TY - CPAPER TI - Comparison of Vision-based Object Detectors and an Audio Transformer for Whale Sound Event Detection AU - Maaz Saeed AU - Bidarouni, Amir Latifi AU - Lukashevich, Hanna AU - Abeßer, Jakob PY - 2026 T2 - Proceedings of the 11th Workshop on Detection and Classication of Acoustic Scenes and Events (DCASE 2026) SP - 176-180 PB - Zenodo SN - 979-8-234-25461-0 DO - 10.5281/zenodo.22954971 UR - https://fis.uni-bamberg.de/handle/uniba/117980 LA - eng ER - TY - CHAP TI - Audio-Based Tools for Decontextualisation Detection AU - Gerhardt, Milica AU - Cuccovillo, Luca AU - Aichroth, Patrick AU - Abeßer, Jakob PY - 2026 T2 - Countering Disinformation in the Era of Generative AI SP - 317-341 PB - Springer Nature Switzerland CY - Cham SN - 9783032117816 DO - 10.1007/978-3-032-11782-3_11 UR - https://fis.uni-bamberg.de/handle/uniba/115155 LA - eng ER - TY - JOUR TI - JSD : A Dataset for Structure Analysis in Jazz Music AU - Balke, Stefan AU - Reck, Julian AU - Weiß, Christof AU - Abeßer, Jakob AU - Müller, Meinard PY - 2026 SP - 156-172 PB - Otto-Friedrich-Universität CY - Bamberg UR - https://fis.uni-bamberg.de/handle/uniba/112997 LA - eng AB - Given a music recording, music structure analysis aims at identifying important structural elements and segmenting the recording according to these elements. In jazz music, a performance is often structured by repeating harmonic schemata (known as choruses), which lay the foundation for improvisation by soloists. Within the fields of music information retrieval (MIR) and computational musicology, the Weimar Jazz Database (WJD) has turned out to be an extremely valuable resource for jazz research. Containing high-quality solo transcriptions for 456 solo sections, the dataset opened up new avenues for the understanding of creative processes in jazz improvisation using computational methods. In this paper, we complement this dataset by introducing the Jazz Structure Dataset (JSD), which provides annotations on structure and instrumentation of entire recordings. The JSD comprises 340 recordings with more than 3000 annotated segments, along with a segment-wise encoding of the solo and accompanying instruments. These annotations provide the basis for training, testing, and evaluating models for various important MIR tasks, including structure analysis, solo detection, or instrument recognition. As an example application, we consider the task of structure boundary detection. Based on a traditional novelty-based as well as a more recent data-driven approach using deep learning, we indicate the potential of the JSD while critically reflecting on some evaluation aspects of structure analysis. In this context, we also demonstrate how the JSD annotations and analysis results can be made accessible in a user-friendly way via web-based interfaces for data inspection and visualization. All annotations, experimental results, and code for reproducibility are made publicly available for research purposes. KW - Dataset KW - Music Structure Analysis KW - Boundary Detection KW - Instrument Recognition KW - Jazz ER - TY - CPAPER TI - Towards Interpreting and Improving the Latent Space for Musical Chord Recognition AU - Nadar, Christon R. AU - Taenzer, Michael AU - Abeßer, Jakob PY - 2026 SP - 74-79 PB - Otto-Friedrich-Universität CY - Bamberg UR - https://fis.uni-bamberg.de/handle/uniba/112996 LA - eng AB - Automatic chord recognition (ACR) naturally faces musical ambiguities between chord classes. These can be responsible for many misclassifications, especially in large chord vocabularies. In this paper, we propose a metric learning approach utilizing a triplet loss for the task of ACR in order to reduce chord ambiguities. In particular, we investigate how metric learning with different triplet sampling strategies re-aligns the distances between different chord classes in the latent space. Our main finding is that metric learning significantly improves the ACR performance for two taxonomies with five and nine chord classes. KW - Musical Chord Recognition ER - TY - CPAPER TI - How Reliable are Posterior Class Probabilties in Automatic Music Classification? AU - Lukashevich, Hanna AU - Grollmisch, Sascha AU - Abeßer, Jakob AU - Stober, Sebastian AU - Bös, Joachim PY - 2026 SP - 45-50 PB - Otto-Friedrich-Universität CY - Bamberg