@techreport{Berg_LU_task3_report,
    Author = "Berg, Axel and Engman, Johanna and Gulin, Jens and Astrom, Karl and Oskarsson, Magnus",
    title = "THE LU SYSTEM FOR DCASE 2024 SOUND EVENT LOCALIZATION AND DETECTION CHALLENGE",
    institution = "DCASE2024 Challenge",
    year = "2024",
    month = "June",
    abstract = "This technical report gives an overview of our submission to task 3 of the DCASE 2024 challenge. We present a sound event localization and detection (SELD) system using input features based on trainable neural generalized cross-correlations with phase transform (NGCC-PHAT). With these features together with spectrograms as input to a Transformer-based network, we achieve significant improvements over the baseline method. In addition, we also present an audio-visual version of our system, where distance predictions are updated using depth maps from the panorama video frames."
}

@techreport{Berghi_SURREY_task3b_report,
    Author = "Berghi, Davide and Jackson, Philip J. B.",
    title = "LEVERAGING REVERBERATION AND VISUAL DEPTH CUES FOR SOUND EVENT LOCALIZATION AND DETECTION WITH DISTANCE ESTIMATION",
    institution = "DCASE2024 Challenge",
    year = "2024",
    month = "June",
    abstract = "This report describes our systems submitted for the DCASE2024 Task 3 challenge: Audio and Audiovisual Sound Event Localization and Detection with Source Distance Estimation (Track B). Our main model is based on the audio-visual (AV) Conformer, which processes video and audio embeddings extracted with ResNet50 and with an audio encoder pre-trained on SELD, respectively. This model outperformed the audio-visual baseline of the development set of the STARSS23 dataset by a wide margin, halving its DOAE and improving the F1 by more than 3x. Our second system performs a temporal ensemble from the outputs of the AV-Conformer. We then extended the model with features for distance estimation, such as direct and reverberant signal components extracted from the omnidirectional audio channel, and depth maps extracted from the video frames. While the new system improved the RDE of our previous model by about 3 percentage points, it achieved a lower F1 score. This may be caused by sound classes that rarely appear in the training set and that the more complex system does not detect, as analysis can determine. To overcome this problem, our fourth and final system consists of an ensemble strategy combining the predictions of the other three. Many opportunities to refine the system and training strategy can be tested in future ablation experiments, and likely achieve incremental performance gains for this audio-visual task."
}

@techreport{Chen_ECUST_task3_report,
    Author = "Tao, Jishen and Chen, Ning",
    title = "FEATURE FUSION BASED ON CROSS-FEATURE TRANSFORMER FOR SOUND EVENT LOCALIZATION AND DETECTION WITH SOURCE DISTANCE ESTIMATION",
    institution = "DCASE2024 Challenge",
    year = "2024",
    month = "June",
    abstract = "Since the audio of many sound events contains rich high-frequency components, the Log-Mel, which compacts the high-frequency components seriously, cannot represent the essential feature of sound event entirely. In this paper, the Log-Mel Spectrogram + Intensity Vector (LMSIV) and Magnitude Spectrogram (MS) are fused to solve this problem. First, the Cross-Feature Transformer (CFT) is performed on each feature to inspire the other feature to reinforce itself through directly attending to latent relevance revealed in the other feature to fuse the features while ensuring awareness of their interaction introduced. Then Self-Attention Transformer (SAT) is performed on the concatenation of the obtained embeddings to further prioritize contextual information in it. The experimental results show that our proposed system outperform the baseline system on the development dataset of DCASE 2024 task3."
}

