Slim Essid's research page

Weak supervision sets a practical regime for audio-visual sound source localization as dense spatial annotations are costly to obtain at scale. The task, however, remains challenging, as models must locate sound sources from temporally aligned audio-visual data without pixel-level supervision. Recent large-scale audio-visual retrieval models, trained at unprecedented scale, encode rich multimodal structure. We show their latent representations, though optimized for global alignment, can nonetheless enable fine-grained spatial grounding. While spatial detail is progressively lost in the upper layers of retrieval backbones due to global pooling, intermediate visual tokens retain highly structured spatial information. To exploit this, we introduce LAIP (\emphLocalization via Audio-Informed Pooling), a framework that employs a lightweight \emphAudio-informed Spatial Pooling (AiSP) to replace the standard global aggregation module. By querying intermediate visual tokens with audio aligned at the frame level, LAIP recovers localized spatial information that is otherwise discarded by the retrieval pipeline, with the largest gains observed for PE-AV, a stack with underlying temporal aggregation. Our approach achieves state-of-the-art performance on AVSBench and AVATAR, nearly doubling previous results on the latter, improving average CIoU from 13.21 to 26.22.

We introduce NemotronLabs VoiceChat, an open full-duplex speech-to-speech model with native tool-calling capabilities. NemotronLabs VoiceChat combines a streaming speech encoder and decoder-only language model with parallel specialized output streams for agent text and structured function calls, an auxiliary RNN-T branch for incremental user transcription, and a streaming TTS decoder. This design enables the model to listen, transcribe, reason, invoke tools, and speak within a unified streaming architecture while preserving the temporal behavior required for natural conversation. On Full-Duplex-Bench 1.0, NemotronLabs VoiceChat achieves the lowest pause-handling takeover rates among evaluated open-weight systems, 100% takeover following user interruptions, and a 4.33/5 post-interruption response-quality score. On Full-Duplex-Bench 1.5, it resumes its response after user backchannels in 93% of cases. NemotronLabs VoiceChat obtains a 55.1 normalized average on VoiceBench and, on Full-Duplex-Bench 3.0 (FDB 3.0), achieves 82.5% tool-selection F1, while argument accuracy and end-to-end tool execution remain areas for improvement. These results demonstrate that full-duplex interaction, speech recognition and generation, general language capabilities, and external tool use can be integrated in a single open speech-to-speech model without sacrificing real-time conversational behavior.

We propose LoRA-MCL, a training scheme that extends next-token prediction in language models with a method designed to decode diverse, plausible sentence continuations at inference time. Traditional language modeling is an intrinsically ill-posed problem: given a context, multiple “futures” may be equally plausible. Our approach leverages Multiple Choice Learning (MCL) and the Winner-Takes-All loss to efficiently handle ambiguity through Low-Rank Adaptation. We provide a theoretical interpretation of applying MCL to language modeling, assuming the data is generated from a mixture of distributions. We illustrate the proposed approach using mixtures of Markov chains. We then demonstrate with experiments on visual and audio captioning, as well as machine translation, that our method achieves high diversity and relevance in generated outputs.

Large-scale pre-trained audio and image models demonstrate an unprecedented degree of generalization, making them suitable for a wide range of applications. Here, we tackle the specific task of sound-prompted segmentation, aiming to segment image regions corresponding to objects heard in an audio signal. Most existing approaches tackle this problem by fine-tuning pre-trained models or by training additional modules specifically for the task. We adopt a different strategy: we introduce a training-free approach that leverages Non-negative Matrix Factorization (NMF) to co-factorize audio and visual features from pre-trained models so as to reveal shared interpretable concepts. These concepts are passed on to an open-vocabulary segmentation model for precise segmentation maps. By using frozen pre-trained models, our method achieves high generalization and establishes state-of-the-art performance in unsupervised sound-prompted segmentation, significantly surpassing previous unsupervised methods.

