| 1. | MO Spata; VM Russo; A Ortis; S Battiato: A New Pipeline for Snooping Keystroke Based on Deep Learning Algorithm. In: IEEE Access, 13 , pp. 24498 - 24514, 2025. (Type: Journal Article | Abstract | Links | BibTeX) @article{Spata2025, title = {A New Pipeline for Snooping Keystroke Based on Deep Learning Algorithm}, author = {MO Spata and VM Russo and A Ortis and S Battiato}, editor = {IEEE}, url = {https://ieeexplore.ieee.org/abstract/document/10858134}, doi = {10.1109/ACCESS.2025.3536877}, year = {2025}, date = {2025-01-29}, journal = {IEEE Access}, volume = {13}, pages = {24498 - 24514}, abstract = {This research focuses on the vulnerabilities of keystroke by logging on a physical computer keyboard, known as Snooping Keystroke. This category of attacks occurs recording an audio track with a smartphone while typing on the keyboard, and processing the audio to detect individual pressed keys. To address this issue, mathematical wavelet transforms have been tested, while key recognition has been implemented using the inference test of a deep learning model based on a Temporal Convolutional Network (TCN). The novelty of the proposed pipeline lies in its dynamic audio analysis and keystroke recognition, which splits the wave based on audio signal peaks generated by key presses. This approach enables an attack in real-world conditions without knowing the exact number of keystrokes typed by the user. Experimental results show a peak accuracy of 98.3%.}, keywords = {}, pubstate = {published}, tppubtype = {article} } This research focuses on the vulnerabilities of keystroke by logging on a physical computer keyboard, known as Snooping Keystroke. This category of attacks occurs recording an audio track with a smartphone while typing on the keyboard, and processing the audio to detect individual pressed keys. To address this issue, mathematical wavelet transforms have been tested, while key recognition has been implemented using the inference test of a deep learning model based on a Temporal Convolutional Network (TCN). The novelty of the proposed pipeline lies in its dynamic audio analysis and keystroke recognition, which splits the wave based on audio signal peaks generated by key presses. This approach enables an attack in real-world conditions without knowing the exact number of keystrokes typed by the user. Experimental results show a peak accuracy of 98.3%. |
| 2. | Massimo Orazio Spata, Valerio Maria Russo, Alessandro Ortis, Sebastiano Battiato: Acoustic Side Channel Attack for Keystroke Splitting in the Wild. 2024, ISBN: 979-8-3503-7800-9. (Type: Proceeding | Abstract | Links | BibTeX) @proceedings{Spata2024, title = {Acoustic Side Channel Attack for Keystroke Splitting in the Wild}, author = {Massimo Orazio Spata, Valerio Maria Russo, Alessandro Ortis, Sebastiano Battiato}, editor = {IEEE}, url = {https://ieeexplore.ieee.org/abstract/document/10796234}, doi = {10.1109/MetroXRAINE62247.2024.10796234}, isbn = {979-8-3503-7800-9}, year = {2024}, date = {2024-12-24}, booktitle = {2024 IEEE International Conference on Metrology for eXtended Reality, Artificial Intelligence and Neural Engineering (MetroXRAINE)}, pages = {131-136}, abstract = {With the increasing prevalence of smart devices in our daily lives, concerns about security vulnerabilities have become more prominent. Acoustic Side Channel Attacks (ASCA) have emerged as a significant threat, exploiting sound emissions from devices to infer sensitive information such as keystrokes or conversations. We propose a novel approach leveraging deep learning techniques for audio segmentation in ASCA scenarios. Our method involves preprocessing the audio data, extracting relevant features, and applying a Temporal Convolutional Network (TCN) model for keystroke classification. We conducted experiments on a diverse dataset comprising various types of smart devices and attack scenarios. Our experiments demonstrate the effectiveness of the proposed audio segmentation method as fundamental preparatory step in ASCA attacks. The segmentation model achieved high precision, sensitivity, and specificity values, indicating its robustness in accurately. Furthermore, we observed consistent performance across different types of smart devices and attack scenarios, highlighting the generalizability of our approach in real conditions. The high precision, sensitivity, and specificity values obtained in our evaluation underscore the reliability and practical utility of the proposed approach. Experimental results demonstrate a peak accuracy of 990%, showcasing the effectiveness and precision of the approach. This high level of accuracy underscores the reliability and potential real-world applicability of the findings for keystroke splitting in the wild.