@techreport{Du_NERCSLIP_task3_report,
    Author = "Wang, Qing and Dong, Yuxuan and Hong, Hengyi and Wei, Ruoyu and Hu, Maocheng and Cheng, Shi and Jiang, Ya and Cai, Mingqi and Fang, Xin and Du, Jun",
    title = "THE NERC-SLIP SYSTEM FOR SOUND EVENT LOCALIZATION AND DETECTION WITH SOURCE DISTANCE ESTIMATION OF DCASE 2024 CHALLENGE",
    institution = "DCASE2024 Challenge",
    year = "2024",
    month = "June",
    abstract = "The technical report presents our submission system for Task 3 of the DCASE 2024 Challenge: Audio and Audiovisual Sound Event Localization and Detection (SELD) with Source Distance Estimation (SDE). In addition to direction of arrival estimation (DOAE) of the sound source, this challenge also requires predicting the source distance. We attempted three methods to enable the system to predict both the DOA and the distance of the sound source. First, we proposed two multi-task learning frameworks. One introduces an extra branch to the original SELD model with multi-task learning framework, resulting in a three-branch output to simultaneously predict the DOA and distance of the sound source. The other integrates the sound source distance into the DOA prediction, estimat- ing the absolute position of the sound source. Second, we trained two models for DOAE and SDE respectively, and then used a joint prediction method based on the outputs of the two models. For the audiovisual SELD task with SDE, we used a ResNet-50 model pretrained on ImageNet as the visual feature extractor. Additionally, we simulated audio-visual data and used a teacher-student learning method to train our multi-modal system. We evaluated our methods on the dev-test set of the Sony-TAu Realistic Spatial Soundscapes 2023 (STARSS23) dataset."
}

@techreport{Guan_CQUPT_task3_report,
    Author = "Guan, Xin and Zhou, Yi and Liu, Hongqing and Cao, Yin",
    title = "POWER CUE ENHANCED NETWORK AND AUDIO-VISUAL FUISON FOR SOUND EVENT LOCALIZATION AND DETECTION OF DCASE2024 CHALLENGE",
    institution = "DCASE2024 Challenge",
    year = "2024",
    month = "June",
    abstract = "This technical report describes our submission systems for Task 3 of the DCASE2024 challenge: Sound Event Localization and Detection (SELD) Evaluated in Real Spatial Sound Scenes. To address the audio-only SELD task, we utilize a Resnet-Conformer as the main network. Additionally, we introduce a branch to receive power cue features, specifically log root mean square (log-rms). We employ various data augmentation techniques, including audio channel swapping (ACS), random cutout, time-frequency masking, frequency shifting, and AugMix, to enhance the model’s generalization. For the audio-visual SELD task, we also augment the visual modality in alignment with ACS. The audio and visual embeddings are sent to parallel Cross-Modal Attentive Fusion (CMAF) blocks before concatenation. We evaluate our approach on the dev-test set of the Sony-TAu Realistic Spatial Soundscapes 2023 (STARSS23) dataset."
}

@techreport{Li_BIT_task3a_report,
    Author = "Li, Jiahao",
    title = "THE SYSTEM USING CONVNEXT, CONFORMER, AND DATA AUGMENTATION FOR SOUND EVENT LOCALIZATION AND DETECTION",
    institution = "DCASE2024 Challenge",
    year = "2024",
    month = "June",
    abstract = "This technical report details our submission system for DCASE2024 Task 3: Audio and Audiovisual Sound Event Localization and Detection (SELD) with Source Distance Estimation. To address the audio-only task, we initially apply the Audio Channel Swapping (ACS) method to generate augmented data, enhancing the performance of the proposed system. Subsequently, we introduce the ConvNeXt module for feature extraction and processing. To further enhance feature extraction capabilities, we employ the Squeeze-and-Excitation Block (SEBlock) after ConvNeXt. We then utilize the Conformer to extract additional features and ultimately compute the multi-ACCDOA output. The proposed system significantly outperforms the baseline on the development dataset of DCASE2024 Task 3."
}

