understanding how deep hierarchical models build their knowledge is a key issue in the usage of a... more Understanding how deep hierarchical models build their knowledge is a key issue in the usage of artificial intelligence to interpret the reality behind data. Depending on the discipline and models used, such knowledge may be represented in ways that are more or less intelligible for humans, limiting further improvements on the performance of the existing models. In order to delve into the characterization and modeling of volcano-seismic signals, this article emphasizes the idea of deciphering what and how recurrent neural networks (RNNs) model, and how this knowledge can be used to improve data interpretation. The key to accomplishing these objectives is both analyzing the hidden state dynamics associated with their hidden units as well as pruning/trimming based on the specialization of neurons. In this article, we process, analyze, and visualize the hidden states activation maps of two RNN architectures when managing different types of volcano-seismic events. As a result, the class-dependent discriminative behavior of most active neurons is analyzed, thereby increasing the comprehension of the detection and classification tasks. A representative dataset from the deception island volcano (Antarctica), containing volcano-tectonic earthquakes, long period events, volcanic tremors, and hybrid events, is used to train the models. Experimental analysis shows how neural activity and its associated specialization skills change depending on the architecture chosen and the type of event analyzed.
Annals of Geophysics, 2016
This work describes the automatic picking of the P-phase arrivals of the 3*106 seismic registers ... more This work describes the automatic picking of the P-phase arrivals of the 3*106 seismic registers originated during the TOMO-ETNA experiment. Air-gun shots produced by the vessel “Sarmiento de Gamboa” and contemporary passive seismicity occurring in the island are recorded by a dense network of stations deployed for the experiment. In such scenario, automatic processing is needed given: (i) the enormous amount of data, (ii) the low Signal-to-Noise ratio of many of the available registers and, (iii) the accuracy needed for the velocity tomography resulting from the experiment. A preliminary processing is performed with the records obtained from all stations. Raw data formats from the different types of stations are unified, eliminating defective records and reducing noise through filtering in the band of interest for the phase picking. The Advanced Multiband Picking Algorithm (AMPA) is then used to process the big database obtained and determine the travel times of the seismic phases....
Expert Systems with Applications, 2013
Process mining techniques have been used to analyze event logs from information systems in order ... more Process mining techniques have been used to analyze event logs from information systems in order to derive useful patterns. However, in the big data era, real-life event logs are huge, unstructured, and complex so that traditional process mining techniques have difficulties in the analysis of big logs. To reduce the complexity during the analysis, trace clustering can be used to group similar traces together and to mine more structured and simpler process models for each of the clusters locally. However, a high dimensionality of the feature space in which all the traces are presented poses different problems to trace clustering. In this paper, we study the effect of applying dimensionality reduction (preprocessing) techniques on the performance of trace clustering. In our experimental study we use three popular feature transformation techniques; singular value decomposition (SVD), random projection (RP), and principal components analysis (PCA), and the state-of-the art trace clustering in process mining. The experimental results on the dataset constructed from a real event log recorded from patient treatment processes in a Dutch hospital show that dimensionality reduction can improve trace clustering performance with respect to the computation time and average fitness of the mined local process models.
IEEE Transactions on Geoscience and Remote Sensing, 2022
2018 26th European Signal Processing Conference (EUSIPCO), 2018
Whilst recent advances in the field of artificial neural networks could be applied to monitor vol... more Whilst recent advances in the field of artificial neural networks could be applied to monitor volcanoes, its direct application remains a challenge given the complex geodynamics involved and the size of available datasets. However, Bayesian Neural Networks (BNNs) are probabilistic models that could classify and provide uncertainty estimation for transient seismic sources, even under data scarcity conditions. This research focuses on practical applications of BNNs to classify volcano-seismic signals using two variational learning approaches: Bayes by back-prop and Monte-Carlo dropout. We evaluate classification performance on seven classes of isolated events registered at “Volcán de Fuego”, Colima. Experimental results show an overall improvement for Monte-Carlo dropout approximation when compared to Bayes by backprop. Being at the intersection of Bayesian learning and geophysics, we demonstrate that BNNs provide uncertainty estimations when internal volcano-seismic sources change, w...
Computers & Geosciences, 2020
This is a PDF file of an article that has undergone enhancements after acceptance, such as the ad... more This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Electronics Letters, 2015
A prototype of phonetics learning application based on neurofeedback is presented. The learner... more A prototype of phonetics learning application based on neurofeedback is presented. The learner's auditive discrimination of chosen phonemes is evaluated using his EEG response to auditory contrasts in an oddball paradigm experiment. When auditory contrasts distinction takes place, the well-known mismatch negativity (MMN) potential registered presents a negative amplitude while zero or positive amplitudes indicate that the participant is not able to discriminate among phonemes. MMN can be processed or directly presented as user feedback, depending on the target-user profile (age, special needs, learning abilities etc.). The paradigm of application presented has many potential usages as phonetics learning tool given its capacity to automatically (i) evaluate the phonetic discrimination providing feedback, (ii) adapt to the individual needs/challenges of the user based on that feedback and (iii) keep track of the improvements achieved.
