Petros Maragos
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Petros Maragos has a total of 187 co-authors. Here are the top co-authors by number of shared publications:
- Panagiotis Paraskevas Filntisis
School of Electrical and Computer Engineering, National Technical University of Athens, Athens, Greece
17 shared publications - George Retsinas
National Technical University of Athens
14 shared publications - Niki Efthymiou
National Technical University of Athens
13 shared publications - Athanasia Zlatintsi
National Technical University of Athens
9 shared publications - Petros Koutras
School of E.C.E National Technical University of Athens Greece
9 shared publications - Christos Garoufis
School of ECE National Technical University of Athens 15773 Athens Greece
8 shared publications
Latest Publications
- Open Access
- 30 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Audio, Speech and Language Processing
- Vol. 34, 2724-2737
CVSM: Contrastive Vocal Similarity Modeling
The availability of large, unlabeled datasets across various domains has contributed to the development of a plethora of methods that learn representations for multiple target (downstream) tasks through self-supervised pre-training. In this work, we introduce CVSM (Contrastive Vocal Similarity Modeling), a contrastive self-supervised procedure for music signal representation learning in the audio domain that can be utilized for musical and vocal similarity modeling. Our method operates under a contrastive framework, maximizing the similarity between vocal excerpts and musical mixtures containing the same vocals; we devise both a label-informed protocol, leveraging artist identity information to sample the contrastive pairs, and a label-agnostic scheme, involving artificial mixture creation from randomly sampled vocal and accompaniment excerpts, which are paired with vocals from the same audio segment. We evaluate our proposed method in measuring vocal similarity both objectively, through linear probing on a suite of appropriate downstream tasks, and subjectively, via conducting a user study consisting of pairwise comparisons between different models in a recommendation-by-query setting. Our results indicate that the representations learned through CVSM are effective in musical and vocal similarity modeling, outperforming numerous baselines across both isolated vocals and complete musical mixtures. Moreover, while the availability of artist identity labels during pre-training leads to overall more consistent performance both in the evaluated downstream tasks and the user study, a label-agnostic CVSM variant incorporating hybrid pre-training with real and artificial mixtures achieves comparable performance to the label-informed one in artist identification and perceived vocal similarity.
- 30 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Audio, Speech and Language Processing
- Vol. 34, 2724-2737
Open AccessCVSM: Contrastive Vocal Similarity Modeling
The availability of large, unlabeled datasets across various domains has contributed to the development of a plethora of methods that learn representations for multiple target (downstream) tasks through self-supervised pre-training. In this work, we introduce CVSM (Contrastive Vocal Similarity Modeling), a contrastive self-supervised procedure for music signal representation learning in the audio domain that can be utilized for musical and vocal similarity modeling. Our method operates under a contrastive framework, maximizing the similarity between vocal excerpts and musical mixtures containing the same vocals; we devise both a label-informed protocol, leveraging artist identity information to sample the contrastive pairs, and a label-agnostic scheme, involving artificial mixture creation from randomly sampled vocal and accompaniment excerpts, which are paired with vocals from the same audio segment. We evaluate our proposed method in measuring vocal similarity both objectively, through linear probing on a suite of appropriate downstream tasks, and subjectively, via conducting a user study consisting of pairwise comparisons between different models in a recommendation-by-query setting. Our results indicate that the representations learned through CVSM are effective in musical and vocal similarity modeling, outperforming numerous baselines across both isolated vocals and complete musical mixtures. Moreover, while the availability of artist identity labels during pre-training leads to overall more consistent performance both in the evaluated downstream tasks and the user study, a label-agnostic CVSM variant incorporating hybrid pre-training with real and artificial mixtures achieves comparable performance to the label-informed one in artist identification and perceived vocal similarity.
