Sebastian Strahl, Johannes Zeitler (International Audio Laboratories Erlangen): "Differentiable MIR: Classical Concepts Meet End-to-End Learning”
Many core problems in Music Information Retrieval (MIR) have traditionally been addressed using classical signal processing and alignment concepts. While end-to-end deep learning approaches have achieved impressive results, they often replace these classical methods entirely, reducing interpretability and discarding valuable domain knowledge. In this workshop talk, we argue for a complementary perspective: rethinking classical MIR concepts in differentiable form and integrating them into end-to-end learning pipelines. We present two case studies that illustrate how established methods can be adapted to modern neural systems. The first revisits fundamental frequency estimation by leveraging DSP-inspired representations within a lightweight, differentiable model, combining robustness, efficiency, and interpretability. The second focuses on sequence alignment, showing how differentiable approximations of classical Dynamic Time Warping and its variants enable end-to-end training from unsynchronized data. Together, these examples highlight how classical MIR concepts and end-to-end learning can be meaningfully combined, offering a principled path toward efficient, interpretable, and flexible MIR systems.

Laura Ibáñez-Martínez (UPF MTG): “Evaluating Disentangled Representations for Controllable Music Generation”
Recent approaches in music generation rely on disentangled representations, often labeled as structure and timbre or local and global, to enable controllable synthesis. Yet the underlying properties of these embeddings remain underexplored. In this work, we evaluate such disentangled representations in a set of music audio models for controllable generation using a probing-based framework that goes beyond standard downstream tasks. The selected models reflect diverse unsupervised disentanglement strategies, including inductive biases, data augmentations, adversarial objectives, and staged training procedures. We further isolate specific strategies to analyze their effect. Our analysis spans four key axes: informativeness, equivariance, invariance, and disentanglement, which are assessed across datasets, tasks, and controlled transformations. Our findings reveal inconsistencies between intended and actual semantics of the embeddings, suggesting that current strategies fall short of producing truly disentangled representations, and prompting a re-examination of how controllability is approached in music generation.

Soham Korade (Walmart Global Tech India), Suswara Pochampally (IIT Madras): “Modeling, Visualizing, and Interacting with Raga Structure Using Symbolic Music Representations”
In this talk, we will share a collection of projects exploring how computational methods and interactive systems can help us better understand and engage with Indian classical music. We will present our previous work on modeling and visualizing raga structure using symbolic notation, as well as the development of a large-scale notation dataset designed to support MIR and generative research. We will also briefly talk about our upcoming ICASSP 2026 paper, which investigates how shifts in tonal context influence musical behavior and perception, through an interactive experimental setup. Beyond the research papers, we will showcase ongoing work on building visualization tools, interactive interfaces, and creative systems. Our goal is to attempt new ways of analyzing and experiencing raga-based music.

Gerardo Roa-Dabike (University of Sheffield): “ICASSP 2026 Cadenza Challenge: Predicting Lyric Intelligibility”
Understanding the lyrics in music is key for music enjoyment. People with hearing loss can have difficulties clearly and effortlessly hearing lyrics, however. In speech technology, having metrics to automatically evaluate intelligibility has driven improvements in speech enhancement. We wanted to do the same for music with lyrics. To address this gap we presented the lyric intelligibility challenge. A new dataset, CLIP1, was introduced, comprising audio samples of popular western music paired with listener intelligibility scores. To model diverse listening profiles, samples were processed with no, mild and moderate simulated hearing loss. A total of 27 systems were submitted by 22 teams.