Music understanding and reasoning are central challenges in the Music Information Research field, with applications ranging from retrieval and recommendation to music agents and virtual assistants. Recent Large Audio-Language Models (LALMs) have shown remarkable progress in answering music-related questions by following user instructions. However, their massive scale, often billions of parameters, results in expensive training, slow inference, and limited deployability on edge devices. In this work, we present TinyMU, a lightweight (229M) Music-Language Model (MLM) that achieves performance comparable to much larger LALMs while remaining efficient and compact. To train TinyMU, we introduce MusicSkills-3.5M, a carefully curated, music-grounded question-answering dataset with 3.5M samples. Spanning multiple-choice, binary, and open-ended formats, this dataset provides fine-grained supervision across diverse musical concepts. For its architecture, TinyMU leverages MATPAC++, the SOTA self-supervised audio encoder for fine-grained feature extraction. Paired with a lightweight linear projector, it efficiently aligns audio embeddings with the language model. Through extensive evaluation, we show that TinyMU performs strongly in both basic music understanding and complex reasoning. Notably, on the MuChoMusic benchmark, it achieves 82% of SOTA LALM’s performance despite being 35× smaller, highlighting the potential of small MLMs under constrained computational budgets.

Masked latent prediction has emerged as a leading paradigm in self-supervised learning (SSL), especially for general audio and music representation learning. While recent methods have demonstrated strong performance, the role of the predictor module used at the output of such SSL systems remains mainly overlooked, despite being crucial for solving the pretext task at hand. In particular, this module should be able to deal with the ambiguity inherent in audio content, especially when it is composed of multiple sound sources. This work proposes a novel enhancement: integrating Multiple Choice Learning (MCL) to explicitly model prediction ambiguity and improve representation quality. We build on top of the recently proposed MATPAC system, improving its prediction and unsupervised classification pretext tasks with MCL. We extensively evaluate our method, MATPAC++, through both linear probing across multiple downstream tasks and fine-tuning on AudioSet, employing a unified protocol that enables rigorous and fair comparisons with state-of-the-art SSL approaches. Results show that our proposal achieves state-of-the-art when fine-tuned on AudioSet and overall state-of-the-art scores on downstream tasks. Additionally, we examine domain specialisation by training exclusively on music data, where our model achieves state-of-the-art performance with significantly improved efficiency.

Contrastive Language–Audio Pretraining (CLAP) models learn by aligning audio and text in a shared embedding space, enabling powerful zero-shot recognition. However, their performance is highly sensitive to prompt formulation and language nuances, and they often inherit semantic ambiguities and spurious correlations from noisy pretraining data. While prior work has explored prompt engineering, adapters, and prefix tuning to address these limitations, the use of structured prior knowledge remains largely unexplored. We present iKnow-audio, a framework that integrates knowledge graphs with audio-language models to provide robust semantic grounding. iKnow-audio builds on the Audio-centric Knowledge Graph (AKG), which encodes ontological relations comprising semantic, causal, and taxonomic connections reflective of everyday sound scenes and events. By training knowlege graph embedding models on the AKG and refining CLAP predictions through this structured knowledge, iKnow-audio improves disambiguation of acoustically similar sounds and reduces reliance on prompt engineering. Comprehensive zero-shot evaluations across six benchmark datasets demonstrate consistent gains over baseline CLAP, supported by embedding-space analyses that highlight improved relational grounding.

We introduce Annealed Multiple Choice Learning (aMCL) which combines simulated annealing with MCL. MCL is a learning framework handling ambiguous tasks by predicting a small set of plausible hypotheses. These hypotheses are trained using the Winner-takes-all (WTA) scheme, which promotes the diversity of the predictions. However, this scheme may converge toward an arbitrarily suboptimal local minimum, due to the greedy nature of WTA. We overcome this limitation using annealing, which enhances the exploration of the hypothesis space during training. We leverage insights from statistical physics and information theory to provide a detailed description of the model training trajectory. Additionally, we validate our algorithm by extensive experiments on synthetic datasets on the standard UCI benchmark, and on speech separation.

Multimodal large language models have fueled progress in image captioning. These models, fine-tuned on vast image datasets, exhibit a deep understanding of semantic concepts. In this work, we show that this ability can be re-purposed for audio captioning, where the joint image-language decoder can be leveraged to describe auditory content associated with image sequences within videos featuring audiovisual content. This can be achieved via multimodal alignment. Yet, this multimodal alignment task is non-trivial due to the inherent disparity between audible and visible elements in real-world videos. Moreover, multimodal representation learning often relies on contrastive learning, facing the challenge of the so-called modality gap which hinders smooth integration between modalities. In this work, we introduce a novel methodology for bridging the audiovisual modality gap by matching the distributions of tokens produced by an audio backbone and those of an image captioner. Our approach aligns the audio token distribution with that of the image tokens, enabling the model to perform zero-shot audio captioning in an unsupervised fashion. This alignment allows for the use of either audio or audiovisual input by combining or substituting the image encoder with the aligned audio encoder. Our method achieves significantly improved performances in zero-shot audio captioning, compared to existing approaches.