}, keywords = {}, pubstate = {published}, tppubtype = {proceedings} } With the increasing prevalence of smart devices in our daily lives, concerns about security vulnerabilities have become more prominent. Acoustic Side Channel Attacks (ASCA) have emerged as a significant threat, exploiting sound emissions from devices to infer sensitive information such as keystrokes or conversations. We propose a novel approach leveraging deep learning techniques for audio segmentation in ASCA scenarios. Our method involves preprocessing the audio data, extracting relevant features, and applying a Temporal Convolutional Network (TCN) model for keystroke classification. We conducted experiments on a diverse dataset comprising various types of smart devices and attack scenarios. Our experiments demonstrate the effectiveness of the proposed audio segmentation method as fundamental preparatory step in ASCA attacks. The segmentation model achieved high precision, sensitivity, and specificity values, indicating its robustness in accurately. Furthermore, we observed consistent performance across different types of smart devices and attack scenarios, highlighting the generalizability of our approach in real conditions. The high precision, sensitivity, and specificity values obtained in our evaluation underscore the reliability and practical utility of the proposed approach. Experimental results demonstrate a peak accuracy of 990%, showcasing the effectiveness and precision of the approach. This high level of accuracy underscores the reliability and potential real-world applicability of the findings for keystroke splitting in the wild. |
| 3. | Massimo Orazio Spata, , Valerio Maria Russo, Alessandro Ortis, Sebastiano Battiato: A New Deep Learning Pipeline for Acoustic Attack on Keyboards. 1065 , 2024, ISBN: 978-3-031-66328-4. (Type: Proceeding | Abstract | Links | BibTeX) @proceedings{Spata2024b, title = {A New Deep Learning Pipeline for Acoustic Attack on Keyboards}, author = {Massimo Orazio Spata, , Valerio Maria Russo, Alessandro Ortis, Sebastiano Battiato}, editor = {Springer}, url = {https://link.springer.com/chapter/10.1007/978-3-031-66329-1_26#citeas}, doi = {https://doi.org/10.1007/978-3-031-66329-1_26}, isbn = {978-3-031-66328-4}, year = {2024}, date = {2024-07-31}, booktitle = {Lecture Notes in Networks and Systems ((LNNS,volume 1065))}, volume = {1065}, pages = {402-414}, abstract = {The increasing reliance on services based on recent Artificial Intelligence advancements has elevated concerns about security vulnerabilities, leading to the exploration of novel attack vectors such as keystroke acoustic attacks on keyboards. This research delves into a deep learning approach for such attacks, which exploits acoustic emissions produced during typing to infer sensitive information. Traditional methods of keystroke acoustic attacks have relied on hand-engineered features and shallow classifiers, often failing to capture the intricate patterns within the acoustic data. In contrast, deep learning models have demonstrated remarkable capabilities in learning intricate patterns from complex data sources. We propose the exploitation of a Temporal Convolutional Network (TCN) to process acoustic signals, providing a more sophisticated and adaptive approach for keystroke acoustic attack analysis. The employed deep learning model showcases superior performance in multiple dimensions achieving a peak validation accuracy of 98.3% for keystrokes recorded by phone, and 93.05% for keystrokes recorded via Zoom, obtaining the best performances with respect the related prior art.}, keywords = {}, pubstate = {published}, tppubtype = {proceedings} } The increasing reliance on services based on recent Artificial Intelligence advancements has elevated concerns about security vulnerabilities, leading to the exploration of novel attack vectors such as keystroke acoustic attacks on keyboards. This research delves into a deep learning approach for such attacks, which exploits acoustic emissions produced during typing to infer sensitive information. Traditional methods of keystroke acoustic attacks have relied on hand-engineered features and shallow classifiers, often failing to capture the intricate patterns within the acoustic data. In contrast, deep learning models have demonstrated remarkable capabilities in learning intricate patterns from complex data sources. We propose the exploitation of a Temporal Convolutional Network (TCN) to process acoustic signals, providing a more sophisticated and adaptive approach for keystroke acoustic attack analysis. The employed deep learning model showcases superior performance in multiple dimensions achieving a peak validation accuracy of 98.3% for keystrokes recorded by phone, and 93.05% for keystrokes recorded via Zoom, obtaining the best performances with respect the related prior art. |
2025 |