- Open Access
- 12 December 2025
- Published by JMIR Publications Inc. in JMIR Mental Health
- Vol. 12, e75774
Smartwatch-Derived Digital Phenotypes Relate to Psychopathology Dimensions in Patients With Psychotic Spectrum Disorders: Longitudinal Observational Study
Background: Digital phenotyping refers to the objective measurement of human behavior via devices such as smartphones or watches and constitutes a promising advancement in personalized medicine. Digital phenotypes derived from heart rate, mobility, or sleep schedule data have been used in psychiatry to either diagnose individuals with psychotic disorders or to predict relapse as a binary outcome. Machine learning models so far have achieved predictive accuracies that are significant but not large enough for clinical applications. This could hinge on broad clinical definitions, which encompass heterogeneous symptom and sign ensembles, thus hindering accurate classification. The 5-factor model for the Positive and Negative Syndrome Scale (PANSS), which entails 5 independently varying dimensions, is thought to better capture symptom variability. Using the specific definitions of this refined clinical taxonomy in combination with digital phenotypes could yield more precise results. Objective: This study aims to investigate potential links between digital phenotypes and each dimension of the 5-factor PANSS model. We also assess whether clinical, demographic, and medication variables confound said reactions. Methods: In the e-Prevention study, heart rate, accelerometer, gyroscope, and sleep schedule data were continuously collected via smartwatch for a maximum of 26 months in 38 patients with psychotic spectrum disorders. Obtaining the mean and SD for each patient-month resulted in a database consisting of more than 740 monthly data points. A linear mixed model analysis was used to ascertain connections between monthly aggregated heart rate and mobility features and the 5 symptom dimension scores of PANSS, obtained during monthly clinical interviews. Results: An increase in positive symptoms was associated with a decrease in heart interpulse variation during sleep ( t 570.7 =−3.3 , P <.001, f 2 =0.021), while an increase in negative symptoms was associated with a decrease in accelerometer (mean: t 22.1 =−3.1 , P =.005, f 2 =0.042; SD: t 20 =−2.4 , P =.03, f 2 =0.019), gyroscope (mean: t 22.9 =−2.8 , P =.01, f 2 =0.016), and locomotive motor activity ( t 17.2 =−2.4 , P =.03, f 2 =0.016) during wakefulness. An increase in accelerometer (mean: t 564.4 =2.8 , P =.005, f 2 =0.017; SD: t 551.6 =2.5, P =.01, f 2 =0.015) and gyroscope (mean: t 564.5 =3.2 , P =.001, f 2 =0.022; SD: t 569.2 =2.8 , P =.005, f 2 =0.017) motor activity during sleep was related to an increase in depression/anxiety symptoms as well as excitement/hostility symptoms (accelerometer SD: t 469.7 =3.2 , P =.002, f 2 =0.031; gyroscope mean: t 497 =2.3 , P =.03, f 2 =0.013; SD: t 507.7 =3.2 , P =.001, f 2 =0.029). Excitement/hostility symptoms were further associated with an increase in normalized heart rate during sleep ( t 368.2 =3.2, P =.001, f 2 =0.044) and reduced sleep:wake ratio ( t 562 =−2.7 , P =.007, f 2 =0.013). An increase in cognitive/disorganization symptoms was related to a decrease in the SD of normalized heart rate during wakefulness ( t 574.5 =−3.5 , P <.001, f 2 =0.013). Conclusions: This study provides evidence that biological changes assessed by continuous measurement of digital phenotypes could be characteristic of specific symptom clusters rather than entire diagnostic categories of psychotic disorders. These results support the use of digital phenotypes not only as a means for remote patient monitoring but also as concrete targets for biomarker research in psychotic disorders.