Self-supervised learning (SSL) leverages large datasets of unlabeled speech to reach impressive performance with reduced amounts of annotated data. The high number of proposed approaches fostered the emergence of comprehensive benchmarks that evaluate their performance on a set of downstream tasks exploring various aspects of the speech signal. However, while the number of considered tasks has been growing, most proposals rely upon a single downstream architecture that maps the frozen SSL representations to the task labels. This study examines how benchmarking results are affected by changes in the probing head architecture. Interestingly, we found that altering the downstream architecture structure leads to significant fluctuations in the performance ranking of the evaluated models. Against common practices in speech SSL benchmarking, we evaluate larger-capacity probing heads, showing their impact on performance, inference costs, generalization, and multi-level feature exploitation.

Winner-takes-all training is a simple learning paradigm, in which the multiple predictions of so-called hypotheses are leveraged to tackle ambiguous tasks. Recently, a connection was established between winner-takes-all training and centroidal Voronoi tessellations, showing that, once trained, the hypotheses should quantize optimally the shape of the conditional distribution to predict. However, probabilistic reliability guarantees for the predictions are missing. In this work, we show how to take advantage of the appealing geometrical properties of the winner-takes-all learners for conditional density estimation, without modifying its original training scheme. We then discuss the competitiveness of our estimator based on novel theoretical and experimental results on both synthetic and audio data.

Domain Generalized Semantic Segmentation (DGSS) deals with training a model on a labeled source domain with the aim of generalizing to unseen domains during inference. Existing DGSS methods typically effectuate robust features by means of Domain Randomization (DR). Such an approach is often limited as it can only account for style diversification and not content. In this work, we take an orthogonal approach to DGSS and propose to use an assembly of CoLlaborative FOUndation models for Domain Generalized Semantic Segmentation (CLOUDS). In detail, CLOUDS is a framework that integrates FMs of various kinds: (i) CLIP backbone for its robust feature representation, (ii) generative models to diversify the content, thereby covering various modes of the possible target distribution, and (iii) Segment Anything Model (SAM) for iteratively refining the predictions of the segmentation model. Extensive experiments show that our CLOUDS excels in adapting from synthetic to real DGSS benchmarks and under varying weather conditions, notably outperforming prior methods by 5.6% and 6.7% on averaged miou, respectively.

The task of music structure analysis refers to automatically identifying the location and the nature of musical sections within a song. In the supervised scenario, structural annotations generally result from exhaustive data collection processes, which represents one of the main challenges of this task. Moreover, both the subjectivity of music structure and the hierarchical characteristics it exhibits make the obtained structural annotations not fully reliable, in the sense that they do not convey a "universal ground-truth" unlike other tasks in music information retrieval. On the other hand, the quickly growing quantity of available music data has enabled weakly supervised and self-supervised approaches to achieve impressive results on a wide range of music-related problems. In this work, a self-supervised learning method is proposed to learn robust multi-level music representations prior to structural segmentation using contrastive learning. To this end, sets of frames sampled at different levels of detail are used to train a deep neural network in a disentangled manner. The proposed method is evaluated on both flat and multi-level segmentation. We show that each distinct sub-region of the output embeddings can efficiently account for structural similarity at their own targeted level of detail, which ultimately improves performance of downstream flat and multi-level segmentation. Finally, complementary experiments are carried out to study how the obtained representations can be further adapted to specific datasets using a supervised fine-tuning objective in order to facilitate structure retrieval in domains where human annotations remain scarce.

We introduce Resilient Multiple Choice Learning (rMCL), an extension of the MCL approach for conditional distribution estimation in regression settings where multiple targets may be sampled for each training input. Multiple Choice Learning is a simple framework to tackle multimodal density estimation, using the Winner-Takes-All (WTA) loss for a set of hypotheses. In regression settings, the existing MCL variants focus on merging the hypotheses, thereby eventually sacrificing the diversity of the predictions. In contrast, our method relies on a novel learned scoring scheme underpinned by a mathematical framework based on Voronoi tessellations of the output space, from which we can derive a probabilistic interpretation. After empirically validating rMCL with experiments on synthetic data, we further assess its merits on the sound source localization problem, demonstrating its practical usefulness and the relevance of its interpretation.