Sumit Kumar (Indian Institute of Technology, Kanpur): “Recognizing Ornaments in Vocal Indian Art Music with Active Annotation”
Ornamentations, embellishments, or microtonal inflections are essential to melodic expression across many musical traditions, adding depth, nuance, and emotional impact to performances. Recognizing ornamentations in singing voices is key to MIR, with potential applications in music pedagogy, singer identification, genre classification, and controlled singing voice generation. However, the lack of annotated datasets and specialized modeling approaches remains a major obstacle for progress in this research area. In this work, we introduce Rāga Ornamentation Detection (ROD), a novel dataset comprising Indian classical music recordings curated by expert musicians. The dataset is annotated using a custom Human-in-the-Loop tool for six vocal ornaments labelled as events with start and end timestamps. Using this dataset, we develop an ornamentation detection model based on deep time-series analysis, preserving ornament boundaries during the chunking of long audio recordings. We conduct experiments using different train-test configurations within the ROD dataset and also evaluate our approach on a separate, manually annotated dataset of live concert recordings. Experimental results support the superior performance of our proposed approach over the baseline CRNN model.

Eleonora Mancini (Deezer Research): “Leveraging Whisper Embeddings for Audio-based Lyrics Matching”
Audio-based lyrics matching can be an appealing alternative to other content-based retrieval approaches, but existing methods often suffer from limited reproducibility and inconsistent baselines. In this work, we introduce WEALY, a fully reproducible pipeline that leverages Whisper decoder embeddings for lyrics matching tasks. WEALY establishes robust and transparent baselines, while also exploring multimodal extensions that integrate textual and acoustic features. Through extensive experiments on standard datasets, we demonstrate that WEALY achieves a performance comparable to state-of-the-art methods that lack reproducibility. In addition, we provide ablation studies and analyses on language robustness, loss functions, and embedding strategies. This work contributes a reliable benchmark for future research and underscores the potential of speech technologies for music information retrieval tasks.

Andrea Poltronieri, Pablo Alonso, Martín Rocamora, Dmitry Bogdanov (UPF MTG): “Current approaches to computational analysis of music audio signals (Tutorial)”
Computational analysis of music audio signals lies at the core of Music Information Retrieval (MIR), aiming to extract semantically meaningful information from audio signals to describe, organise, and interact with music at scale. It encompasses a wide range of tasks, such as beat tracking and tempo estimation, pitch and melody extraction, chord and key recognition, instrument and timbre classification, and music tagging. These tasks also underpin numerous real-world applications, including music recommendation, automatic transcription, performance analysis, and content-based retrieval. Despite their diversity, these tasks share challenges specific to the musical domain. Music data are often scarce, unevenly annotated, and biased toward certain genres or cultures, making learning and evaluation difficult. Musical signals also exhibit complex rhythmic and harmonic dependencies that require models capable of capturing relationships across multiple structural levels. Moreover, evaluation remains non-trivial, as metrics must account for musical plausibility and perceptual relevance. These properties have driven the development of MIR-specific representations and architectures that integrate domain knowledge into the learning process. In recent years, MIR has evolved rapidly, following broader trends in signal processing and machine learning, with increasing attention to large-scale pre-training and representation learning paradigms. Recent advances in self-supervised and contrastive learning have led to foundational models trained in large musical corpora, capable of learning rich and transferable representations that support multiple MIR tasks. In addition, multimodal architectures that jointly process audio, symbolic, and textual modalities are expanding the scope of MIR toward cross-modal retrieval, captioning, and question-answering. These advances are reshaping MIR, introducing new large-scale and multimodal paradigms that demand a renewed understanding of how signal processing methods can adapt to and inform modern music analysis. As these approaches increasingly rely on large and diverse musical corpora, the limitations and risks associated with data and pretrained models become even more evident. Unlike other audio domains, music recordings are deeply entangled with issues of copyright, ownership, and performer rights, which restrict the creation and distribution of open datasets and hinder reproducibility. At the same time, large-scale collections often mirror cultural and market imbalances, amplifying biases in genre, geography, and instrumentation. These aspects underscore the importance of transparent data governance and responsible research practices to safeguard openness, reproducibility, and fairness in MIR.

