Jesus Lopez-Ballester
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Jesus Lopez-Ballester has a total of 27 co-authors. Here are the top co-authors by number of shared publications:
- Jaume Segura-Garcia
Department of Computer Sciences, ETSE, University of Valencia, Burjassot, 46100, SPAIN
9 shared publications - Santiago Felici-Castell
University of Valencia
8 shared publications - Maximo Cobos
Departament d’Informàtica, Universitat de València, Burjassot, Spain
7 shared publications - Franscesc J. Ferri
Departament d’Informàtica, Universitat de València, Burjassot, Spain
4 shared publications - Juan Jose Perez Solano
Departament d’Informàtica, Escola Tècnica Superior d’Enginyeria (ETSE), Universitat de València, 46100 Burjassot, Spain
4 shared publications - Carlos Castorena
Universitat de València - Department of Informatics
3 shared publications
Latest Publications
- Open Access
- 15 December 2025
- Published by Springer Nature in EURASIP Journal on Audio, Speech, and Music Processing
- Vol. 2025 (1)
A data-driven exploration of elevation cues in HRTFs: an explainable AI perspective across multiple datasets
Precise elevation perception in binaural audio remains a challenge, despite extensive research on head-related transfer functions (HRTFs) and spectral cues. While prior studies have advanced our understanding of sound localization cues, the interplay between spectral features and elevation perception is still not fully understood. This paper presents a comprehensive analysis of over 600 subjects from 11 diverse public HRTF datasets, employing a convolutional neural network (CNN) model combined with explainable artificial intelligence (XAI) techniques to investigate elevation cues. In addition to testing various HRTF pre-processing methods, we focus on both within-dataset and inter-dataset generalization and explainability, assessing the model’s robustness across different HRTF variations stemming from subjects and measurement setups. By leveraging class activation mapping (CAM) saliency maps, we identify key frequency bands that may contribute to elevation perception, providing deeper insights into the spectral features that drive elevation-specific classification. This study offers new perspectives on HRTF modeling and elevation perception by analyzing diverse datasets and pre-processing techniques, expanding our understanding of these cues across a wide range of conditions.
- 15 December 2025
- Published by Springer Nature in EURASIP Journal on Audio, Speech, and Music Processing
- Vol. 2025 (1)
Open AccessA data-driven exploration of elevation cues in HRTFs: an explainable AI perspective across multiple datasets
Precise elevation perception in binaural audio remains a challenge, despite extensive research on head-related transfer functions (HRTFs) and spectral cues. While prior studies have advanced our understanding of sound localization cues, the interplay between spectral features and elevation perception is still not fully understood. This paper presents a comprehensive analysis of over 600 subjects from 11 diverse public HRTF datasets, employing a convolutional neural network (CNN) model combined with explainable artificial intelligence (XAI) techniques to investigate elevation cues. In addition to testing various HRTF pre-processing methods, we focus on both within-dataset and inter-dataset generalization and explainability, assessing the model’s robustness across different HRTF variations stemming from subjects and measurement setups. By leveraging class activation mapping (CAM) saliency maps, we identify key frequency bands that may contribute to elevation perception, providing deeper insights into the spectral features that drive elevation-specific classification. This study offers new perspectives on HRTF modeling and elevation perception by analyzing diverse datasets and pre-processing techniques, expanding our understanding of these cues across a wide range of conditions.
- 8 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1248-1252
On the Perceptual Validation of Explainable AI-Based HRTF Saliency for Vertical Sound Localization
Head-Related Transfer Functions (HRTFs) play a key role in spatial audio, particularly for vertical sound localization, where interaural time and level differences are negligible. While spectral cues essential for elevation perception have been extensively studied, their precise contributions remain elusive. In this work, we investigate the saliency of spectral cues by leveraging Explainable Artificial Intelligence (XAI) techniques applied to a deep learning model trained for HRTF-based elevation classification. Using saliency maps, we identify and systematically ablate key frequency bands in perceptual experiments to assess their role in human sound localization along the median plane. Our preliminary results suggest that removing high-saliency frequency bands degrades localization accuracy, supporting the model's predictions. However, the degree of impairment varies across conditions, indicating a complex interplay between spectral cues and perceptual processing. These findings highlight the potential of XAI for interpreting spatial hearing models and motivate further investigation into the perceptual significance of HRTF saliency.