- 12 December 2025
- Published by JMIR Publications Inc. in JMIR Mental Health
- Vol. 12, e75774
Open AccessSmartwatch-Derived Digital Phenotypes Relate to Psychopathology Dimensions in Patients With Psychotic Spectrum Disorders: Longitudinal Observational Study
Background: Digital phenotyping refers to the objective measurement of human behavior via devices such as smartphones or watches and constitutes a promising advancement in personalized medicine. Digital phenotypes derived from heart rate, mobility, or sleep schedule data have been used in psychiatry to either diagnose individuals with psychotic disorders or to predict relapse as a binary outcome. Machine learning models so far have achieved predictive accuracies that are significant but not large enough for clinical applications. This could hinge on broad clinical definitions, which encompass heterogeneous symptom and sign ensembles, thus hindering accurate classification. The 5-factor model for the Positive and Negative Syndrome Scale (PANSS), which entails 5 independently varying dimensions, is thought to better capture symptom variability. Using the specific definitions of this refined clinical taxonomy in combination with digital phenotypes could yield more precise results. Objective: This study aims to investigate potential links between digital phenotypes and each dimension of the 5-factor PANSS model. We also assess whether clinical, demographic, and medication variables confound said reactions. Methods: In the e-Prevention study, heart rate, accelerometer, gyroscope, and sleep schedule data were continuously collected via smartwatch for a maximum of 26 months in 38 patients with psychotic spectrum disorders. Obtaining the mean and SD for each patient-month resulted in a database consisting of more than 740 monthly data points. A linear mixed model analysis was used to ascertain connections between monthly aggregated heart rate and mobility features and the 5 symptom dimension scores of PANSS, obtained during monthly clinical interviews. Results: An increase in positive symptoms was associated with a decrease in heart interpulse variation during sleep ( t 570.7 =−3.3 , P <.001, f 2 =0.021), while an increase in negative symptoms was associated with a decrease in accelerometer (mean: t 22.1 =−3.1 , P =.005, f 2 =0.042; SD: t 20 =−2.4 , P =.03, f 2 =0.019), gyroscope (mean: t 22.9 =−2.8 , P =.01, f 2 =0.016), and locomotive motor activity ( t 17.2 =−2.4 , P =.03, f 2 =0.016) during wakefulness. An increase in accelerometer (mean: t 564.4 =2.8 , P =.005, f 2 =0.017; SD: t 551.6 =2.5, P =.01, f 2 =0.015) and gyroscope (mean: t 564.5 =3.2 , P =.001, f 2 =0.022; SD: t 569.2 =2.8 , P =.005, f 2 =0.017) motor activity during sleep was related to an increase in depression/anxiety symptoms as well as excitement/hostility symptoms (accelerometer SD: t 469.7 =3.2 , P =.002, f 2 =0.031; gyroscope mean: t 497 =2.3 , P =.03, f 2 =0.013; SD: t 507.7 =3.2 , P =.001, f 2 =0.029). Excitement/hostility symptoms were further associated with an increase in normalized heart rate during sleep ( t 368.2 =3.2, P =.001, f 2 =0.044) and reduced sleep:wake ratio ( t 562 =−2.7 , P =.007, f 2 =0.013). An increase in cognitive/disorganization symptoms was related to a decrease in the SD of normalized heart rate during wakefulness ( t 574.5 =−3.5 , P <.001, f 2 =0.013). Conclusions: This study provides evidence that biological changes assessed by continuous measurement of digital phenotypes could be characteristic of specific symptom clusters rather than entire diagnostic categories of psychotic disorders. These results support the use of digital phenotypes not only as a means for remote patient monitoring but also as concrete targets for biomarker research in psychotic disorders.
- Open Access
CVSM: Contrastive Vocal Similarity Modeling
The availability of large, unlabeled datasets across various domains has contributed to the development of a plethora of methods that learn representations for multiple target (downstream) tasks through self-supervised pre-training. In this work, we introduce CVSM (Contrastive Vocal Similarity Modeling), a contrastive self-supervised procedure for music signal representation learning in the audio domain that can be utilized for musical and vocal similarity modeling. Our method operates under a contrastive framework, maximizing the similarity between vocal excerpts and musical mixtures containing the same vocals; we devise both a label-informed protocol, leveraging artist identity information to sample the contrastive pairs, and a label-agnostic scheme, involving artificial mixture creation from randomly sampled vocal and accompaniment excerpts, which are paired with vocals from the same audio segment. We evaluate our proposed method in measuring vocal similarity both objectively, through linear probing on a suite of appropriate downstream tasks, and subjectively, via conducting a user study consisting of pairwise comparisons between different models in a recommendation-by-query setting. Our results indicate that the representations learned through CVSM are effective in musical and vocal similarity modeling, outperforming numerous baselines across both isolated vocals and complete musical mixtures. Moreover, while the availability of artist identity labels during pre-training leads to overall more consistent performance both in the evaluated downstream tasks and the user study, a label-agnostic CVSM variant incorporating hybrid pre-training with real and artificial mixtures achieves comparable performance to the label-informed one in artist identification and perceived vocal similarity.