Hyon Kim (UPF-MTG): “Velocity2DMs: A Contextual Modeling Approach to Dynamics Marking Prediction in Piano Performance”
Expressive dynamics in music performance are sub-jective and context-dependent, yet most symbolic models treat Dynamics Markings (DMs) as static with fixed MIDI velocities. This paper proposes a method for predicting DMs in piano performance by combining MusicXML score information with performance MIDI data through a novel tokenization scheme and an adapted RoBERTa-based Masked Language Model (MLM). Our approach focuses on contextual aggregated MIDI velocities and corresponding DMs, accounting for subjective interpretations of pianists. Note-level features are serialized and translated into a sequence of tokens to predict both constant (e.g., mp, ff ) and non-constant DMs (e.g., crescendo, fp). Evaluation across three expert performance datasets shows that the model effectively learns dynamics transitions from contextual note blocks and generalizes beyond constant markings. This is the first study to model both constant and non-constant dynamics in a unified framework using contextual sequence learning. The results suggest promising applications for expressive music analysis, performance modeling, and computer-assisted music education. 

Minhui Lu: “Learning Vocal-Tract Area and Radiation with a Physics-Informed Webster Model”
We present a physics-informed voiced backend renderer for singing-voice synthesis. Given synthetic single-channel audio and a fund-amental--frequency trajectory, we train a time-domain Webster model as a physics-informed neural network to estimate an interpretable vocal-tract area function and an open-end radiation coefficient. Training enforces partial differential equation and boundary consistency; a lightweight DDSP path is used only to stabilize learning, while inference is purely physics-based. On sustained vowels (/a/, /i/, /u/), parameters rendered by an independent finite-difference time-domain Webster solver reproduce spectral envelopes competitively with a compact DDSP baseline and remain stable under changes in discretization, moderate source variations, and about ten percent pitch shifts. The in-graph waveform remains breathier than the reference, motivating periodicity-aware objectives and explicit glottal priors in future work.

Marius Miron (Earth Species Project): “Multi-layer attentive probing improves transfer of audio representations for bioacoustics”
Probing heads map the representations learned from audio by a machine learning model to downstream task labels and are a key component in evaluating representation learning. Most bioacoustic benchmarks use a fixed, low-capacity probe, such as a linear layer on the final encoder layer. While this standardization enables model comparisons, it may bias results by overlooking the interaction between encoder features and probe design. In this work, we systematically study different probing strategies across two bioacoustic benchmarks, BEANs and BirdSet. We evaluate last- and multi-layer probing, across linear and attention probes. We show that larger probe heads that leverage time information have superior performance. Our results suggest that current benchmarks may misrepresent encoder quality when relying on a last-layer probing setup. Multi-layer probing improves downstream task performance across all tested models, while attention probing has superior performance to linear probing for transformer models.   

Pedro Ramoneda (UPF MTG): “Can Audio Reveal Music Performance Difficulty? Insights From the Piano Syllabus Dataset”
Automatically estimating the performance difficulty of a music piece represents a key process in music education to create tailored curricula according to the individual needs of the students. Given its relevance, the Music Information Retrieval (MIR) field comprises some proof-of-concept works addressing this task that mainly focus on high-level music abstractions such as machine-readable scores or music sheet images. In this regard, the potential of directly analyzing audio recordings has generally been neglected. This work addresses this gap in the field with two contributions: (i) PSyllabus, the first audio-based difficulty estimation dataset—collected from Piano Syllabus community—featuring 7,901 piano pieces across 11 difficulty levels from 1,233 composers as well as two additional benchmark datasets particularly compiled for evaluation purposes; and (ii) a recognition framework capable of managing different input representations—both in unimodal and multimodal manners—derived from audio to perform the difficulty estimation task. The comprehensive experimentation comprising different pre-training schemes, input modalities, and multi-task scenarios proves the validity of the hypothesis and establishes PSyllabus as a reference dataset for audio-based difficulty estimation in the MIR field. The dataset, developed code, and trained models are publicly shared to promote further research in the field.

