Florian Schmid

Florian Schmid, Khaled Koutini, Gerhard Widmer

Audio Spectrogram Transformer models rule the field of Audio Tagging, outrunning previously dominating Convolutional Neural Networks (CNNs). Their superiority is based on the ability to scale up and exploit large-scale datasets such as AudioSet. However, Transformers are demanding in terms of model size and computational requirements compared to CNNs. We propose a training procedure for efficient CNNs based on offline Knowledge Distillation (KD) from high-performing yet complex transformers. The proposed training schema and the efficient CNN design based on MobileNetV3 results in models outperforming previous solutions in terms of parameter and computational efficiency and prediction performance. We provide models of different complexity levels, scaling from low-complexity models up to a new state-of-the-art performance of .483 mAP on AudioSet. Source Code available at: https://github.com/fschmid56/EfficientAT

Khaled Koutini, Shahed Masoudian, Florian Schmid et al.

The success of supervised deep learning methods is largely due to their ability to learn relevant features from raw data. Deep Neural Networks (DNNs) trained on large-scale datasets are capable of capturing a diverse set of features, and learning a representation that can generalize onto unseen tasks and datasets that are from the same domain. Hence, these models can be used as powerful feature extractors, in combination with shallower models as classifiers, for smaller tasks and datasets where the amount of training data is insufficient for learning an end-to-end model from scratch. During the past years, Convolutional Neural Networks (CNNs) have largely been the method of choice for audio processing. However, recently attention-based transformer models have demonstrated great potential in supervised settings, outperforming CNNs. In this work, we investigate the use of audio transformers trained on large-scale datasets to learn general-purpose representations. We study how the different setups in these audio transformers affect the quality of their embeddings. We experiment with the models' time resolution, extracted embedding level, and receptive fields in order to see how they affect performance on a variety of tasks and datasets, following the HEAR 2021 NeurIPS challenge evaluation setup. Our results show that representations extracted by audio transformers outperform CNN representations. Furthermore, we will show that transformers trained on Audioset can be extremely effective representation extractors for a wide range of downstream tasks.

Florian Schmid, Khaled Koutini, Gerhard Widmer

The introduction of large-scale audio datasets, such as AudioSet, paved the way for Transformers to conquer the audio domain and replace CNNs as the state-of-the-art neural network architecture for many tasks. Audio Spectrogram Transformers are excellent at exploiting large datasets, creating powerful pre-trained models that surpass CNNs when fine-tuned on downstream tasks. However, current popular Audio Spectrogram Transformers are demanding in terms of computational complexity compared to CNNs. Recently, we have shown that, by employing Transformer-to-CNN Knowledge Distillation, efficient CNNs can catch up with and even outperform Transformers on large datasets. In this work, we extend this line of research and increase the capacity of efficient CNNs by introducing dynamic CNN blocks, constructed of dynamic non-linearities, dynamic convolutions and attention mechanisms. We show that these dynamic CNNs outperform traditional efficient CNNs, in terms of the performance-complexity trade-off and parameter efficiency, at the task of audio tagging on the large-scale AudioSet. Our experiments further indicate that the introduced dynamic CNNs achieve better performance on downstream tasks and scale up well, attaining Transformer performance and even outperforming them on AudioSet and several downstream tasks.

Florian Schmid, Paul Primus, Toni Heittola et al.

This paper presents the Low-Complexity Acoustic Scene Classification with Device Information Task of the DCASE 2025 Challenge, along with its baseline system. Continuing the focus on low-complexity models, data efficiency, and device mismatch from previous editions (2022-2024), this year's task introduces a key change: recording device information is now provided at inference time. This enables the development of device-specific models that leverage device characteristics-reflecting real-world deployment scenarios in which a model is designed with awareness of the underlying hardware. The training set matches the 25% subset used in the corresponding DCASE 2024 challenge, with no restrictions on external data use, highlighting transfer learning as a central topic. The baseline achieves 50.72% accuracy with a device-agnostic model, improving to 51.89% when incorporating device-specific fine-tuning. The task attracted 31 submissions from 12 teams, with 11 teams outperforming the baseline. The top-performing submission achieved an accuracy gain of more than 8 percentage points over the baseline on the evaluation set.

Paul Primus, Florian Schmid, Gerhard Widmer

Dual-encoder-based audio retrieval systems are commonly optimized with contrastive learning on a set of matching and mismatching audio-caption pairs. This leads to a shared embedding space in which corresponding items from the two modalities end up close together. Since audio-caption datasets typically only contain matching pairs of recordings and descriptions, it has become common practice to create mismatching pairs by pairing the audio with a caption randomly drawn from the dataset. This is not ideal because the randomly sampled caption could, just by chance, partly or entirely describe the audio recording. However, correspondence information for all possible pairs is costly to annotate and thus typically unavailable; we, therefore, suggest substituting it with estimated correspondences. To this end, we propose a two-staged training procedure in which multiple retrieval models are first trained as usual, i.e., without estimated correspondences. In the second stage, the audio-caption correspondences predicted by these models then serve as prediction targets. We evaluate our method on the ClothoV2 and the AudioCaps benchmark and show that it improves retrieval performance, even in a restricting self-distillation setting where a single model generates and then learns from the estimated correspondences. We further show that our method outperforms the current state of the art by 1.6 pp. mAP@10 on the ClothoV2 benchmark.