UR - https://fis.uni-bamberg.de/handle/uniba/112894 LA - eng AB - Music classification algorithms use signal processing and machine learning approaches to extract and enrich metadata for audio recordings in music archives. Common tasks include music genre classification, where each song is assigned a single label (such as Rock, Pop, or Jazz), and musical instrument classification. Since music metadata can be ambiguous, classification algorithms cannot always achieve fully accurate predictions. Therefore, our focus extends beyond the correctly estimated class labels to include realistic confidence values for each potential genre or instrument label. In practice, many state-of-the-art classification algorithms based on deep neural networks exhibit overconfident predictions, complicating the interpretation of the final output values. In this work, we examine whether the issue of overconfident predictions and, consequently, non-representative confidence values is also relevant to music genre classification and musical instrument classification. Moreover, we describe techniques to mitigate this behavior and assess the impact of deep ensembles and temperature scaling in generating more realistic confidence outputs, which can be directly employed in real-world music tagging applications. KW - music information retrieval KW - music classifcation KW - uncertainty KW - temperature scaling KW - deep ensembles ER - TY - JOUR TI - How Robust are Audio Embeddings for Polyphonic Sound Event Tagging? AU - Abeßer, Jakob AU - Grollmisch, Sascha AU - Müller, Meinard PY - 2026 SP - 2658-2667 PB - Otto-Friedrich-Universität CY - Bamberg UR - https://fis.uni-bamberg.de/handle/uniba/112893 LA - eng AB - Sound classification algorithms are challenged by the natural variability of everyday sounds, particularly for large sound class taxonomies. In order to be applicable in real-life environments, such algorithms must also be able to handle polyphonic scenarios, where simultaneously occurring and overlapping sound events need to be classified. With the rapid progress of deep learning, several deep audio embeddings (DAEs) have been proposed as pre-trained feature representations for sound classification. In this article, we analyze the embedding spaces of two non-trainable audio representations (NTARs) and five DAEs for sound classification in polyphonic scenarios (sound event tagging) and make several contributions. First, we compare general properties like the inter-correlation between feature dimensions and the scattering of sound classes in the embedding spaces. Second, we test the robustness of the embeddings against several audio degradations and propose two sensitivity measures based on a class-agnostic and a class-centric view on the resulting drift in the embedding space. Finally, as a central contribution, we study how a blending between pairs of sounds maps to embedding space trajectories and how the path of these trajectories can cause classification errors due to their proximity to other sound classes. Throughout our analyses, the PANN embeddings have shown the best overall performance for low-polyphony sound event tagging. KW - sound event tagging KW - sound polyphony KW - deep audio embeddings KW - embedding space ER - TY - CPAPER TI - Enhancing Multiscale Features for Efficient Acoustic Scene Classification with One-Dimensional Separate CNN AU - He, Yuxuan AU - Raake, Alexander AU - Abeßer, Jakob PY - 2026 SP - 125-129 PB - Otto-Friedrich-Universität CY - Bamberg UR - https://fis.uni-bamberg.de/handle/uniba/112891 LA - eng AB - Acoustic Scene Classifcation (ASC) is a fundamental task in audio signal processing, aiming to classify the location of an environmental audio recording. Recent advances focus on improving ASC model effciency, particularly in resource-constrained environments. Convolutional neural networks (CNNs) remain the dominant approach due to their high performance, with recent focus on 1D kernels, such as in the Time-Frequency Separate Network (TFSepNet), for reducing model complexity and computational cost. However,TFSepNet performs feature extraction using a fixed receptive field in both time and frequency dimensions, which restrictsits ability to capture multiscale contextual patterns. In this study, we investigate the integration of