Sofiene Kammoun (CentraleSupélec), Xavier Alameda-Pineda (Inria), Simon Leglaive (CentraleSupélec): “Modeling strategies for speech enhancement in the latent space of a neural audio codec”
Neural audio codecs (NACs) provide compact latent speech representations in the form of sequences of continuous vectors or discrete tokens. In this work, we investigate how these two types of speech representations compare when used as training targets for supervised speech enhancement. We consider both autoregressive and non-autoregressive speech enhancement models based on the Conformer architecture, as well as a simple baseline where the NAC encoder is simply fine-tuned for speech enhancement. Our experiments reveal three key findings: predicting continuous latent representations consistently outperforms discrete token prediction; autoregressive models achieve higher quality but at the expense of intelligibility and efficiency, making non-autoregressive models more attractive in practice; and adding encoder fine-tuning yields the strongest enhancement metrics overall, though at the cost of degraded codec reconstruction.

Riccardo Casciotti (Tampere University): “Incremental Learning for Audio Classification with Hebbian Deep Neural Networks”
The ability of humans for lifelong learning is an inspiration for deep learning methods and in particular for continual learning. In this work, we apply Hebbian learning, a biologically inspired learning process, to sound classification. We propose a kernel plasticity approach that selectively modulates network kernels during incremental learning, acting on selected kernels to learn new information and on others to retain previous knowledge. Using the ESC-50 dataset, the proposed method achieves 76.3% overall accuracy over five incremental steps, outperforming a baseline without kernel plasticity (68.7%) and demonstrating significantly greater stability across tasks.

Dongmin Kim (MALer Lab), Jongmin Jung (Neutune/MALer Lab), Sihun Lee (MALer Lab): “U-MusT: A Unified Framework for Cross-modal Translation of Score Images, Symbolic Music, and Performance Audio”
Music exists in various modalities, such as score images, symbolic scores, MIDI, and audio. Translations between such modalities are established as core tasks of music information retrieval, such as automatic music transcription (audio-to-MIDI) and optical music recognition (score image to symbolic score). However, most past work on multimodal translation utilizes specialized models trained for each translation task. In this paper, we propose a unified framework based on a common tokenization strategy. We use dedicated separate models for the Image-to-Audio and Audio-to-Image directions, sharing an identical encoder-decoder architecture to handle each task within a coherent framework. Two key factors make this unified approach viable: a new large-scale dataset, and the tokenization of each modality. Firstly, we propose a new dataset that consists of more than 1,300 hours of paired audio-score image data collected from YouTube videos, which is an order of magnitude larger than any existing music modal translation datasets. Secondly, our unified tokenization framework discretizes score images, audio, MIDI, and MusicXML into a sequence of tokens, enabling standard encoder-decoder Transformers to tackle multiple crossmodal translation as one coherent sequence-to-sequence task. Experimental results confirm that our unified framework improves upon single-task baselines in several key areas, notably reducing the symbol error rate for optical music recognition from 24.58% to a state-of-the-art 13.67%, while also seeing substantial improvements across the other translation tasks. Notably, our approach achieves the first musically-coherent score-image-conditioned audio generation, marking a significant breakthrough in cross-modal music generation.

Francesca Ronchini, Luca Comanducci (Politecnico di Milano): “Mitigating data replication in text-to-audio generative diffusion models through anti-memorization guidance”
A persistent challenge in generative audio models is data replication, where the model unintentionally generates parts of its training data during inference. In this work, we address this issue in text-to-audio diffusion models by exploring the use of anti-memorization strategies. We adopt Anti-Memorization Guidance (AMG), a technique that modifies the sampling process of pre-trained diffusion models to discourage memorization. Our study explores three types of guidance within AMG, each designed to reduce replication while preserving generation quality. We use Stable Audio Open as our backbone, leveraging its fully open-source architecture and training dataset. Our comprehensive experimental analysis suggests that AMG significantly mitigates memorization in diffusion-based text-to-audio generation without compromising audio fidelity or semantic alignment.