- 8 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1248-1252
On the Perceptual Validation of Explainable AI-Based HRTF Saliency for Vertical Sound Localization
Head-Related Transfer Functions (HRTFs) play a key role in spatial audio, particularly for vertical sound localization, where interaural time and level differences are negligible. While spectral cues essential for elevation perception have been extensively studied, their precise contributions remain elusive. In this work, we investigate the saliency of spectral cues by leveraging Explainable Artificial Intelligence (XAI) techniques applied to a deep learning model trained for HRTF-based elevation classification. Using saliency maps, we identify and systematically ablate key frequency bands in perceptual experiments to assess their role in human sound localization along the median plane. Our preliminary results suggest that removing high-saliency frequency bands degrades localization accuracy, supporting the model's predictions. However, the degree of impairment varies across conditions, indicating a complex interplay between spectral cues and perceptual processing. These findings highlight the potential of XAI for interpreting spatial hearing models and motivate further investigation into the perceptual significance of HRTF saliency.
- Open Access
- 29 January 2025
- Published by Dirección de Bienes Públicos del Conocimiento, Universidad Autonoma del Estado de Mexico in Ideas en Ciencias de la Ingeniería
- Vol. 3 (1) , 48
Exploring Selective Sampling Bounds for Speaker-Dependent Speech Emotion Recognition
Emotion recognition in speech is crucial for various applications, ranging from mental health to human-machine interaction. Although deep learning-based models have proven effective in these tasks, their performance can be affected when faced with new individuals, even if they were trained with large and diverse datasets. In this study, we analyze the importance of fine-tuning and personalizing these models to improve emotional recognition accuracy. Despite the good initial results of general models, we observe a decline in their predictive ability when applied to new subjects. Through this work, we demonstrate how personalizing a model with a small number of samples from a new individual can significantly improve its performance, using techniques such as fine-tuning and transfer learning. Finally, we discuss the implications of these findings in real-world applications and suggest potential directions for future research in emotion detection through artificial intelligence.
- 29 January 2025
- Published by Dirección de Bienes Públicos del Conocimiento, Universidad Autonoma del Estado de Mexico in Ideas en Ciencias de la Ingeniería
- Vol. 3 (1) , 48
Open AccessExploring Selective Sampling Bounds for Speaker-Dependent Speech Emotion Recognition
Emotion recognition in speech is crucial for various applications, ranging from mental health to human-machine interaction. Although deep learning-based models have proven effective in these tasks, their performance can be affected when faced with new individuals, even if they were trained with large and diverse datasets. In this study, we analyze the importance of fine-tuning and personalizing these models to improve emotional recognition accuracy. Despite the good initial results of general models, we observe a decline in their predictive ability when applied to new subjects. Through this work, we demonstrate how personalizing a model with a small number of samples from a new individual can significantly improve its performance, using techniques such as fine-tuning and transfer learning. Finally, we discuss the implications of these findings in real-world applications and suggest potential directions for future research in emotion detection through artificial intelligence.
- 13 December 2024
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 81 (1)
Edge computing for driving safety: evaluating deep learning models for cost-effective sound event detection
- 13 December 2024
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 81 (1)
Edge computing for driving safety: evaluating deep learning models for cost-effective sound event detection
- Open Access
- 3 July 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-4
Evaluation of Real-Time Acoustic Event Detection Models in Driving Scenarios
- 3 July 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-4
Open AccessEvaluation of Real-Time Acoustic Event Detection Models in Driving Scenarios
- Open Access
- 2 September 2023
- Published by MDPI AG in Applied Sciences
- Vol. 13 (17) , 9930
Accurate Estimation of Air Pollution in Outdoor Routes for Citizens and Decision Making
There is clear evidence of the effects of air pollution on health. In this paper, we present an innovative application designed to assess Air Quality (AQ) exposure based on the World Health Organization’s AQ Guidelines, analysing pollutants and their concentrations independently. Our aim is to provide this information to citizens based on their health profile (medical history or requirements) before and during outdoor trips of their choice, both walking and cycling, empowering them to proactively make informed personal decisions about their route choices and identifying potentially unhealthy travel environments. For this purpose, we have access to official data from AQ monitoring stations that are updated periodically every 10 min. Then, by using spatial interpolation techniques (with Ordinary Kriging), we estimate each pollutant over a grid superimposed on the city map. Once the pollutants have been mapped on each route, they are analysed in order to consider the different alternatives for deciding and planning changes in speed or trajectory. We evaluated the application in the city of Valencia (Spain) as a use case under different scenarios, and showed the results to assess exposure to pollution on the routes of citizens.