multiscale feature extraction modules into TF-SepNet, with the aim of improving model effciency by balancing accuracy and complexity. We propose three architectures, TFSepDCD-Net, TFSepSPP-Net, andTFSepASPP-Net, each with two architectural variants based on TF-SepNet: one replaces its max pooling layers, and the other replaces its final convolutional layer. Each architecture has three confgurations corresponding to different model sizes - small, medium, and large - to explore the tradeoff between accuracy and model complexity. Our experiments show that in corporating multiscale modules allows smaller models to achieve comparable or even superior accuracy to larger baselines. These fndings highlight the potential of multiscale representations for improving the effciency of CNN-based ASC systems, especially in 1D separate architectures like TF-SepNet. KW - acoustic scene classifcation KW - time-frequency separate network KW - 1D kernels KW - multiscale feature extraction ER - TY - CPAPER TI - A Three-Level Evaluation Protocoll for Acoustic Scene Understanding of Large Language Audio Models AU - Dilip, Harish AU - Abeßer, Jakob PY - 2026 SP - 185-189 PB - Otto-Friedrich-Universität CY - Bamberg UR - https://fis.uni-bamberg.de/handle/uniba/112887 LA - eng AB - Reaching a semantic understanding of complex acoustic scenes requires computational models to capture the temporal-spatial sound source composition as well as individual sound events. This is a great challenge for computational models due to the large variety of everyday sound events and the extensive temporal-spectral overlap in real-life acoustic scenes. In this work, we aim to evaluate the acoustic scene understanding capabilities of two large audio-language models (LALMs). As a challenging scenario, we use the USM dataset, which features synthetic urban soundscapes with 2-6 overlapping sound sources per mixture. Our main contribution is a novel three-layer evaluation protocol, which includes four analysis tasks for low-level sound event understanding (sound event tagging), mid-level understanding and reasoning (sound polyphony estimation, sound source loudness ranking), as well as high-level scene understanding (audio captioning). We apply standardized metrics to assess the model's performances for each task. The proposed mulit-layer protocol allows for a fine-grained analysis of model behavior across soundscapes of various complexity levels. Our results indicate that despite their rmarkable controllability using textual instructions, the ability of state-of-the-art LALMs to understand acoustic scenes is still limited as the performance on individual analysis tasks desgrades with increasing sound polyphony. KW - audio captioning KW - large audio-language models KW - sound event tagging KW - sound polyphony estimation KW - sound source loudness ranking ER - TY - JOUR TI - Sound Recurrence Analysis for Acoustic Scene Classification AU - Abeßer, Jakob AU - Liang, Zhiwei AU - Seeber, Bernhard PY - 2026 SP - 1-15 PB - Otto-Friedrich-Universität CY - Bamberg UR - https://fis.uni-bamberg.de/handle/uniba/112771 LA - eng AB - In everyday life, people experience different soundscapes in which natural sounds, animal noises, and man-made sounds blend together. Although there have been several studies on the importance of recurring sound patterns in music and language, the relevance of this phenomenon in natural soundscapes is still largely unexplored. In this article, we study the repetition patterns of harmonic and transient sound events as potential cues for acoustic scene classification (ASC). In the first part of our study, our aim is to identify acoustic scene classes that exhibit characteristic sound repetition patterns concerning harmonic and transient sounds. We propose three metrics to measure the overall prevalence of sound repetitions as well as their repetition periods and temporal stability. In the second part, we evaluate three strategies to incorporate self-similarity matrices as an additional input feature to a convolutional neural network architecture for ASC. We observe the characteristic repetition of transient sounds in recordings of “park” and “street traffic” as well as harmonic sound repetitions in acoustic scene classes related to public transportation. In the ASC experiments, hybrid