Yinghao Ma (C4DM QMUL): “AutoMV: An Automatic Multi-Agent System for Music Video Generation”
Music-to-Video (M2V) generation for full-length songs faces significant challenges. Existing methods produce short, disjointed clips, failing to align visuals with musical structure, beats, or lyrics, and lack temporal consistency. We propose AutoMV, a multi-agent system that generates full music videos (MVs) directly from a song. AutoMV first applies music processing tools to extract musical attributes, such as structure, vocal tracks, and time-aligned lyrics, and constructs these features as contextual inputs for following agents. The screenwriter Agent and director Agent then use this information to design short script, define character profiles in a shared external bank, and specify camera instructions. Subsequently, these agents call the image generator for keyframes and different video generators for "story" or "singer" scenes. A Verifier Agent evaluates their output, enabling multi-agent collaboration to produce a coherent longform MV. To evaluate M2V generation, we further propose a benchmark with four high-level categories (Music Content, Technical, Post-production, Art) and twelve ine-grained criteria. This benchmark was applied to compare commercial products, AutoMV, and human-directed MVs with expert human raters: AutoMV outperforms current baselines significantly across all four categories, narrowing the gap to professional MVs. Finally, we investigate using large multimodal models as automatic MV judges; while promising, they still lag behind human expert, highlighting room for future work.

Javier Nistal (Sony CSL): “Context-Aware Stem Generation in Practice: Diff-A-Riff's GUI and Future Controls”
Diff-A-Riff is a latent-diffusion system for high-quality accompaniment generation. This talk will introduce and live-demo our current GUI, which makes the model usable in an iterative creative workflow: users can flexibly generate and audition stems for any instrument given arbitrary musical context, using a set of practical controls. The GUI reflects the capabilities in the published work, and we will then give a brief overview of ongoing research directions: inference-time feature manipulation by intervening on the gradient field to steer sampling without retraining, training with corrupted source-separated stems to expand paired training data under realistic artifacts, and information-theoretic conditional signals aimed at disentangling and exposing meaningful controls.

Ivan Shanin: “Audio-to-Score Jazz Solo Transcription with the Rhythm Perceiver”
Monophonic audio-to-score transcription remains a challenging task, especially for improvised jazz solos. Prior work commonly follows a modular pattern: combining audio-to-performance-MIDI transcription, state-of-the-art beat tracking, and a post-processing algorithm that assigns metrical durations to notes (note values). This approach is prone to error accumulation at every step. Motivated by how expert transcribers work, we propose an end-to-end approach that inverts the logical order of subtasks: first, we model rhythm perception by transcribing the rhythmic structure of the melody at the beat level; then, we estimate the pitch for each note within the predicted rhythmic structures. Technically, our model introduces a rhythm-aware transformer architecture with perceiver-style cross-attention that jointly aligns frame-level acoustic features with beat-synchronous rhythmic embeddings. The proposed solution achieves state-of-the-art results on a challenging corpus of alto saxophone solos by Charlie Parker (the ``Omnibook'').

Oriol Nieto (Adobe Research): “Sound Design AI with Structure and Generation”
Sound design requires tools that support both meaningful audio representation and flexible transformation of sound materials. This talk presents four works that address these needs through structured metadata and generative modeling techniques designed for professional sound design workflows. We first introduce audiocards, structured metadata grounded in acoustic attributes and sonic descriptors, which improve audio–language modeling for sound retrieval, captioning, and metadata generation. We then present two approaches to generative sound transformation, enabling controllable audio extension and perceptually coherent sound morphing, including sound infusions learned without curated morphing datasets. Finally, we introduce PromptSep, a generative framework for sound separation that supports both extraction and removal, and extends text-based control with vocal imitation as an intuitive conditioning modality. Together, these works demonstrate how structured representations and targeted generative methods can better align audio models with the practical and creative needs of sound designers.