- 2 September 2023
- Published by MDPI AG in Applied Sciences
- Vol. 13 (17) , 9930
Open AccessAccurate Estimation of Air Pollution in Outdoor Routes for Citizens and Decision Making
There is clear evidence of the effects of air pollution on health. In this paper, we present an innovative application designed to assess Air Quality (AQ) exposure based on the World Health Organization’s AQ Guidelines, analysing pollutants and their concentrations independently. Our aim is to provide this information to citizens based on their health profile (medical history or requirements) before and during outdoor trips of their choice, both walking and cycling, empowering them to proactively make informed personal decisions about their route choices and identifying potentially unhealthy travel environments. For this purpose, we have access to official data from AQ monitoring stations that are updated periodically every 10 min. Then, by using spatial interpolation techniques (with Ordinary Kriging), we estimate each pollutant over a grid superimposed on the city map. Once the pollutants have been mapped on each route, they are analysed in order to consider the different alternatives for deciding and planning changes in speed or trajectory. We evaluated the application in the city of Valencia (Spain) as a use case under different scenarios, and showed the results to assess exposure to pollution on the routes of citizens.
- Open Access
- 24 June 2023
- Published by MDPI AG in Sustainability
- Vol. 15 (13) , 10016
Sustainable Soundscape Monitoring of Modified Psycho-Acoustic Annoyance Model with Edge Computing for 5G IoT Systems
Next-generation IoT systems will allow sustainable performance in long-term monitoring systems. This sustainability concept applies to soundscape description, as it allows monitoring in urban environments. In this work, the implementation of psycho-acoustic annoyance models in a 5G-enabled IoT system is proposed, applying two edge-computing approaches. A modified Zwicker’s model is adopted in this research, introducing a term that takes into account the tonal component of the captured sound. These implementations have been validated in a measurement campaign where several IoT devices have been deployed to evaluate different sound environments of a university campus. Then, the analysis of the sound-quality metrics is conducted in a different location, showing that if tonality is present in a noisy environment, it results in greater subjective annoyance. Moreover, the Just-Noticeable Difference of these results is derived from Zwicker’s psycho-acoustic annoyance to establish a limitation for this metric.
- 24 June 2023
- Published by MDPI AG in Sustainability
- Vol. 15 (13) , 10016
Open AccessSustainable Soundscape Monitoring of Modified Psycho-Acoustic Annoyance Model with Edge Computing for 5G IoT Systems
Next-generation IoT systems will allow sustainable performance in long-term monitoring systems. This sustainability concept applies to soundscape description, as it allows monitoring in urban environments. In this work, the implementation of psycho-acoustic annoyance models in a 5G-enabled IoT system is proposed, applying two edge-computing approaches. A modified Zwicker’s model is adopted in this research, introducing a term that takes into account the tonal component of the captured sound. These implementations have been validated in a measurement campaign where several IoT devices have been deployed to evaluate different sound environments of a university campus. Then, the analysis of the sound-quality metrics is conducted in a different location, showing that if tonality is present in a noisy environment, it results in greater subjective annoyance. Moreover, the Just-Noticeable Difference of these results is derived from Zwicker’s psycho-acoustic annoyance to establish a limitation for this metric.
- Open Access
- 30 April 2023
- Published by MDPI AG in Preprints.org
Operation of the Worldwide Lightning Location Network over Spain
This study analyses the detection efficiency of the Worldwide Lightning Location Network (WWLLN) of sensors over Spain, a comparison is made with the Meteorological Spanish Agency (AEMET) Lightning Detection Network. The WWLLN operates a planetary distributed network of Very Low Frequency (VLF) antennas which detect lightning VLF signals around the Earth. Very high currents from lightning strokes radiate strong VLF signals in the band 6-22 kHz, which are detected up to 10,000 km, therefore WWLLN receiving sensors can be sparsely spaced. However the detection capability of VLF signals depends on the environment, and clean areas with a good ground and no metal parts around must be selected for proper reception. For the area of Spain and Portugal we had two WWLLN sensors in 2012, and three sensors later on for the data of 2020, 2021 and 2022. The sensors in Iberian Peninsula are at a distance of around 760 km; that is very close in comparison with other WWLLN sensors in the Earth which are around 5.000-15.000 km. The WWLLN sensors send the timing of the detection to the central sensor where the location of the lightning is determined with an improved algorithm when a mini-mum of 5 sensors detect the VLF signal. We show that detection efficiency was excellent and consistent with previous studies. Lightning location data from 1st January 2012 to 30st April 2012 from AEMET was used to analyze time and space shared events with the WWLLN network with 0.5s time difference and 20 km distance. The results improved in 2020, 2021 and 2022. More than 40,000 lightning strikes were analyzed in the present work, and we found a detection efficiency > 38%.