network architectures, which combine spectrogram features and features from sound recurrence analysis, show increased accuracy for those classes with prominent sound repetition patterns. Our findings provide additional perspective on the distinctions among acoustic scenes previously primarily ascribed in the literature to their spectral features. KW - Acoustic scene classification KW - Sound recurrence analysis KW - Sound repetition patterns KW - Self-similarity matrix KW - Harmonic-percussive source separation KW - Result fusion KW - Ensemble models ER - TY - CPAPER TI - Improvement and Cross-Domain Evaluation of SlowFast Networks AU - Kumar, Ravi AU - Grollmisch, Sascha AU - Abeßer, Jakob PY - 2025 T2 - 2025 AES International Conference on Artificial Intelligence and Machine Learning for Audio: Queen Mary University of London 8th to 10th, September 2025 : Proceedings SP - 211-219 UR - https://fis.uni-bamberg.de/handle/uniba/113068 LA - eng ER - TY - CPAPER TI - From CNN to Reservoir Computing : A New Perspective on Acoustic Scene Classification AU - He, Yuxuan AU - Hosseini, Alireza Molla Ali AU - Abeßer, Jakob AU - Jaurigue, Lina AU - Raake, Alexander AU - Lüdge, Kathy PY - 2025 T2 - 2025 AES International Conference on Artificial Intelligence and Machine Learning for Audio: Queen Mary University of London 8th to 10th, September 2025 : Proceedings SP - 201-210 UR - https://fis.uni-bamberg.de/handle/uniba/113069 LA - eng ER - TY - CPAPER TI - NeuroSensEar : Neuromorphic Acoustic Sensing for the High-Performance Hearing Aids of Tomorrow AU - Lenk, Claudia AU - He, Yuxuan AU - Hosseini, Alireza AU - Zheng, Fang AU - Mukherjee, Aayushmi AU - Abeßer, Jakob AU - Jaurigue, Lina AU - Lüdge, Kathy AU - Raake, Alexander AU - Ivanov, Tzvetan AU - Pieper, Iko AU - Fritzsche, Paul AU - Küller, Jan AU - Beer, Daniel AU - Schöneich, Stefan AU - Nowotny, Manuela AU - Uziel, Sebastian AU - Hutschenreuther, Tino AU - Ziegler, Martin PY - 2025 T2 - Proceedings of DAS/DAGA 2025 : 51st Annual Meeting on Acoustics : March 17-20, 2025, Copenhagen SP - 509-512 CY - Berlin SN - 978-3-939296-23-2 DO - 10.71568/dasdaga2025.523 UR - https://fis.uni-bamberg.de/handle/uniba/111571 LA - eng ER - TY - CPAPER TI - Enhancing Multiscale Features for Efficient Acoustic Scene Classification with One-Dimensional Separate CNN AU - He, Yuxuan AU - Raake, Alexander AU - Abeßer, Jakob PY - 2025 T2 - Proceedings of the 10th Workshop on Detection and Classification of Acoustic Scenes and Events (DCASE 2025), October 2025 SP - 125-129 PB - Zenodo SN - 978-84-09-77652-8 DO - 10.5281/zenodo.17251589 UR - https://fis.uni-bamberg.de/handle/uniba/111568 LA - eng AB - Acoustic Scene Classifcation (ASC) is a fundamental task in audio signal processing, aiming to classify the location of an environmental audio recording. Recent advances focus on improving ASC model effciency, particularly in resource-constrained environments. Convolutional neural networks (CNNs) remain the dominant approach due to their high performance, with recent focus on 1D kernels, such as in the Time-Frequency Separate Network (TFSepNet), for reducing model complexity and computational cost. However,TFSepNet performs feature extraction using a fixed receptive field in both time and frequency dimensions, which restrictsits ability to capture multiscale contextual patterns. In this study, we investigate the integration of multiscale feature extraction modules into TF-SepNet, with the aim of improving model effciency by balancing accuracy and complexity. We propose three architectures, TFSepDCD-Net, TFSepSPP-Net, andTFSepASPP-Net, each with two architectural variants based on TF-SepNet: one replaces its max pooling layers, and the other replaces its final convolutional layer. Each architecture has three confgurations corresponding to different model sizes - small, medium, and large - to explore the tradeoff between accuracy and model complexity. Our experiments show that in corporating multiscale modules allows smaller models to achieve comparable or even superior accuracy to larger baselines. These fndings highlight the potential of multiscale representations for improving the effciency of CNN-based ASC systems, especially in 1D separate architectures like TF-SepNet. KW - acoustic scene classifcation KW - time-frequency