Brandon Carone (NYU-MARL): “The MUSE Benchmark: Probing Music Perception and Auditory Relational Reasoning in Audio LLMs”
Multimodal Large Language Models (MLLMs) have demonstrated capabilities in audio understanding, but current evaluations do not demonstrate relational reasoning. We introduce the Music Understanding and Structural Evaluation (MUSE) Benchmark, consisting of 10 tasks designed to probe fundamental music perception skills. We evaluate four SOTA models (Gemini Pro and Flash, Qwen2.5-Omni, and Audio-Flamingo 3) against humans (N=200). Results reveal a wide variance in SOTA capabilities and a persistent gap with human experts. While Gemini Pro succeeds on basic perception, Qwen and Audio Flamingo 3 perform at or near chance, exposing severe perceptual deficits. Furthermore, we find Chain-of-Thought (CoT) prompting provides inconsistent, often detrimental results. MUSE is a critical tool for evaluating invariant musical representations and driving development of more robust AI systems.

Sripathi Sridhar (New Jersey Institute of Technology): “Audiocards: Structured Metadata Improves Audio Language Models for Sound Design”
Sound designers search for sounds in large sound effects libaries using aspects such as sound class or visual context. However, the metadata needed for such search is often missing or incomplete and requires significant manual effort to add. Existing solutions to automate this task by generating metadata, i.e. captioning, and search using learned embeddings, i.e. text-audio retrieval, are not trained on metadata with the structure and information pertinent to sound design. To this end we propose audiocards, structured metadata grounded in acoustic attributes and sonic descriptors, by exploiting the world knowledge of LLMs. We show that training on audiocards improves downstream text-audio retrieval, descriptive captioning, and metadata generation on professional sound effects libraries. Moreover, audiocards also improve performance on general audio captioning and retrieval over the baseline single-sentence captioning approach. We release a curated dataset of sound effects audiocards to invite further research in audio language modeling for sound design.

Manu Harju (Tampere University): “Self-labeling sounds using optimal transport”
Self-labeling is a method to simultaneously learn representations and classes using unlabeled data. The naive approach to self-labeling leads to a degenerate solution, and the model-generated labels require regularization to serve as useful training targets. In this work, we adapt a self-labeling method using optimal transport to the audio domain using the FSD50K dataset. We analyze the structure of the learned representations and compare the emergent classes with the reference annotations. We compare the learned representations with the ones produced using Bootstrap Your Own Latent for Audio (BYOL-A) across several downstream tasks. Our findings indicate that the method learns to group perceptually similar sounds without supervision. The results show that the method is a viable approach for audio representation learning, and that the learned embeddings are as effective for downstream tasks as the ones obtained with the benchmark method. As an additional outcome, the generated classifications give valuable insight into what the model learns, promoting explainability in feature learning.

Parampreet Singh (Indian Institute of Technology, Kanpur): “Explainable Deep Learning Analysis for Raga Identification in Indian Art Music”
Raga identification is an important problem within the domain of Indian Art music, as Ragas are fundamental to its composition and performance, playing a crucial role in music retrieval, preservation, and education. Few studies that have explored this task employ approaches such as signal processing, Machine Learning (ML), and more recently, Deep Learning (DL) based methods. However, a key question remains unanswered in all these works: do these ML/DL methods learn and interpret Ragas in a manner similar to human experts? Besides, a significant roadblock in this research is the unavailability of an ample supply of rich, labeled datasets, which drives these ML/DL-based methods. In this paper, firstly we curate a dataset comprising 191 hours of Hindustani Classical Music (HCM) recordings, annotate it for Raga and tonic labels, and train a CNN-LSTM model for the task of Automatic Raga Identification (ARI). We achieve a chunk-wise f1-measure of 0.89 for a subset of 12 Raga classes. Following this, we make one of the first attempts to employ model explainability techniques: SoundLIME and GradCAM++ for Raga identification, to evaluate whether the classifier's predictions align with human understanding of Ragas. We compare the generated explanations with human expert annotations and further analyze individual test examples to understand the role of regions highlighted by explanations in making correct or incorrect predictions made by the model. Our results demonstrate a significant alignment of the model's understanding with human understanding, and the thorough analysis validates the effectiveness of our approach.