Journal of Volcanology and Geothermal Research, 2014
Automatic recognition of volcano-seismic events is becoming one of the most demanded features in ... more Automatic recognition of volcano-seismic events is becoming one of the most demanded features in the early warning area at continuous monitoring facilities. While human-driven cataloguing is time-consuming and often an unreliable task, an appropriate machine framework allows expert technicians
IEEE Transactions on Geoscience and Remote Sensing, 2019
Over the past few years, deep learning has emerged as an important tool in the fields of volcano ... more Over the past few years, deep learning has emerged as an important tool in the fields of volcano and earthquake seismology. However, these methods have been applied without performing thorough analyses of the associated uncertainties. Here we propose a solution to enhance volcano-seismic monitoring systems, through probabilistic Bayesian Deep Learning; we implement and demonstrate a workflow for waveform classification, rapid quantification of the associated uncertainty, and link these uncertainties to changes in volcanic unrest. Specifically, we introduce Bayesian Neural Networks (BNNs) to perform event identification, classification, and their estimate uncertainty on data gathered at two active volcanoes, Mount St. Helens, USA, and Bezymianny, Russia. We demonstrate how BNNs achieve excellent performance (92.08 %) in discriminating both the type of event and its origin when the two datasets are merged together and no additional training information is provided. Finally, we demonstrate that the data representations learned by the BNNs are transferable across different eruptive periods. We also find that the estimated uncertainty is related to changes in the state of unrest at the volcanoes, and propose that it could be used to gauge whether the learned models may be exported to other eruptive scenarios.
Speech Communication, 2000
Recent research in Automatic Speech Recognition (ASR) technologies has shown the keyword spotting... more Recent research in Automatic Speech Recognition (ASR) technologies has shown the keyword spotting (KWS) systems as one of the most interesting options for accessing information using speech. KWS systems can accept spontaneous speech, which allows potential users to ask for information without learning complex protocols for the human±machine communication. One of the most relevant aspects in KWS systems is the veri®cation of keyword candidates. Utterances detected as keywords could be either`false alarms' (non-keywords or incorrectly recognized keywords) or`correct key-words'. The use of con®dence measurements allows (by additional processing of the spoken sentence) the veri®cation of the candidates and the decision as to whether each utterance must be accepted as a correctly recognized keyword or rejected as a false alarm. In this work we propose a novel method for veri®cation in those KWS systems based on phone models. Under our new approach, a phonematic speech recognizer decodes the spoken sentence in parallel with the KWS recognizer. The ®rst one produces a phone string as output while the second one generates a key-word/®ller-model string. By aligning both strings, a set of characteristics is extracted which are used to verify the putatives keyword. For that we have built two classi®ers; in the ®rst one the euclidean metric is modi®ed and adapted in a local and iterative way in order to give greater importance to the most discriminate directions between the classes. The second is a vector quantizer which was trained using adaptative technique learning. We have applied the proposed method to several KWS tasks. Experimental results presented in this paper show that the proposed veri®cation method improves the performance of the KWS systems by reducing the false alarm rate without a signi®cant increase in the rejection of correctly detected keywords.
Speech Communication, 2002
The Discriminative Feature Extraction (DFE) method provides an appropriate formalism for the desi... more The Discriminative Feature Extraction (DFE) method provides an appropriate formalism for the design of the frontend feature extraction module in pattern classification systems. In the recent years, this formalism has been successfully applied to different speech recognition problems, like classification of vowels, classification of phonemes or isolated word recognition. The DFE formalism can be applied to weight the contribution of the components in the feature vector. This variant of DFE, that we call Discriminative Feature Weighting (DFW), improves the pattern classification systems by enhancing those components more relevant for the discrimination among the different classes. This paper is dedicated to the application of the DFW formalism to Continuous Speech Recognizers (CSR) based on Hidden Markov Models (HMMs). Two different types of HMM-based speech recognizers are considered: recognizers based on Discrete-HMMs (DHMMs) (for which the acoustic evaluation is based on an Euclidean distance measure) and Semi-Continuous-HMMs (SCHMMs) (for which the acoustic evaluation is performed making use of a mixture of multivariated Gaussians). We report how the components can be weighted and how the weights can be discriminatively trained and applied to the speech recognizers. We present recognition results for several continuous speech recognition tasks. The experimental results show the utility of DFW for HMM-based continuous speech recognizers.
IEEE Geoscience and Remote Sensing Letters