Sound Event Detection and Localization (SELD) is a combined task of identifying sound events and their corresponding direction-of-arrival (DOA). While this task has numerous applications and has been extensively researched in recent years, it fails to provide full information about the sound source position. In this paper, we overcome this problem by extending the task to Sound Event Detection, Localization with Distance Estimation (3D SELD). We study two ways of integrating distance estimation within the SELD core - a multi-task approach, in which the problem is tackled by a separate model output, and a single-task approach obtained by extending the multi-ACCDOA method to include distance information. We investigate both methods for the Ambisonic and binaural versions of STARSS23: Sony-TAU Realistic Spatial Soundscapes 2023. Moreover, our study involves experiments on the loss function related to the distance estimation part. Our results show that it is possible to perform 3D SELD without any degradation of performance in sound event detection and DOA estimation.
In Self-Supervised Learning (SSL), Audio-Visual Correspondence (AVC) is a popular task to learn deep audio and video features from large unlabeled datasets. The key step in AVC is to randomly sample audio and video clips from the dataset and learn to minimize the feature distance between the positive pairs (corresponding audio-video pair) while maximizing the distance between the negative pairs (non-corresponding audio-video pairs). The learnt features are shown to be effective on various downstream tasks. However, these methods achieve subpar performance when the size of the dataset is rather small. In this paper, we investigate the effect of utilizing class label information in the AVC feature learning task. We modified various positive and negative data sampling techniques of SSL based on class label information to investigate the effect on the feature quality. We propose a new sampling approach which we call soft-positive sampling, where the positive pair for one audio sample is not from the exact corresponding video, but from a video of the same class. Experimental results suggest that when the dataset size is small in SSL setup, features learnt through the soft-positive sampling method significantly outperform those from the traditional SSL sampling approaches. This trend holds in both in-domain and out-of-domain downstream tasks, and even outperforms supervised classification. Finally, experiments show that class label information can easily be obtained using a publicly available classifier network and then can be used to boost the SSL performance without adding extra data annotation burden.
In this paper, we propose a method for class-incremental learning of potentially overlapping sounds for solving a sequence of multi-label audio classification tasks. We design an incremental learner that learns new classes independently of the old classes. To preserve knowledge about the old classes, we propose a cosine similarity-based distillation loss that minimizes discrepancy in the feature representations of subsequent learners, and use it along with a Kullback-Leibler divergence-based distillation loss that minimizes discrepancy in their respective outputs. Experiments are performed on a dataset with 50 sound classes, with an initial classification task containing 30 base classes and 4 incremental phases of 5 classes each. After each phase, the system is tested for multi-label classification with the entire set of classes learned so far. The proposed method obtains an average F1-score of 40.9% over the five phases, ranging from 45.2% in phase 0 on 30 classes, to 36.3% in phase 4 on 50 classes. Average performance degradation over incremental phases is only 0.7 percentage points from the initial F1-score of 45.2%.
Classification systems are normally trained by minimizing the cross-entropy between system outputs and reference labels, which makes the Kullback-Leibler divergence a natural choice for measuring how closely the system can follow the data. Precision and recall provide another perspective for measuring the performance of a classification system. Non-binary references can arise from various sources, and it is often beneficial to use the soft labels for training instead of the binarized data. However, the existing definitions for precision and recall require binary reference labels, and binarizing the data can cause erroneous interpretations. We present a novel method to calculate precision, recall and F-score without quantizing the data. The proposed metrics extend the well established metrics as the definitions coincide when used with binary labels. To understand the behavior of the metrics we show simple example cases and an evaluation of different sound event detection models trained on real data with soft labels.
In this paper, we propose a method for incremental learning of two distinct tasks over time: acoustic scene classification (ASC) and audio tagging (AT). We use a simple convolutional neural network (CNN) model as an incremental learner to solve the tasks. Generally, incremental learning methods catastrophically forget the previous task when sequentially trained on a new task. To alleviate this problem, we use independent learning and knowledge distillation (KD) between the timesteps in learning. Experiments are performed on TUT 2016/2017 dataset, containing 4 acoustic scene classes and 25 sound event classes. The proposed incremental learner solves the AT task with an F1 score of 54.4% and the ASC task with an accuracy of 88.9% in an incremental time step, outperforming a multi-task system which solves ASC and AT at the same time. The ASC task performance degrades only by 5.1% from the initial time ASC accuracy of 94.0%.