@techreport{Li_SHU_task3b_report,
    Author = "Li, Yongbo and Wang, Chuan and Huang, Qinghua",
    title = "Data Augmentation and Cross-Fusion for Audiovisual Sound Event Localization and Detection with Source Distance Estimation",
    institution = "DCASE2024 Challenge",
    year = "2024",
    month = "June",
    abstract = "This technical report describes a system participating in the DCASE2024 challenge Task 3: Sound Event Localization and Detection with Source Distance Estimation-Track B: Audio-Visual Reasoning. A system based on the official baseline system is developed and improved in terms of network architecture and data augmentation. The convolutional recurrent neural network (CRNN) is substituted by a ResNet-Conformer block pre-trained on an audio-only network. Audio Channel Swapping (ACS) is applied to the DCASE 2024 official audio dataset to generate more audio data. A simulated audio dataset is also created. Video Pixel Swapping (VPS) is performed on the original video data to obtain more video data. Experimental results show that our system outperforms the baseline method on the Sony-TAU Real Spatial Soundscape 2024 (STARSS24) development dataset. A series of experiments are implemented only on the First-Order Ambisonics (FOA) dataset."
}

@inproceedings{Politis_TAU_task3a_report,
    author = "Politis, Archontis and Shimada, Kazuki and Sudarsanam, Parthasaarathy and Adavanne, Sharath and Krause, Daniel and Koyama, Yuichiro and Takahashi, Naoya and Takahashi, Shusuke and Mitsufuji, Yuki and Virtanen, Tuomas",
    title = "{STARSS22}: {A} dataset of spatial recordings of real scenes with spatiotemporal annotations of sound events",
    booktitle = "Proceedings of the 8th Detection and Classification of Acoustic Scenes and Events 2022 Workshop (DCASE2022)",
    institution = "DCASE2023 Challenge",
    address = "Nancy, France",
    month = "November",
    year = "2022",
    pages = "125--129",
    abstract = "This report presents the Sony-TAu Realistic Spatial Soundscapes 2022 (STARSS22) dataset of spatial recordings of real sound scenes collected in various interiors at two different sites. The dataset is captured with a high resolution spherical microphone array and delivered in two 4-channel formats, first-order Ambisonics and tetrahedral microphone array. Sound events belonging to 13 target classes are annotated both temporally and spatially through a combination of human annotation and optical tracking. STARSS22 serves as the development and evaluation dataset for Task 3 (Sound Event Localization and Detection) of the DCASE2022 Challenge and it introduces significant new challenges with regard to the previous iterations, which were based on synthetic data. Additionally, the report introduces the baseline system that accompanies the dataset with emphasis on its differences to the baseline of the previous challenge. Baseline results indicate that with a suitable training strategy a reasonable detection and localization performance can be achieved on real sound scene recordings. The dataset is available in https://zenodo.org/record/6600531.",
    url = "https://dcase.community/workshop2022/proceedings"
}

@techreport{Qian_IASP_task3a_report,
    Author = "Qian, Yuanhang and Zheng, Tianqin and Zeng, Yichen and Huang, Gongping",
    title = "THE IASP SUBMISSION FOR SOUND EVENT LOCALIZATION AND DETECTION OF DCASE2024 CHALLENGE",
    institution = "DCASE2024 Challenge",
    year = "2024",
    month = "June",
    abstract = "The technical report describes the submission systems developed for task 3a of the DCASE2024 challenge: Audio Sound Event Localization and Detection with Source Distance Estimation. To enhance the performance of the audio-only task, we implement audio channel swapping as a data augmentation technique. We adopt the Resnet-Conformer model for the network architecture, which is well-suited for capturing First-Order Ambisonics (FOA) format data patterns. Additionally, the approach utilizes the Multi- ACCDDOA method to concurrently predict the event type and estimate the source distance. This comprehensive strategy yielded superior results compared to the baseline system."
}

