This paper examines audio self-supervised learning through the alignment between pretraining objectives, architectural inductive biases, and downstream applications, and relates these demands to the biases of CNNs, recurrent and State Space Models, Transformers, and hybrid architectures.
Abstract
This paper examines audio self-supervised learning (SSL) through the alignment between pretraining objectives, architectural inductive biases, and downstream applications. Rather than treating SSL methods as a chronological sequence of pretext tasks or model families, we ask how different supervisory signals shape the representations that models are expected to learn. The discussion is organized around five paradigms: auxiliary tasks, contrastive learning, generative reconstruction, discrete token prediction, and multimodal alignment. These objectives place different demands on the model, from local structural sensitivity and contrastive invariance to contextual inference, discrete semantic abstraction, and multimodal grounding. We relate these demands to the biases of CNNs, recurrent and State Space Models, Transformers, and hybrid architectures, showing how local acoustic compression, sequential state propagation, content-dependent global routing, and local--global integration support different forms of audio SSL. The same view is then used to interpret downstream applications in speech processing, environmental sound analysis, music information retrieval, medical and bioacoustic analysis, and multimodal audio understanding as practical tests of whether learned representations and architectural choices generalize across domains. We also review benchmark protocols and open challenges, including tokenization bottlenecks, long-context efficiency, robustness, and secure multimodal deployment, and discuss how codec-based tokenization and audio-language modeling extend this objective--architecture--application pipeline. The accompanying repository is released at https://github.com/colaudiolab/Awesome-Self-Supervised-Audio-Learning.
Neural audio representation learning has enabled a range of content-oriented applications, but the resulting features remain limited for tasks involving audio processing. Furthermore, it is not obvious what processing-aware representations should capture: the processing itself, abstracted away from source content, or the processed audio that retains it. Existing approaches implicitly commit to one or the other and also differ in their models, data, and evaluation, obscuring which design choices drive their behavior. We address both questions within a unified framework of three objectives: processing consistency, description alignment, and equivariance via forward prediction. We compare all combinations of the objectives under a controlled setup and reveal their relative strengths and interactions. Our framework produces both a transformation embedding and a processed-audio embedding, and we find that the two play complementary roles: distance-based tasks favor the former, while probe-based tasks favor the latter. Combined with improvements in network architecture and training pipeline, our representations outperform prior baselines across retrieval, probe-based evaluation, and style transfer.
Sungho Lee, Marco A. Mart'inez-Ram'irez, Junghyun Koo et al.· 0 citations
NAPE (Next-Audio-Patch-Embedding prediction), a self-supervised framework in which a causal Transformer predicts each next patch embedding of a log-mel spectrogram from the previous ones, using causal masking and stop-gradient as its sole training signal is introduced.
Umberto Cappellazzo, Xubo Liu, Stavros Petridis et al.· 0 citations
Self-supervised learning (SSL) models, such as Wav2Vec2, HuBERT, and WavLM, have become foundational across a wide range of speech and audio tasks. Despite their success, understanding their internal layer-wise dynamics remains an ongoing challenge. To address this, we propose a two-part model-centric framework called InsideSSL. First, we establish a task-agnostic analysis from three intrinsic per-layer perspectives: compression (entropy), geometry (curvature), and robustness to perturbations. We show that varying training objectives induce distinct regimes of acoustic compression and manifold unfolding. Second, we introduce the cross-layer Generative Compatibility Matrix (GCM) to evaluate functional transferability, exposing stable phonetic cores, identity volatility, and deep-layer semantic pruning. In addition to these evaluations, linear probing connects the model-centric perspective to downstream tasks, demonstrating how layer topology dictates phoneme, pitch, and speaker encoding.
A multi-layer feature fusion (MLF) adapter that aggregates information from all encoder layers before projecting them into the language model is proposed and shows that MICL does not emerge naturally in ALLMs, but can be effectively acquired through targeted contextual biasing training.
Xiaoyu Yang, Xuenan Xu, Wenyi Yu et al.· 0 citations
We present DINO-A, an adaptation of self-distillation from vision to general audio representation learning. While DINO has become a canonical method in self-supervised vision and prior audio work has explored latent prediction (BYOL-A) and masked modeling (Audio-MAE, BEATs), no prior work has brought canonical DINO to general audio classification in the way BYOL-A brought BYOL. DINO-A retains DINO's multi-crop, EMA teacher, and high-dimensional projection, replacing only the input modality and augmentations with log-mel spectrograms and the BYOL-A v2 augmentation block. We pretrain three backbones, two Vision Transformers with 8x8 and 16x16 patches and a convolutional encoder, on FSD50K and evaluate them with linear probing on ESC-50, Speech Commands v2, UrbanSound8K, and GTZAN. Three findings characterize the resulting representations. Patch resolution within the Vision Transformer family has consistent effect on representation quality, with smaller patches winning across all four tasks. The choice between Vision Transformer and convolutional backbone interacts with task type: convolutional networks lead on speech while Vision Transformers lead on environmental sounds and music. Under identical pretraining and evaluation conditions, DINO-A and BYOL-A v2 differ by 11.96 percentage points on average, and we trace this difference to two mechanisms: the interaction between DINO's high-dimensional projection space and FSD50K's limited scale, and the additional cost of multi-crop augmentation, which DINO uses but BYOL-A v2 does not. The high-dimensional projection space, central to DINO's success in vision, becomes a liability at FSD50K scale.
Tomasz Radzikowski, M. Modrzejewski, Przemyslaw Rokita· 0 citations
Compared to prior CLIP-enhancement methods, MLLMCLIP achieves state-of-the-art compositional accuracy while delivering consistent gains on standard zero-shot classification and image-text retrieval, showing that feature-level distillation strengthens both compositional and general vision-language representation capability.
Jongsuk Kim, Qiyu Wu, Zhuoyuan Mao et al.· 0 citations