In this paper, we study the use of soft labels to train a system for sound event detection (SED). Soft labels can result from annotations which account for human uncertainty about categories, or emerge as a natural representation of multiple opinions in annotation. Converting annotations to hard labels results in unambiguous categories for training, at the cost of losing the details about the labels distribution. This work investigates how soft labels can be used, and what benefits they bring in training a SED system. The results show that the system is capable of learning information about the activity of the sounds which is reflected in the soft labels and is able to detect sounds that are missed in the typical binary target training setup. We also release a new dataset produced through crowdsourcing, containing temporally strong labels for sound events in real-life recordings, with both soft and hard labels.
In Self-Supervised Learning (SSL), various pretext tasks are designed for learning feature representations through contrastive loss. However, previous studies have shown that this loss is less tolerant to semantically similar samples due to the inherent defect of instance discrimination objectives, which may harm the quality of learned feature embeddings used in downstream tasks. To improve the discriminative ability of feature embeddings in SSL, we propose a new loss function called Angular Contrastive Loss (ACL), a linear combination of angular margin and contrastive loss. ACL improves contrastive learning by explicitly adding an angular margin between positive and negative augmented pairs in SSL. Experimental results show that using ACL for both supervised and unsupervised learning significantly improves performance. We validated our new loss function using the FSDnoisy18k dataset, where we achieved 73.6% and 77.1% accuracy in sound event classification using supervised and self-supervised learning, respectively.
Sound event detection (SED) and Acoustic scene classification (ASC) are two widely researched audio tasks that constitute an important part of research on acoustic scene analysis. Considering shared information between sound events and acoustic scenes, performing both tasks jointly is a natural part of a complex machine listening system. In this paper, we investigate the usefulness of several spatial audio features in training a joint deep neural network (DNN) model performing SED and ASC. Experiments are performed for two different datasets containing binaural recordings and synchronous sound event and acoustic scene labels to analyse the differences between performing SED and ASC separately or jointly. The presented results show that the use of specific binaural features, mainly the Generalized Cross Correlation with Phase Transform (GCC-phat) and sines and cosines of phase differences, result in a better performing model in both separate and joint tasks as compared with baseline methods based on logmel energies only.
This paper analyzes the outcome of the Low-Complexity Acoustic Scene Classification task in DCASE 2022 Challenge. The task is a continuation from the previous years. In this edition, the requirement for low-complexity solutions were modified including: a limit of 128 K on the number of parameters, including the zero-valued ones, imposed INT8 numerical format, and a limit of 30 million multiply-accumulate operations at inference time. The provided baseline system is a convolutional neural network which employs post-training quantization of parameters, resulting in 46512 parameters, and 29.23 million multiply-and-accumulate operations, well under the set limits of 128K and 30 million, respectively. The baseline system has a 42.9% accuracy and a log-loss of 1.575 on the development data consisting of audio from 9 different devices. An analysis of the submitted systems will be provided after the challenge deadline.
Learning from audio-visual data offers many possibilities to express correspondence between the audio and visual content, similar to the human perception that relates aural and visual information. In this work, we present a method for self-supervised representation learning based on audio-visual spatial alignment (AVSA), a more sophisticated alignment task than the audio-visual correspondence (AVC). In addition to the correspondence, AVSA also learns from the spatial location of acoustic and visual content. Based on 360$^\text{o}$ video and Ambisonics audio, we propose selection of visual objects using object detection, and beamforming of the audio signal towards the detected objects, attempting to learn the spatial alignment between objects and the sound they produce. We investigate the use of spatial audio features to represent the audio input, and different audio formats: Ambisonics, mono, and stereo. Experimental results show a 10 $\%$ improvement on AVSA for the first order ambisonics intensity vector (FOA-IV) in comparison with log-mel spectrogram features; the addition of object-oriented crops also brings significant performance increases for the human action recognition downstream task. A number of audio-only downstream tasks are devised for testing the effectiveness of the learnt audio feature representation, obtaining performance comparable to state-of-the-art methods on acoustic scene classification from ambisonic and binaural audio.