separate network KW - 1D kernels KW - multiscale feature extraction ER - TY - CPAPER TI - A Three-Level Evaluation Protocoll for Acoustic Scene Understanding of Large Language Audio Models AU - Dilip, Harish AU - Abeßer, Jakob PY - 2025 T2 - Proceedings of the 10th Workshop on Detection and Classification of Acoustic Scenes and Events (DCASE 2025), October 2025 SP - 185-189 PB - zenodo SN - 978-84-09-77652-8 DO - 10.5281/zenodo.17251589 UR - https://fis.uni-bamberg.de/handle/uniba/111565 LA - eng AB - Reaching a semantic understanding of complex acoustic scenes requires computational models to capture the temporal-spatial sound source composition as well as individual sound events. This is a great challenge for computational models due to the large variety of everyday sound events and the extensive temporal-spectral overlap in real-life acoustic scenes. In this work, we aim to evaluate the acoustic scene understanding capabilities of two large audio-language models (LALMs). As a challenging scenario, we use the USM dataset, which features synthetic urban soundscapes with 2-6 overlapping sound sources per mixture. Our main contribution is a novel three-layer evaluation protocol, which includes four analysis tasks for low-level sound event understanding (sound event tagging), mid-level understanding and reasoning (sound polyphony estimation, sound source loudness ranking), as well as high-level scene understanding (audio captioning). We apply standardized metrics to assess the model's performances for each task. The proposed mulit-layer protocol allows for a fine-grained analysis of model behavior across soundscapes of various complexity levels. Our results indicate that despite their rmarkable controllability using textual instructions, the ability of state-of-the-art LALMs to understand acoustic scenes is still limited as the performance on individual analysis tasks desgrades with increasing sound polyphony. KW - audio captioning KW - large audio-language models KW - sound event tagging KW - sound polyphony estimation KW - sound source loudness ranking ER - TY - CPAPER TI - Semi-Supervised Learning for Acoustic Scene Classification using FixMatch AU - Grollmisch, Sascha AU - Kumar, Ravi AU - Abeßer, Jakob PY - 2025 T2 - Proceedings of DAS/DAGA 2025 : 51st Annual Meeting on Acoustics : March 17-20, 2025, Copenhagen SP - 109-111 CY - Berlin SN - 978-3-939296-23-2 UR - https://fis.uni-bamberg.de/handle/uniba/111564 LA - eng ER - TY - JOUR TI - Fortschritte in der automatischen Erkennung von Vogelstimmen AU - Abeßer, Jakob AU - Lukashevich, Hanna AU - Ziegler, Sebastian AU - Bös, Joachim PY - 2025 JO - Akustik Journal IS - 3 SP - 7-16 CY - Berlin SN - 2569–1597 UR - https://fis.uni-bamberg.de/handle/uniba/111563 LA - deu ER - TY - JOUR TI - Sound Recurrence Analysis for Acoustic Scene Classification AU - Abeßer, Jakob AU - Liang, Zhiwei AU - Seeber, Bernhard PY - 2025 JO - EURASIP Journal on audio, speech, and music processing : EURASIP JASMP VL - 1 IS - 1 SP - 1-15 PB - Springer CY - Heidelberg SN - 1687-4722 DO - 10.1186/s13636-024-00390-2 UR - https://fis.uni-bamberg.de/handle/uniba/111561 LA - eng AB - In everyday life, people experience different soundscapes in which natural sounds, animal noises, and man-made sounds blend together. Although there have been several studies on the importance of recurring sound patterns in music and language, the relevance of this phenomenon in natural soundscapes is still largely unexplored. In this article, we study the repetition patterns of harmonic and transient sound events as potential cues for acoustic scene classification (ASC). In the first part of our study, our aim is to identify acoustic scene classes that exhibit characteristic sound repetition patterns concerning harmonic and transient sounds. We propose three metrics to measure the overall prevalence of sound repetitions as well as their repetition periods and temporal stability. In the second part, we evaluate three strategies to incorporate self-similarity matrices as an additional input feature to a convolutional neural network architecture for ASC. We observe the characteristic repetition of transient sounds in recordings of “park” and “street traffic” as well as harmonic sound repetitions in acoustic scene classes related to public transportation. In