@inproceedings{Shimada_SONY_task3b_report,
    author = "Shimada, Kazuki and Politis, Archontis and Sudarsanam, Parthasaarathy and Krause, Daniel and Uchida, Kengo and Adavanne, Sharath and Hakala, Aapo and Koyama, Yuichiro and Takahashi, Naoya and Takahashi, Shusuke and Virtanen, Tuomas and Mitsufuji, Yuki",
    title = "{STARSS23}: {A}n Audio-Visual Dataset of Spatial Recordings of Real Scenes with Spatiotemporal Annotations of Sound Events",
    booktitle = "Proceedings of the 37th International Conference on Neural Information Processing Systems (NeurIPS 2023)",
    address = "New Orleans, USA",
    month = "December",
    pages = "72931--72957",
    abstract = "While direction of arrival (DOA) of sound events is generally estimated from multichannel audio data recorded in a microphone array, sound events usually derive from visually perceptible source objects, e.g., sounds of footsteps come from the feet of a walker. This paper proposes an audio-visual sound event localization and detection (SELD) task, which uses multichannel audio and video information to estimate the temporal activation and DOA of target sound events. Audio-visual SELD systems can detect and localize sound events using signals from a microphone array and audio-visual correspondence. We also introduce an audio-visual dataset, Sony-TAu Realistic Spatial Soundscapes 2023 (STARSS23), which consists of multichannel audio data recorded with a microphone array, video data, and spatiotemporal annotation of sound events. Sound scenes in STARSS23 are recorded with instructions, which guide recording participants to ensure adequate activity and occurrences of sound events. STARSS23 also serves human-annotated temporal activation labels and human-confirmed DOA labels, which are based on tracking results of a motion capture system. Our benchmark results show that the audio-visual SELD system achieves lower localization error than the audio-only system. The data is available at https://zenodo.org/record/7880637.",
    year = "2023",
    url = "https://proceedings.neurips.cc/paper\_files/paper/2023/hash/e6c9671ed3b3106b71cafda3ba225c1a-Abstract-Datasets\_and\_Benchmarks.html"
}

@techreport{Sun_JLESS_task3a_report,
    Author = "Sun, Wenqiang and Zhang, Dongzhe and Bai, Jisheng and Chen, Jianfeng",
    title = "JLESS SUBMISSION TO DCASE2024 TASK3: Conformer with Data Augmentation for Sound Event Localization and Detection with Source Distance Estimation",
    institution = "DCASE2024 Challenge",
    year = "2024",
    month = "June",
    abstract = "This technical report, we describe our proposed system for DCASE2024 task3: Sound Event Localization and Detec tion(SELD) with Source Distance Estimation in Real Spatial Sound Scenes. At first, we review the famous deep learning methods in SELD. To augment our dataset, we employ channel rotation techniques. In addition to existing features, we introduce a novel feature: the sine value of the inter-channel phase difference. Finally, we validate the effectiveness of our approach on the Sony-TAU Realistic Spatial Soundscapes 2023 (STARSS23) dataset and the results demonstrate that our method outperforms the baseline across multiple metrics."
}

@techreport{Vo_DU_task3a_report,
    Author = "Vo, Quoc Thinh and Han, David K.",
    title = "RESNET-CONFORMER NETWORK WITH SHARED WEIGHTS AND ATTENTION MECHANISM FOR SOUND EVENT LOCALIZATION, DETECTION, AND DISTANCE ESTIMATION",
    institution = "DCASE2024 Challenge",
    year = "2024",
    month = "June",
    abstract = "This technical report outlines our approach to Task 3A of the Detection and Classification of Acoustic Scenes and Events (DCASE) 2024, focusing on Sound Event Localization and Detection (SELD). SELD provides valuable insights by estimating sound event localization and detection, aiding in various machine cognition tasks such as environmental inference, navigation, and other sound localization-related applications. This year’s challenge evaluates models using either audio-only (Track A) or audiovisual (Track B) inputs on annotated recordings of real sound scenes. A notable change this year is the introduction of distance estimation, with evaluation metrics adjusted accordingly for a comprehensive assessment. Our submission is for Task A of the Challenge, which focuses on the audio-only track. Our approach utilizes log-mel spectrograms, intensity vectors, and employs multiple data augmentations. We proposed an EINV2-based network architecture, achieving improved results: an F-score of 40.2\%, Angular Error (DOA) of 17.7°, and Relative Distance Error (RDE) of 0.32 on the test set of the Development Dataset."
}