MO Spata; VM Russo; A Ortis; S Battiato A New Pipeline for Snooping Keystroke Based on Deep Learning Algorithm Journal Article IEEE Access, 13 , pp. 24498 - 24514, 2025. Abstract | Links | BibTeX | Tags: Acoustic side channel attack, Deep learning, laptop keystroke attacks, snooping keystroke attacks, user security and privacy, zoom-based acoustic attacks @article{Spata2025, title = {A New Pipeline for Snooping Keystroke Based on Deep Learning Algorithm}, author = {MO Spata and VM Russo and A Ortis and S Battiato}, editor = {IEEE}, url = {https://ieeexplore.ieee.org/abstract/document/10858134}, doi = {10.1109/ACCESS.2025.3536877}, year = {2025}, date = {2025-01-29}, journal = {IEEE Access}, volume = {13}, pages = {24498 - 24514}, abstract = {This research focuses on the vulnerabilities of keystroke by logging on a physical computer keyboard, known as Snooping Keystroke. This category of attacks occurs recording an audio track with a smartphone while typing on the keyboard, and processing the audio to detect individual pressed keys. To address this issue, mathematical wavelet transforms have been tested, while key recognition has been implemented using the inference test of a deep learning model based on a Temporal Convolutional Network (TCN). The novelty of the proposed pipeline lies in its dynamic audio analysis and keystroke recognition, which splits the wave based on audio signal peaks generated by key presses. This approach enables an attack in real-world conditions without knowing the exact number of keystrokes typed by the user. Experimental results show a peak accuracy of 98.3%.}, keywords = {Acoustic side channel attack, Deep learning, laptop keystroke attacks, snooping keystroke attacks, user security and privacy, zoom-based acoustic attacks}, pubstate = {published}, tppubtype = {article} } This research focuses on the vulnerabilities of keystroke by logging on a physical computer keyboard, known as Snooping Keystroke. This category of attacks occurs recording an audio track with a smartphone while typing on the keyboard, and processing the audio to detect individual pressed keys. To address this issue, mathematical wavelet transforms have been tested, while key recognition has been implemented using the inference test of a deep learning model based on a Temporal Convolutional Network (TCN). The novelty of the proposed pipeline lies in its dynamic audio analysis and keystroke recognition, which splits the wave based on audio signal peaks generated by key presses. This approach enables an attack in real-world conditions without knowing the exact number of keystrokes typed by the user. Experimental results show a peak accuracy of 98.3%. |
2024 |
Massimo Orazio Spata, Valerio Maria Russo, Alessandro Ortis, Sebastiano Battiato Acoustic Side Channel Attack for Keystroke Splitting in the Wild Proceeding 2024, ISBN: 979-8-3503-7800-9. Abstract | Links | BibTeX | Tags: Acoustic side channel attack, Deep learning, laptop keystroke attacks, user security and privacy @proceedings{Spata2024, title = {Acoustic Side Channel Attack for Keystroke Splitting in the Wild}, author = {Massimo Orazio Spata, Valerio Maria Russo, Alessandro Ortis, Sebastiano Battiato}, editor = {IEEE}, url = {https://ieeexplore.ieee.org/abstract/document/10796234}, doi = {10.1109/MetroXRAINE62247.2024.10796234}, isbn = {979-8-3503-7800-9}, year = {2024}, date = {2024-12-24}, booktitle = {2024 IEEE International Conference on Metrology for eXtended Reality, Artificial Intelligence and Neural Engineering (MetroXRAINE)}, pages = {131-136}, abstract = {With the increasing prevalence of smart devices in our daily lives, concerns about security vulnerabilities have become more prominent. Acoustic Side Channel Attacks (ASCA) have emerged as a significant threat, exploiting sound emissions from devices to infer sensitive information such as keystrokes or conversations. We propose a novel approach leveraging deep learning techniques for audio segmentation in ASCA scenarios. Our method involves preprocessing the audio data, extracting relevant features, and applying a Temporal Convolutional Network (TCN) model for keystroke classification. We conducted experiments on a diverse dataset comprising various types of smart devices and attack scenarios. Our experiments demonstrate the effectiveness of the proposed audio segmentation method as fundamental preparatory step in ASCA attacks. The segmentation model achieved high precision, sensitivity, and specificity values, indicating its robustness in accurately. Furthermore, we observed consistent performance across different types of smart devices and attack scenarios, highlighting the generalizability of our approach in real conditions. The high precision, sensitivity, and specificity values obtained in our evaluation underscore the reliability and practical utility of the proposed approach. Experimental results demonstrate a peak accuracy of 990%, showcasing the effectiveness and precision of the approach. This high level of accuracy underscores the reliability and potential real-world applicability of the findings for keystroke splitting in the wild.