the ASC experiments, hybrid network architectures, which combine spectrogram features and features from sound recurrence analysis, show increased accuracy for those classes with prominent sound repetition patterns. Our findings provide additional perspective on the distinctions among acoustic scenes previously primarily ascribed in the literature to their spectral features. KW - Acoustic scene classification KW - Sound recurrence analysis KW - Sound repetition patterns KW - Self-similarity matrix KW - Harmonic-percussive source separation KW - Result fusion KW - Ensemble models ER - TY - CPAPER TI - Towards Measuring and Forecasting Noise Exposure at the VELTINS-Arena in Gelsenkirchen, Germany AU - Ngamthipwatthana, Pitchapa AU - Götze, Marco AU - Kátai, András AU - Abeßer, Jakob PY - 2024 T2 - 2024 IEEE 5th International Symposium on the Internet of Sounds (IS2) : proceedings SP - 137-144 PB - IEEE DO - 10.1109/IS262782.2024.10704088 UR - https://fis.uni-bamberg.de/handle/uniba/111590 LA - eng ER - TY - CPAPER TI - Selbstüberwachtes Vortraining zur Verbesserung automatischer Audioklassifikationsalgorithmen AU - Grollmisch, Sascha AU - Abeßer, Jakob AU - Bös, Joachim PY - 2024 T2 - Fortschritte der Akustik - DAGA 2024". Tagungsband - Proceedings: DAGA 2024 - 50. Jahrestagung für Akustik, 18.-21. März 2024, Hannover SP - 1406-1409 CY - Berlin SN - 978-3-939296-22-5 UR - https://fis.uni-bamberg.de/handle/uniba/111589 LA - deu ER - TY - CPAPER TI - Towards Domain Shift in Location-Mismatch Scenarios for Bird Activity Detection AU - Latifi Bidarouni, Amir AU - Abeßer, Jakob PY - 2024 T2 - 32nd European Signal Processing Conference (EUSIPCO 2024) : proceedings SP - 1267-1271 PB - IEEE CY - Piscataway, NJ SN - 978-94-645936-1-7 DO - 10.23919/EUSIPCO63174.2024.10715313 UR - https://fis.uni-bamberg.de/handle/uniba/111587 LA - eng ER - TY - CPAPER TI - Visual and Audio Scene Classification for Detecting Discrepancies in Video : A Baseline Method and Experimental Protocol AU - Apostolidis, Konstantinos AU - Abeßer, Jakob AU - Cuccovillo, Luca AU - Mezaris, Vasileios PY - 2024 T2 - MAD '24: Proceedings of the 3rd ACM International Workshop on Multimedia AI against Disinformation SP - 30-36 PB - New York, NYUnited States CY - Association for Computing Machinery SN - 979-8-4007-0552-6 DO - 10.1145/3643491.3660287 UR - https://fis.uni-bamberg.de/handle/uniba/111586 LA - eng ER - TY - CPAPER TI - How Reliable are Posterior Class Probabilties in Automatic Music Classification? AU - Lukashevich, Hanna AU - Grollmisch, Sascha AU - Abeßer, Jakob AU - Stober, Sebastian AU - Bös, Joachim PY - 2023 T2 - AM '23: Proceedings of the 18th International Audio Mostly Conference SP - 45-50 PB - ACM SN - 979-8-4007-0818-3 DO - 10.1145/3616195.3616228 UR - https://fis.uni-bamberg.de/handle/uniba/111597 LA - eng AB - Music classification algorithms use signal processing and machine learning approaches to extract and enrich metadata for audio recordings in music archives. Common tasks include music genre classification, where each song is assigned a single label (such as Rock, Pop, or Jazz), and musical instrument classification. Since music metadata can be ambiguous, classification algorithms cannot always achieve fully accurate predictions. Therefore, our focus extends beyond the correctly estimated class labels to include realistic confidence values for each potential genre or instrument label. In practice, many state-of-the-art classification algorithms based on deep neural networks exhibit overconfident predictions, complicating the interpretation of the final output values. In this work, we examine whether the issue of overconfident predictions and, consequently, non-representative confidence values is also relevant to music genre classification and musical instrument classification. Moreover, we describe techniques to mitigate this behavior and assess the impact of deep ensembles and temperature scaling in generating more realistic confidence outputs, which can be directly employed in real-world music tagging applications. KW - music information retrieval KW - music classifcation KW - uncertainty KW - temperature scaling KW - deep ensembles ER - TY - CPAPER TI - Quantifying Uncertainty in Music Genre Classification AU - Lukashevich, Hanna AU - Grollmisch, Sascha AU - Abeßer, Jakob PY - 2023 T2 - Tagungsband, DAGA 2023 - 49. Jahrestagung für Akustik : 06.