@techreport{Yeow_NTU_task3a_report,
    Author = "Yeow, Jun Wei and Tan, Ee-Leng and Bai, Jisheng and Peksi, Santi and Gan, Woon-Seng",
    title = "SQUEEZE-AND-EXCITE RESNET-CONFORMERS FOR SOUND EVENT LOCALIZATION, DETECTION, AND DISTANCE ESTIMATION FOR DCASE2024 CHALLENGE",
    institution = "DCASE2024 Challenge",
    year = "2024",
    month = "June",
    abstract = "This technical report details our systems submitted for Task 3 of the DCASE 2024 Challenge: Audio and Audiovisual Sound Event Localization and Detection (SELD) with Source Distance Estimation (SDE). We address only the audio-only SELD with SDE (SELDDE) task in this report. We propose to improve the existing ResNet- Conformer architectures with Squeeze-and-Excitation blocks in order to introduce additional forms of channel- and spatial-wise attention. In order to improve SELD performance, we also utilize the Spatial Cue-Augmented Log-Spectrogram (SALSA) features over the commonly used log-mel spectra features for polyphonic SELD. We complement the existing Sony-TAu Realistic Spatial Soundscapes 2023 (STARSS23) dataset with the audio channel swapping technique and synthesize additional data using the SpatialScaper generator. We also perform distance scaling in order to prevent large distance errors from contributing more towards the loss function. Finally, we evaluate our approach on the evaluation subset of the STARSS23 dataset."
}

@techreport{Yu_HYUNDAI_task3a_report,
    Author = "Yu, Hogeon",
    title = "DOA AND EVENT GUIDANCE SYSTEM FOR SOUND EVENT LOCALIZATION AND DETECTION WITH SOURCE DISTANCE ESTIMATION",
    institution = "DCASE2024 Challenge",
    year = "2024",
    month = "June",
    abstract = "This technical report describes the proposed system submitted to the DCASE2024 Task3: Sound Event Localization and Detection with Source Distance Estimation. There are two tracks, and we participate in the audio-only track. At first, we adopt the CST block, a transformer-based network, to extract meaningful features for predicting sub-tasks DOA and SED. Next, DOA and EVENT guidance attention blocks are introduced to boost the performance on a Multi-ACCDOA-based single-task system for the SELD tasks. We only apply the data augmentation method, a multi-channel simulation technique to complement the sparsity of training data provided by the challenge. Tested on the dev-test set of the Sony-TAu Realistic Spatial Soundscapes 2023 (STARSS23) dataset, our proposed systems outperform the baseline system."
}

@techreport{Zhang_BUPT_task3a_report,
    Author = "Mu, Da and Sun, Huamei and Yue, Haobo and Jiang, Yuanyuan and Wang, Zehao and Zhang, Zhicheng and Yin, Jianqin",
    title = "MULTI-SCALE FEATURE FUSION FOR SOUND EVENT LOCALIZATION AND DETECTION",
    institution = "DCASE2024 Challenge",
    year = "2024",
    month = "June",
    abstract = "This technical report describes our submission system for task 3 of the DCASE2024 challenge: Sound Event Localization and Detection with Source Distance Estimation. Our experiment specifically focused on analyzing the first-order ambisonics (FOA) dataset. Building upon our previous work, we utilized a three-stage network structure known as the Multi-scale Feature Fusion (MFF) module. This module allowed us to efficiently extract multi-scale features across the spectral, spatial, and temporal domains. In this report, we introduce the implementation of the MFF module as the encoder and Conformer Blocks as the decoder within a single-branch neu- ral network named MFF-Conformer. This configuration enables us to generate Multi-ACCDOA labels as the output. Compared to the baseline system, our approach exhibits significant improvements in F20° and DOAE metrics and demonstrates its effectiveness on the development dataset of DCASE task 3."
}