Zachary Novack (University of California San Diego): “Sonic BOOM! The Path Towards Fast and Interactive Sound Diffusion”
Diffusion-based generative models have revolutionized multimedia generation, and in particular, such models for music and sound have expanded the possibilities of creative sonic artistic practice. However, such models remain generally slow, non-iteractive fixed artifacts; using these in real, interactive setups remains generally underexplored, and has shifted attention overly towards large offline systems. To counteract this, a growing body of work has seeked to rapidly accelerate such models into interactive tools. In this talk, I will go over a number of works on accelerating diffusion-based generative models, in music, sound design, and beyond. Particularly, we will analyze what sort of methodological algorithmic tools are used to turn slow diffusion models into fast generators, practical considerations of using such models in low-resource settings, and finally how such models can be turned into interactive multimedia art pieces.

Simon Schwär, Manuel Peters: “Differentiable Wind Instrument Synthesis with Pulsetables and Beyond”
Pulsetable synthesis is a concatenative audio synthesis method that generates harmonic tones by repeating short waveform prototypes at regular intervals. While related to the more commonly used wavetable method, using pulsetables aligns closely with the physical excitation mechanism of some reed and brass instruments, making it well suited for efficient modeling of those sounds. However, a convincing synthesis result typically requires extensive parameter tuning. In this paper, we explore pulsetable synthesis in the context of differentiable digital signal processing (DDSP). We introduce a differentiable pulsetable synthesizer in which pulse prototypes are directly optimized via gradient descent, jointly trained with a lightweight neural network to select pulses given a target pitch and velocity. Building on this approach, we propose a framework for wind instrument synthesis that provides interpretable control parameters and can be trained in an unsupervised way using only a few minutes of recordings from a target instrument. Initial experiments with resynthesis and timbre transfer demonstrate that our approach enables expressive synthesis with only 60,000 trainable parameters, offering an efficient alternative to other instrument synthesis methods. In this framework, we further explore differences and commonalities between DDSP synthesis methods, such as additive synthesis, frequency modulation, and wavetables.

David López (UPF MTG): “AI-Generated Music Detection in Broadcast Monitoring”
AI music generators have advanced to the point where their outputs are often indistinguishable from human compositions. While detection methods have emerged, they are typically designed and validated in music streaming contexts with clean, full-length tracks. Broadcast audio, however, poses a different challenge: music appears as short excerpts, often masked by dominant speech, conditions under which existing detectors fail. In this work, we introduce AI-OpenBMAT, the first dataset tailored to broadcast-style AI-music detection. It contains 3,294 one-minute audio excerpts (54.9 hours) that follow the duration patterns and loudness relations of real television audio, combining human-made production music with stylistically matched continuations generated with Suno v3.5. We benchmark a CNN baseline and state-of-the-art SpectTTTra models to assess SNR and duration robustness, and evaluate on a full broadcast scenario. Across all settings, models that excel in streaming scenarios suffer substantial degradation, with F1-scores dropping below 60% when music is in the background or has a short duration. These results highlight speech masking and short music length as critical open challenges for AI music detection, and position AI-OpenBMAT as a benchmark for developing detectors capable of meeting industrial broadcast requirements.

Alain Riou (Universal Music Group): “Automatic Music Sample Detection with Multi-Track Contrastive Learning”
Sampling, the technique of reusing pieces of existing audio tracks to create new music content, is a very common practice in modern music production. In this paper, we tackle the challenging task of automatic sample identification, that is, detecting such sampled content and retrieving the material from which it originates. To do so, we adopt a self-supervised learning approach that leverages a multi-track dataset to create positive pairs of artificial mixes, and design a novel contrastive learning objective. We show that such method significantly outperforms previous state-of-the-art baselines, that is robust to various genres, and that scales well when increasing the number of noise songs in the reference database. In addition, we extensively analyze the contribution of the different components of our training pipeline and highlight, in particular, the need for high-quality separated stems for this task.