-09. März 2023, Hamburg SP - 1378-1381 CY - Berlin SN - 978-3-939296-21-8 UR - https://fis.uni-bamberg.de/handle/uniba/111596 LA - eng ER - TY - CPAPER TI - Uncertainty in Semi-Supervised Audio Classification : A Novel Extension for FixMatch AU - Grollmisch, Sascha AU - Cano, Estefanía AU - Lukashevich, Hanna AU - Abeßer, Jakob PY - 2023 T2 - 31st European Signal Processing Conference (EUSIPCO 2024) : proceedings SP - 161-165 PB - IEEE CY - Piscataway, NJ DO - 10.23919/eusipco58844.2023.10289789 UR - https://fis.uni-bamberg.de/handle/uniba/111594 LA - eng ER - TY - CPAPER TI - Unsupervised Feature-Space Domain Adaptation applied for Audio Classification AU - Bidarouni, Amir Latifi AU - Abeßer, Jakob PY - 2023 T2 - 4th International Symposium on the Internet of Sounds : Congress Center Le Benedettine, Pisa, 26-27 October 2023 SP - 1-7 PB - IEEE SN - 979-8-3503-8254-9 DO - 10.1109/ieeeconf59510.2023.10335455 UR - https://fis.uni-bamberg.de/handle/uniba/111593 LA - eng ER - TY - JOUR TI - Human and Machine Performance in Counting Sound Classes in Single-Channel Soundscapes AU - Abeßer, Jakob AU - Ullah, Asad AU - Ziegler, Sebastian AU - Grollmisch, Sascha PY - 2023 JO - Journal of the Audio Engineering Society VL - 71 IS - 12 SP - 860-872 PB - AES CY - New York, NY SN - 1549-4950 UR - https://fis.uni-bamberg.de/handle/uniba/111592 LA - eng ER - TY - JOUR TI - How Robust are Audio Embeddings for Polyphonic Sound Event Tagging? AU - Abeßer, Jakob AU - Grollmisch, Sascha AU - Müller, Meinard PY - 2023 JO - IEEE ACM transactions on audio, speech, and language processing : TASLP VL - 31 SP - 2658-2667 PB - IEEE CY - New York, NY SN - 2329-9290 DO - 10.1109/taslp.2023.3293032 UR - https://fis.uni-bamberg.de/handle/uniba/111591 LA - eng AB - Sound classification algorithms are challenged by the natural variability of everyday sounds, particularly for large sound class taxonomies. In order to be applicable in real-life environments, such algorithms must also be able to handle polyphonic scenarios, where simultaneously occurring and overlapping sound events need to be classified. With the rapid progress of deep learning, several deep audio embeddings (DAEs) have been proposed as pre-trained feature representations for sound classification. In this article, we analyze the embedding spaces of two non-trainable audio representations (NTARs) and five DAEs for sound classification in polyphonic scenarios (sound event tagging) and make several contributions. First, we compare general properties like the inter-correlation between feature dimensions and the scattering of sound classes in the embedding spaces. Second, we test the robustness of the embeddings against several audio degradations and propose two sensitivity measures based on a class-agnostic and a class-centric view on the resulting drift in the embedding space. Finally, as a central contribution, we study how a blending between pairs of sounds maps to embedding space trajectories and how the path of these trajectories can cause classification errors due to their proximity to other sound classes. Throughout our analyses, the PANN embeddings have shown the best overall performance for low-polyphony sound event tagging. KW - sound event tagging KW - sound polyphony KW - deep audio embeddings KW - embedding space ER - TY - CPAPER TI - Construction-sAIt : Multi-modal AI-driven technologies for construction site monitoring AU - Abeßer, Jakob AU - Loos, Alexander AU - Sharma, Prachi PY - 2022 T2 - Tagungsband, DAGA 2022 - 48. Jahrestagung für Akustik : 21.-24. März 2022, Stuttgart und Online SP - 90-93 CY - Berlin SN - 978-3-939296-20-1 UR - https://fis.uni-bamberg.de/handle/uniba/111761 LA - eng ER - TY - CPAPER TI - Analyzing Bird and Bat Activity in Agricultural Environments using AI-driven Audio Monitoring AU - Abeßer, Jakob AU - Wang, Xiaoyi AU - Bänsch, Svenja AU - Scherber, Christoph AU - Lukashevich, Hanna PY - 2022 T2 - Fortschritte der Akustik - DAGA 2022 : 48. Jahrestagung für Akustik, 21. - 24. März 2022, Stuttgart und Online SP - 123-126 CY - Berlin SN - 978-3-939296-20-1 UR - https://fis.uni-bamberg.de/handle/uniba/111762 LA - eng ER - TY - JOUR TI - JSD : A Dataset for Structure Analysis in Jazz Music AU - Balke, Stefan AU - Reck, Julian AU - Weiß, Christof AU - Abeßer, Jakob AU - Müller, Meinard PY - 2022 JO - Transactions of the International Society for Music Information Retrieval VL - 5 IS - 1 SP - 156-172 PB - Ubiquity Press, Ltd. CY - London SN - 2514-3298 DO - 10.5334/tismir.131 UR - https://fis.uni-bamberg.de/handle/uniba/111763 LA - eng AB - Given a music recording, music structure analysis aims at identifying important structural elements and segmenting the recording according to these elements. In jazz music, a performance is often structured by repeating harmonic schemata (known as choruses), which lay the foundation for improvisation by soloists. Within the fields of music information retrieval (MIR) and computational musicology, the Weimar Jazz Database (WJD) has turned out to be an extremely valuable resource for jazz research. Containing high-quality solo transcriptions for 456 solo sections, the dataset opened up new avenues for the understanding of creative processes in jazz improvisation using computational methods. In this paper, we complement this dataset by introducing the Jazz Structure Dataset (JSD), which provides annotations on structure and instrumentation of entire recordings. The JSD comprises 340 recordings with more than 3000 annotated segments, along with a segment-wise encoding of the solo and accompanying instruments. These annotations provide the basis for training, testing, and evaluating models for various important MIR tasks, including structure analysis, solo detection, or instrument recognition. As an example application, we consider the task of structure boundary detection. Based on a traditional novelty-based as well as a more recent data-driven approach using deep learning, we indicate the potential of the JSD while critically reflecting on some evaluation aspects of structure analysis. In this context, we also demonstrate how the JSD annotations and analysis results can be made accessible in a user-friendly way via web-based interfaces for data inspection and visualization. All annotations, experimental results, and code for reproducibility are made publicly available for research purposes. KW - Dataset KW - Music Structure Analysis KW - Boundary Detection KW - Instrument Recognition KW - Jazz ER - TY - CPAPER TI - Towards Interpreting and Improving the Latent Space for Musical Chord Recognition AU - Nadar, Christon R. AU - Taenzer, Michael AU - Abeßer, Jakob PY - 2022 T2 - Standing wave : ICMC 2022 : International Computer Music Conference, University of Limerick, Ireland, 2022 SP - 74-79 PB - International Computer Music Association, Inc. CY - San Francisco, California, USA SN - 978-0-9713192-6-4 UR - https://fis.uni-bamberg.de/handle/uniba/111765 LA - eng AB - Automatic chord recognition (ACR) naturally faces musical ambiguities between chord classes. These can be responsible for many misclassifications, especially in large chord vocabularies. In this paper, we propose a metric learning approach utilizing a triplet loss for the task of ACR in order to reduce chord ambiguities. In particular, we investigate how metric learning with different triplet sampling strategies re-aligns the distances between different chord classes in the latent space. Our main finding is that metric learning significantly improves the ACR performance for two taxonomies with five and nine chord classes. KW - Musical Chord Recognition ER - TY - CPAPER TI - Multi-Input Architecture and Disentangled Representation Learning for Multi-Dimensional Modeling of Music Similarity AU - Ribecky, Sebastian AU - Lukashevich, Hanna AU - Abeßer, Jakob PY - 2022 PB - AES CY - New York, NY UR - https://fis.uni-bamberg.de/handle/uniba/111766 LA - eng ER - TY - CPAPER TI - Classifying Sounds in Polyphonic Urban Sound Scenes AU - Abeßer, Jakob PY - 2022 PB - AES CY - New York, NY UR - https://fis.uni-bamberg.de/handle/uniba/111760 LA - eng ER - TY - CPAPER TI - Potentials and Challenges of AI-based Audio Analysis in Industrial Sound Analysis AU - Gourishetti, Saichand AU - Grollmisch, Sascha AU - Abeßer, Jakob AU - Liebetrau, Judith PY - 2022 T2 - Tagungsband, DAGA 2022 - 48. Jahrestagung für Akustik : 21.-24. März 2022, Stuttgart und Online SP - 94-97 CY - Berlin SN - 978-3-939296-20-1 UR - https://fis.uni-bamberg.de/handle/uniba/111764 LA - eng ER -