}, keywords = {Acoustic side channel attack, Deep learning, laptop keystroke attacks, user security and privacy}, pubstate = {published}, tppubtype = {proceedings} } With the increasing prevalence of smart devices in our daily lives, concerns about security vulnerabilities have become more prominent. Acoustic Side Channel Attacks (ASCA) have emerged as a significant threat, exploiting sound emissions from devices to infer sensitive information such as keystrokes or conversations. We propose a novel approach leveraging deep learning techniques for audio segmentation in ASCA scenarios. Our method involves preprocessing the audio data, extracting relevant features, and applying a Temporal Convolutional Network (TCN) model for keystroke classification. We conducted experiments on a diverse dataset comprising various types of smart devices and attack scenarios. Our experiments demonstrate the effectiveness of the proposed audio segmentation method as fundamental preparatory step in ASCA attacks. The segmentation model achieved high precision, sensitivity, and specificity values, indicating its robustness in accurately. Furthermore, we observed consistent performance across different types of smart devices and attack scenarios, highlighting the generalizability of our approach in real conditions. The high precision, sensitivity, and specificity values obtained in our evaluation underscore the reliability and practical utility of the proposed approach. Experimental results demonstrate a peak accuracy of 990%, showcasing the effectiveness and precision of the approach. This high level of accuracy underscores the reliability and potential real-world applicability of the findings for keystroke splitting in the wild. |
Massimo Orazio Spata, , Valerio Maria Russo, Alessandro Ortis, Sebastiano Battiato A New Deep Learning Pipeline for Acoustic Attack on Keyboards Proceeding 1065 , 2024, ISBN: 978-3-031-66328-4. Abstract | Links | BibTeX | Tags: Acoustic side channel attack, Deep learning, laptop keystroke attacks, user security and privacy, zoom-based acoustic attacks @proceedings{Spata2024b, title = {A New Deep Learning Pipeline for Acoustic Attack on Keyboards}, author = {Massimo Orazio Spata, , Valerio Maria Russo, Alessandro Ortis, Sebastiano Battiato}, editor = {Springer}, url = {https://link.springer.com/chapter/10.1007/978-3-031-66329-1_26#citeas}, doi = {https://doi.org/10.1007/978-3-031-66329-1_26}, isbn = {978-3-031-66328-4}, year = {2024}, date = {2024-07-31}, booktitle = {Lecture Notes in Networks and Systems ((LNNS,volume 1065))}, volume = {1065}, pages = {402-414}, abstract = {The increasing reliance on services based on recent Artificial Intelligence advancements has elevated concerns about security vulnerabilities, leading to the exploration of novel attack vectors such as keystroke acoustic attacks on keyboards. This research delves into a deep learning approach for such attacks, which exploits acoustic emissions produced during typing to infer sensitive information. Traditional methods of keystroke acoustic attacks have relied on hand-engineered features and shallow classifiers, often failing to capture the intricate patterns within the acoustic data. In contrast, deep learning models have demonstrated remarkable capabilities in learning intricate patterns from complex data sources. We propose the exploitation of a Temporal Convolutional Network (TCN) to process acoustic signals, providing a more sophisticated and adaptive approach for keystroke acoustic attack analysis. The employed deep learning model showcases superior performance in multiple dimensions achieving a peak validation accuracy of 98.3% for keystrokes recorded by phone, and 93.05% for keystrokes recorded via Zoom, obtaining the best performances with respect the related prior art.}, keywords = {Acoustic side channel attack, Deep learning, laptop keystroke attacks, user security and privacy, zoom-based acoustic attacks}, pubstate = {published}, tppubtype = {proceedings} } The increasing reliance on services based on recent Artificial Intelligence advancements has elevated concerns about security vulnerabilities, leading to the exploration of novel attack vectors such as keystroke acoustic attacks on keyboards. This research delves into a deep learning approach for such attacks, which exploits acoustic emissions produced during typing to infer sensitive information. Traditional methods of keystroke acoustic attacks have relied on hand-engineered features and shallow classifiers, often failing to capture the intricate patterns within the acoustic data. In contrast, deep learning models have demonstrated remarkable capabilities in learning intricate patterns from complex data sources. We propose the exploitation of a Temporal Convolutional Network (TCN) to process acoustic signals, providing a more sophisticated and adaptive approach for keystroke acoustic attack analysis. The employed deep learning model showcases superior performance in multiple dimensions achieving a peak validation accuracy of 98.3% for keystrokes recorded by phone, and 93.05% for keystrokes recorded via Zoom, obtaining the best performances with respect the related prior art. |