Pablo Alonso (UPF MTG): “Benchmarking Music Autotagging with MGPHot Expert Annotations vs. Generic Tag Datasets”
Music autotagging aims to automatically assign descriptive tags, such as genre, mood, or instrumentation, to audio recordings. Due to its challenges, diversity of semantic descriptions, and practical value in various applications, it has become a common downstream task for evaluating the performance of general-purpose music representations learned from audio data. We introduce a new benchmarking dataset based on the recently published MGPHot dataset, which includes expert musicological annotations, allowing for additional insights and comparisons with results obtained on common generic tag datasets. While MGPHot annotations have been shown to be useful for computational musicology, the original dataset neither includes audio nor provides evaluation setups for its use as a standardized autotagging benchmark. To address this, we provide a curated set of YouTube URLs with retrievable audio, and propose a train/val/test split for standardized evaluation, and precomputed representations for seven state-of-the-art models. Using these resources, we evaluated these models in MGPHot and standard reference tag datasets, highlighting key differences between expert and generic tag annotations. Altogether, our contributions provide a more advanced benchmarking framework for future research in music understanding.

Bernardo Torres (Telecom Paris, Institut Polytechnique de Paris): “Learning Linearity in Audio Consistency Autoencoders via Implicit Regularization”
Audio autoencoders learn useful, compressed audio representations, but their non-linear latent spaces prevent intuitive algebraic manipulation such as mixing or scaling. We introduce a simple training methodology to induce linearity in a high-compression Consistency Autoencoder (CAE) by using data augmentation, thereby inducing homogeneity (equivariance to scalar gain) and additivity (the decoder preserves addition) without altering the model's architecture or loss function. When trained with our method, the CAE exhibits linear behavior in both the encoder and decoder while preserving reconstruction fidelity. We test the practical utility of our learned space on music source composition and separation via simple latent arithmetic. This work presents a straightforward technique for constructing structured latent spaces, enabling more intuitive and efficient audio processing.

Zhanhong He (University of Western Australia): “Joint Estimation of Piano Dynamics & Structure”
We present a compact multi-task, multi-scale network that jointly predicts piano dynamic levels, change points, beats, and downbeats. This multi-scale design captures both long- and short-term trends from the same input, benefiting all tasks. Using Bark-scale specific loudness (BSSL) features, our model reduces parameters from 14.7M (log-Mel baseline) to 0.5M while improving performance. Trained on the Chopin MazurkaBL dataset, it provides an efficient tool for adding labels and analyzing piano dynamics. We open-source the project on GitHub.

Hanyu Meng (The University of New South Wales): “Biologically Inspired Adaptive Front-end Paradigm for Robust Audio Signal Processing”
Robust machine hearing in real-world acoustic environments remains challenging because of non-stationary noise, reverberation, and multiple moving sound sources. Humans handle such conditions effectively through adaptive processing in the auditory periphery, supported by feedback mechanisms that regulate auditory sensitivity and frequency selectivity. In machine hearing systems, the front-end extracts sound representations while the back-end performs perception and decision making. However, most existing audio front-ends operate with fixed processing during inference, limiting their ability to adapt to changing acoustic environments. This presentation will present the research for the past decade about how we address this limitation by developing biologically inspired adaptive audio front-end models and novel temporal, spectral, and even spatial acoustic feature representations that improve robustness in diverse acoustic environments. The recent paper, which will present in ICASSP2026 about adaptive dynamic range compression front-end for robust monaural audio signal processing, will also be included in this presentation.

Magdalena Fuentes (New York University): “Investigating Modality Contribution In Audio LLMs for Music”
Audio Large Language Models (Audio LLMs) enable human-like conversation about music, yet it is unclear if they are truly listening to the audio or just using textual reasoning, as recent benchmarks suggest. This paper investigates this issue by quantifying the contribution of each modality to a model’s output. We adapt the MMSHAP framework, a performance-agnostic score based on Shapley values that quantifies the relative contribution of each modality to a model’s prediction. We evaluate two models on the MuChoMusic benchmark and find that the model with higher accuracy relies more on text to answer questions, but further inspection shows that even if the overall audio contribution is low, models can successfully localize key sound events, suggesting that audio is not entirely ignored. Our study is the first application of MM-SHAP to Audio LLMs and we hope it will serve as a foundational step for future research in explainable AI and audio reasoning.