- 30 April 2023
- Published by MDPI AG in Preprints.org
Open AccessOperation of the Worldwide Lightning Location Network over Spain
This study analyses the detection efficiency of the Worldwide Lightning Location Network (WWLLN) of sensors over Spain, a comparison is made with the Meteorological Spanish Agency (AEMET) Lightning Detection Network. The WWLLN operates a planetary distributed network of Very Low Frequency (VLF) antennas which detect lightning VLF signals around the Earth. Very high currents from lightning strokes radiate strong VLF signals in the band 6-22 kHz, which are detected up to 10,000 km, therefore WWLLN receiving sensors can be sparsely spaced. However the detection capability of VLF signals depends on the environment, and clean areas with a good ground and no metal parts around must be selected for proper reception. For the area of Spain and Portugal we had two WWLLN sensors in 2012, and three sensors later on for the data of 2020, 2021 and 2022. The sensors in Iberian Peninsula are at a distance of around 760 km; that is very close in comparison with other WWLLN sensors in the Earth which are around 5.000-15.000 km. The WWLLN sensors send the timing of the detection to the central sensor where the location of the lightning is determined with an improved algorithm when a mini-mum of 5 sensors detect the VLF signal. We show that detection efficiency was excellent and consistent with previous studies. Lightning location data from 1st January 2012 to 30st April 2012 from AEMET was used to analyze time and space shared events with the WWLLN network with 0.5s time difference and 20 km distance. The results improved in 2020, 2021 and 2022. More than 40,000 lightning strikes were analyzed in the present work, and we found a detection efficiency > 38%.
- Open Access
- 23 April 2023
- Published by MDPI AG in Preprints.org
Sustainable Soundscape Monitoring of Modified Psycho-Acoustic Annoyance Model With Edge Computing for 5G IoT Systems
Next Generation IoT systems will allow sustainable performance in long-term monitoring systems. This sustainability concept is applicable to soundscape description, as it allows monitoring in urban environments. In this work, the implementation of psycho-acoustic annoyance models in a 5G-enabled IoT system is proposed applying two Edge Computing approaches. A modified Zwicker’s model is adopted in this research, introducing a term that takes into account the tonal component of the captured sound. These implementations have been validated in a measurement campaign where several IoT devices have been deployed to evaluate different sound environments of a University Campus. Then, the analysis of the sound quality metrics is done in different location showing that if tonality is present in a noisy environment, it results in greater subjective annoyance. Moreover, the Just Noticeable Difference of these results are derived for the Zwicker’s psycho-acoustic annoyance to establish a limitation for this metric.
- 23 April 2023
- Published by MDPI AG in Preprints.org
Open AccessSustainable Soundscape Monitoring of Modified Psycho-Acoustic Annoyance Model With Edge Computing for 5G IoT Systems
Next Generation IoT systems will allow sustainable performance in long-term monitoring systems. This sustainability concept is applicable to soundscape description, as it allows monitoring in urban environments. In this work, the implementation of psycho-acoustic annoyance models in a 5G-enabled IoT system is proposed applying two Edge Computing approaches. A modified Zwicker’s model is adopted in this research, introducing a term that takes into account the tonal component of the captured sound. These implementations have been validated in a measurement campaign where several IoT devices have been deployed to evaluate different sound environments of a University Campus. Then, the analysis of the sound quality metrics is done in different location showing that if tonality is present in a noisy environment, it results in greater subjective annoyance. Moreover, the Just Noticeable Difference of these results are derived for the Zwicker’s psycho-acoustic annoyance to establish a limitation for this metric.