cs.SDNov 6, 2026

AudioSphere: Towards Self-Supervised Spatial Audio Representation Models

Authors: Goksenin Yuksel, Marcel van Gerven, Kiki van der Heijden

Organizations: Donders Institute, Radboud University, Nijmegen, the Netherlands · Mortimer B Zuckerman Institute, Columbia University, New York, United States

Abstract

Recent audio embedding models learn representations that transfer across clip- and frame-level tasks, yet they operate almost exclusively on single-channel signals and are blind to spatial structure. Spatial audio models capture that structure but rely on clip-level supervision or contrastive objectives, yielding embeddings unsuitable for frame-level tasks like sound event detection and localization (SELD). We argue this dichotomy is not fundamental: patch-level spatial self-supervision can jointly learn acoustic and spatial features at both temporal granularities. We introduce AudioSphere, a multi-channel masked autoencoder trained on Ambisonics recordings that reconstructs both spectral content and directional cues via a joint spatiotemporal masking strategy that masks both time-frequency patches and intensity vectors. AudioSphere supports frame-level predictions while producing strong clip-level representations via aggregation. Alongside the model, we release RealSELD, a standardized benchmark for evaluating frozen backbones on real-world multi-channel SELD. AudioSphere enables, for the first time, SELD with a frozen self-supervised backbone, while matching or exceeding prior self-supervised audio embedding models on clip-level classification tasks spanning environmental sound, speech, and music, and on frame-level sound event detection. All code and data are available at https://github.com/labhamlet/AudioSphere and https://huggingface.co/labhamlet.

Figures & tables

Appendix figures & tables16 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Sep 19, 2026

Bearings: Self-Supervised Soundfield Embeddings from First-Order Ambisonics

Recently proposed self-supervised audio encoders learn powerful general-purpose representations of sound scenes, yet they are spatially blind. To supply the missing spatial representation of sound scenes, we introduce Bearings. Bearings is a self-supervised framework that learns soundfield embeddings from unlabeled first-order Ambisonics. We pre-train a masked auto-encoder paired with a decoder conditioned on frozen acoustic embeddings from an off-the-shelf single-channel audio encoder. Our results show that the resulting soundfield embeddings form a reusable stream that can be attached to frozen acoustic encoders with a lightweight trainable fusion head. On sound event localization and detection, concatenating our soundfield embeddings with acoustic representations provides the missing spatial information and enables joint detection and localization, raising the location-dependent F-score from below 4 to 50 on TAU-NIGENS 2021 and 39 on STARSS23. To our knowledge, Bearings is the first self-supervised soundfield encoder whose embeddings plug into frozen acoustic encoders without retraining either model.
Oct 30, 2026

Self-Supervised Object-Centric Representation for Multi-Task Audio Analysis

This paper studies the capability of an object-centric audio representation to support multiple downstream tasks within a unified framework. The object-centric model is trained in a self-supervised setup using slot attention by reconstructing Audio Spectrogram Transformer (AST) patch embeddings from a set of latent slots. It is trained on synthetic audio mixtures, which provides controlled multi-source audio scenes. These slot representations are evaluated for the classification, detection, and separation of sound events by training task specific output layers. On the classification task, the model achieves an F1-score of 0.8, comparable to clip-level AST embeddings which achieve 0.7. For sound event detection, it reaches a macro F1-score of 0.7, outperforming clip-level AST embeddings at 0.4 and matching AST patch embeddings at 0.7, which encode finer time-frequency information. In source separation, it achieves an SI-SDR improvement (SI-SDRi) of 18.4 dB. Furthermore, in a zero-shot source separation setting, the model achieves an SI-SDRi of 14.5 dB. These results demonstrate that slot-based representations capture transferable and compositional structure in audio scenes, enabling generalization across tasks and datasets.
Oct 25, 2026

AudioFM: Towards General Purpose Audio Representation Learning

Learning audio representations that effectively bridge holistic global soundscapes with fine-grained acoustic moments is a long-standing goal in self-supervised learning. Recording-level contrastive objectives encourage agreement between pooled audio representations, but do not explicitly supervise local time-frequency representations. We introduce AudioFM, a self-supervised framework that combines recording-level contrast with patch-wise masked latent discrimination. Using visible acoustic context, a student predicts representations of masked patches and identifies their corresponding full-view teacher targets among candidates from other recordings. We further incorporate weaker supervision from eligible temporal neighbors to encourage local temporal consistency. Extensive evaluations show that AudioFM achieves state-of-the-art performance across diverse audio domains in linear probing, fine-tuning, and zero-shot transfer, demonstrating superior event detection, robust scene-level understanding, and strong domain generalizability.