Open AccessCVSM: Contrastive Vocal Similarity Modeling
The availability of large, unlabeled datasets across various domains has contributed to the development of a plethora of methods that learn representations for multiple target (downstream) tasks through self-supervised pre-training. In this work, we introduce CVSM (Contrastive Vocal Similarity Modeling), a contrastive self-supervised procedure for music signal representation learning in the audio domain that can be utilized for musical and vocal similarity modeling. Our method operates under a contrastive framework, maximizing the similarity between vocal excerpts and musical mixtures containing the same vocals; we devise both a label-informed protocol, leveraging artist identity information to sample the contrastive pairs, and a label-agnostic scheme, involving artificial mixture creation from randomly sampled vocal and accompaniment excerpts, which are paired with vocals from the same audio segment. We evaluate our proposed method in measuring vocal similarity both objectively, through linear probing on a suite of appropriate downstream tasks, and subjectively, via conducting a user study consisting of pairwise comparisons between different models in a recommendation-by-query setting. Our results indicate that the representations learned through CVSM are effective in musical and vocal similarity modeling, outperforming numerous baselines across both isolated vocals and complete musical mixtures. Moreover, while the availability of artist identity labels during pre-training leads to overall more consistent performance both in the evaluated downstream tasks and the user study, a label-agnostic CVSM variant incorporating hybrid pre-training with real and artificial mixtures achieves comparable performance to the label-informed one in artist identification and perceived vocal similarity.
- 25 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Audio, Speech and Language Processing
- Vol. PP, 1-12
Separate and Transcribe: Deep Guitar String Separation and its Application for Tablature Transcription Enhancement
Guitar string separation constitutes a source separation task in which the individual stems to be retrieved are the signals produced by each distinct guitar string. Certain hardware solutions for string-wise signal manipulation are gaining traction among the guitarist community for applications such as MIDI control and string-level sound effects. However, software techniques for string separation are lagging behind, even though they could form a competitive alternative. This work applies established deep learning architectures commonly used for standard music source separation, to address the task of guitar string separation. It opts for a waveform-to-waveform approach using a multi-channel version of Wave-U-Net. It further suggests that this model can serve as a bridge from separation to transcription by facilitating guitar tablature inference. Tablature transcription enhancement proved feasible using a method that relies on feeding the separated signals to standard tablature inference models, by simply modifying their initial convolution layer to handle multi-source input. Moreover, this work seeks to address the challenges encountered in the more specific string separation scenario where no target signals are readily available for training, meaning that these signals are not provided directly through sophisticated hardware equipment such as polyphonic pickups. Instead, data manipulation and augmentation techniques that produce sample-level synthesized targets are proposed. Thus, two newly created datasets are introduced, namely GS-Aux and ADGP, based on standard guitar datasets: GuitarSet and DadaGP.