Tobias Morocutti, Jonathan Greif, Paul Primus et al.

Spatial semantic segmentation of sound scenes (S5) involves the accurate identification of active sound classes and the precise separation of their sources from complex acoustic mixtures. Conventional systems rely on a two-stage pipeline - audio tagging followed by label-conditioned source separation - but are often constrained by the absence of fine-grained temporal information critical for effective separation. In this work, we address this limitation by introducing a novel approach for S5 that enhances the synergy between the event detection and source separation stages. Our key contributions are threefold. First, we fine-tune a pre-trained Transformer to detect active sound classes. Second, we utilize a separate instance of this fine-tuned Transformer to perform sound event detection (SED), providing the separation module with detailed, time-varying guidance. Third, we implement an iterative refinement mechanism that progressively enhances separation quality by recursively reusing the separator's output from previous iterations. These advancements lead to significant improvements in both audio tagging and source separation performance, as demonstrated by our system's second-place finish in Task 4 of the DCASE Challenge 2025. Our implementation and model checkpoints are available in our GitHub repository: https://github.com/theMoro/dcase25task4 .

Tobias Morocutti, Florian Schmid, Jonathan Greif et al.

We target the problem of developing new low-complexity networks for the sound event detection task. Our goal is to meticulously analyze the performance-complexity trade-off, aiming to be competitive with the large state-of-the-art models, at a fraction of the computational requirements. We find that low-complexity convolutional models previously proposed for audio tagging can be effectively adapted for event detection (which requires frame-wise prediction) by adjusting convolutional strides, removing the global pooling, and, importantly, adding a sequence model before the (now frame-wise) classification heads. Systematic experiments reveal that the best choice for the sequence model type depends on which complexity metric is most important for the given application. We also investigate the impact of enhanced training strategies such as knowledge distillation. In the end, we show that combined with an optimized training strategy, we can reach event detection performance comparable to state-of-the-art transformers while requiring only around 5% of the parameters. We release all our pre-trained models and the code for reproducing this work to support future research in low-complexity sound event detection at https://github.com/theMoro/EfficientSED.

Paul Primus, Florian Schmid, Gerhard Widmer

Learning to associate audio with textual descriptions is valuable for a range of tasks, including pretraining, zero-shot classification, audio retrieval, audio captioning, and text-conditioned audio generation. Existing contrastive language-audio pretrained models are typically trained using global, clip-level descriptions, which provide only weak temporal supervision. We hypothesize that CLAP-like language-audio models - particularly, if they are expected to produce frame-level embeddings - can benefit from a stronger temporal supervision. To confirm our hypothesis, we curate a novel dataset of approximately 12,000 audio recordings from Freesound, each annotated with single-sentence free-text descriptions linked to a specific temporal segment in an audio recording. We use large language models to clean these annotations by removing references to non-audible events, transcribed speech, typos, and annotator language bias. We further propose a frame-wise contrastive training strategy that learns to align text descriptions with temporal regions in an audio recording and demonstrate that our model has better temporal text-audio alignment abilities compared to models trained only on global captions when evaluated on the AudioSet Strong benchmark. The dataset and our source code are available on Zenodo and GitHub, respectively.

Tobias Morocutti, Florian Schmid, Khaled Koutini et al.

Knowledge Distillation (KD) is a widespread technique for compressing the knowledge of large models into more compact and efficient models. KD has proved to be highly effective in building well-performing low-complexity Acoustic Scene Classification (ASC) systems and was used in all the top-ranked submissions to this task of the annual DCASE challenge in the past three years. There is extensive research available on establishing the KD process, designing efficient student models, and forming well-performing teacher ensembles. However, less research has been conducted on investigating which teacher model attributes are beneficial for low-complexity students. In this work, we try to close this gap by studying the effects on the student's performance when using different teacher network architectures, varying the teacher model size, training them with different device generalization methods, and applying different ensembling strategies. The results show that teacher model sizes, device generalization methods, the ensembling strategy and the ensemble size are key factors for a well-performing student network.

Tobias Morocutti, Florian Schmid, Khaled Koutini et al.

The ability to generalize to a wide range of recording devices is a crucial performance factor for audio classification models. The characteristics of different types of microphones introduce distributional shifts in the digitized audio signals due to their varying frequency responses. If this domain shift is not taken into account during training, the model's performance could degrade severely when it is applied to signals recorded by unseen devices. In particular, training a model on audio signals recorded with a small number of different microphones can make generalization to unseen devices difficult. To tackle this problem, we convolve audio signals in the training set with pre-recorded device impulse responses (DIRs) to artificially increase the diversity of recording devices. We systematically study the effect of DIR augmentation on the task of Acoustic Scene Classification using CNNs and Audio Spectrogram Transformers. The results show that DIR augmentation in isolation performs similarly to the state-of-the-art method Freq-MixStyle. However, we also show that DIR augmentation and Freq-MixStyle are complementary, achieving a new state-of-the-art performance on signals recorded by devices unseen during training.