- 25 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Audio, Speech and Language Processing
- Vol. PP, 1-12
Separate and Transcribe: Deep Guitar String Separation and its Application for Tablature Transcription Enhancement
Guitar string separation constitutes a source separation task in which the individual stems to be retrieved are the signals produced by each distinct guitar string. Certain hardware solutions for string-wise signal manipulation are gaining traction among the guitarist community for applications such as MIDI control and string-level sound effects. However, software techniques for string separation are lagging behind, even though they could form a competitive alternative. This work applies established deep learning architectures commonly used for standard music source separation, to address the task of guitar string separation. It opts for a waveform-to-waveform approach using a multi-channel version of Wave-U-Net. It further suggests that this model can serve as a bridge from separation to transcription by facilitating guitar tablature inference. Tablature transcription enhancement proved feasible using a method that relies on feeding the separated signals to standard tablature inference models, by simply modifying their initial convolution layer to handle multi-source input. Moreover, this work seeks to address the challenges encountered in the more specific string separation scenario where no target signals are readily available for training, meaning that these signals are not provided directly through sophisticated hardware equipment such as polyphonic pickups. Instead, data manipulation and augmentation techniques that produce sample-level synthesized targets are proposed. Thus, two newly created datasets are introduced, namely GS-Aux and ADGP, based on standard guitar datasets: GuitarSet and DadaGP.
- Open Access
Shared-Weights Extender and Gradient Voting for Neural Network Expansion
Expanding neural networks during training is a promising way to augment capacity without retraining larger models from scratch. However, newly added neurons often fail to adjust to a trained network and become inactive, providing no contribution to capacity growth. We propose the Shared-Weights Extender (SWE), a novel method explicitly designed to prevent inactivity of new neurons by coupling them with existing ones for smooth integration. In parallel, we introduce the Steepest Voting Distributor (SVoD), a gradient-based method for allocating neurons across layers during deep network expansion. Our extensive benchmarking on four datasets shows that our method can effectively suppress neuron inactivity and achieve better performance compared to other expanding methods and baselines.
Open AccessShared-Weights Extender and Gradient Voting for Neural Network Expansion
Expanding neural networks during training is a promising way to augment capacity without retraining larger models from scratch. However, newly added neurons often fail to adjust to a trained network and become inactive, providing no contribution to capacity growth. We propose the Shared-Weights Extender (SWE), a novel method explicitly designed to prevent inactivity of new neurons by coupling them with existing ones for smooth integration. In parallel, we introduce the Steepest Voting Distributor (SVoD), a gradient-based method for allocating neurons across layers during deep network expansion. Our extensive benchmarking on four datasets shows that our method can effectively suppress neuron inactivity and achieve better performance compared to other expanding methods and baselines.
- Open Access
- 16 September 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-8
Text-to-Sign Language Production via Intermediate Skeletal Representations using Transformers and Neural Rendering
- 16 September 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-8
Open AccessText-to-Sign Language Production via Intermediate Skeletal Representations using Transformers and Neural Rendering
- Open Access
Classical Guitar Duet Separation using GuitarDuets -- a Dataset of Real and Synthesized Guitar Recordings
Recent advancements in music source separation (MSS) have focused in the multi-timbral case, with existing architectures tailored for the separation of distinct instruments, overlooking thus the challenge of separating instruments with similar timbral characteristics. Addressing this gap, our work focuses on monotimbral MSS, specifically within the context of classical guitar duets. To this end, we introduce the GuitarDuets dataset, featuring a combined total of approximately three hours of real and synthesized classical guitar duet recordings, as well as note-level annotations of the synthesized duets. We perform an extensive cross-dataset evaluation by adapting Demucs, a state-of-the-art MSS architecture, to monotimbral source separation. Furthermore, we develop a joint permutation-invariant transcription and separation framework, to exploit note event predictions as auxiliary information. Our results indicate that utilizing both the real and synthesized subsets of GuitarDuets leads to improved separation performance in an independently recorded test set compared to utilizing solely one subset. We also find that while the availability of ground-truth note labels greatly helps the performance of the separation network, the predicted note estimates result only in marginal improvement. Finally, we discuss the behavior of commonly utilized metrics, such as SDR and SI-SDR, in the context of monotimbral MSS.
Open AccessClassical Guitar Duet Separation using GuitarDuets -- a Dataset of Real and Synthesized Guitar Recordings
Recent advancements in music source separation (MSS) have focused in the multi-timbral case, with existing architectures tailored for the separation of distinct instruments, overlooking thus the challenge of separating instruments with similar timbral characteristics. Addressing this gap, our work focuses on monotimbral MSS, specifically within the context of classical guitar duets. To this end, we introduce the GuitarDuets dataset, featuring a combined total of approximately three hours of real and synthesized classical guitar duet recordings, as well as note-level annotations of the synthesized duets. We perform an extensive cross-dataset evaluation by adapting Demucs, a state-of-the-art MSS architecture, to monotimbral source separation. Furthermore, we develop a joint permutation-invariant transcription and separation framework, to exploit note event predictions as auxiliary information. Our results indicate that utilizing both the real and synthesized subsets of GuitarDuets leads to improved separation performance in an independently recorded test set compared to utilizing solely one subset. We also find that while the availability of ground-truth note labels greatly helps the performance of the separation network, the predicted note estimates result only in marginal improvement. Finally, we discuss the behavior of commonly utilized metrics, such as SDR and SI-SDR, in the context of monotimbral MSS.
- Open Access
- 25 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 38-45
Indoor Turn-By-Turn Navigation Assistance for Robotic Rollators
- 25 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 38-45
Open AccessIndoor Turn-By-Turn Navigation Assistance for Robotic Rollators
- Open Access
- 25 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 237-243
A Transformer-Based Framework for Greek Sign Language Production using Extended Skeletal Motion Representations
- 25 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 237-243
Open AccessA Transformer-Based Framework for Greek Sign Language Production using Extended Skeletal Motion Representations
- 30 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. PP (2162237X) , 1-15
Revisiting Tropical Polynomial Division: Theory, Algorithms, and Application to Neural Networks
Tropical geometry has recently found several applications in the analysis of neural networks with piecewise linear activation functions. This article presents a new look at the problem of tropical polynomial division and its application to the simplification of neural networks. We analyze tropical polynomials with real coefficients, extending earlier ideas and methods developed for polynomials with integer coefficients. We first prove the existence of a unique quotient-remainder pair and characterize the quotient in terms of the convex bi-conjugate of a related function. Interestingly, the quotient of tropical polynomials with integer coefficients does not necessarily have integer coefficients. Furthermore, we develop a relationship of tropical polynomial division with the computation of the convex hull of unions of convex polyhedra and use it to derive an exact algorithm for tropical polynomial division. An approximate algorithm is also presented, based on an alternation between data partition and linear programming. We also develop special techniques to divide composite polynomials, described as sums or maxima of simpler ones. Finally, we provide numerical results to demonstrate the efficiency of the proposed algorithms, using the MNIST handwritten digits, SVHN, CIFAR-10, and CIFAR-100 datasets, along with an application example in learning model predictive control (LMPC).
- 30 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. PP (2162237X) , 1-15
Revisiting Tropical Polynomial Division: Theory, Algorithms, and Application to Neural Networks
Tropical geometry has recently found several applications in the analysis of neural networks with piecewise linear activation functions. This article presents a new look at the problem of tropical polynomial division and its application to the simplification of neural networks. We analyze tropical polynomials with real coefficients, extending earlier ideas and methods developed for polynomials with integer coefficients. We first prove the existence of a unique quotient-remainder pair and characterize the quotient in terms of the convex bi-conjugate of a related function. Interestingly, the quotient of tropical polynomials with integer coefficients does not necessarily have integer coefficients. Furthermore, we develop a relationship of tropical polynomial division with the computation of the convex hull of unions of convex polyhedra and use it to derive an exact algorithm for tropical polynomial division. An approximate algorithm is also presented, based on an alternation between data partition and linear programming. We also develop special techniques to divide composite polynomials, described as sums or maxima of simpler ones. Finally, we provide numerical results to demonstrate the efficiency of the proposed algorithms, using the MNIST handwritten digits, SVHN, CIFAR-10, and CIFAR-100 datasets, along with an application example in learning model predictive control (LMPC).