Gabriel Meseguer-Brocal

Rachel M. Bittner, Juan José Bosch, David Rubinstein et al.

Automatic Music Transcription (AMT) has been recognized as a key enabling technology with a wide range of applications. Given the task's complexity, best results have typically been reported for systems focusing on specific settings, e.g. instrument-specific systems tend to yield improved results over instrument-agnostic methods. Similarly, higher accuracy can be obtained when only estimating frame-wise $f_0$ values and neglecting the harder note event detection. Despite their high accuracy, such specialized systems often cannot be deployed in the real-world. Storage and network constraints prohibit the use of multiple specialized models, while memory and run-time constraints limit their complexity. In this paper, we propose a lightweight neural network for musical instrument transcription, which supports polyphonic outputs and generalizes to a wide variety of instruments (including vocals). Our model is trained to jointly predict frame-wise onsets, multipitch and note activations, and we experimentally show that this multi-output structure improves the resulting frame-level note accuracy. Despite its simplicity, benchmark results show our system's note estimation to be substantially better than a comparable baseline, and its frame-level accuracy to be only marginally below those of specialized state-of-the-art AMT systems. With this work we hope to encourage the community to further investigate low-resource, instrument-agnostic AMT systems.

Dorian Desblancs, Gabriel Meseguer-Brocal, Romain Hennequin et al.

Cloned voices of popular singers sound increasingly realistic and have gained popularity over the past few years. They however pose a threat to the industry due to personality rights concerns. As such, methods to identify the original singer in synthetic voices are needed. In this paper, we investigate how singer identification methods could be used for such a task. We present three embedding models that are trained using a singer-level contrastive learning scheme, where positive pairs consist of segments with vocals from the same singers. These segments can be mixtures for the first model, vocals for the second, and both for the third. We demonstrate that all three models are highly capable of identifying real singers. However, their performance deteriorates when classifying cloned versions of singers in our evaluation set. This is especially true for models that use mixtures as an input. These findings highlight the need to understand the biases that exist within singer identification systems, and how they can influence the identification of voice deepfakes in music.

Gabriel Meseguer-Brocal, Dorian Desblancs, Romain Hennequin

Self-supervised learning has emerged as a powerful way to pre-train generalizable machine learning models on large amounts of unlabeled data. It is particularly compelling in the music domain, where obtaining labeled data is time-consuming, error-prone, and ambiguous. During the self-supervised process, models are trained on pretext tasks, with the primary objective of acquiring robust and informative features that can later be fine-tuned for specific downstream tasks. The choice of the pretext task is critical as it guides the model to shape the feature space with meaningful constraints for information encoding. In the context of music, most works have relied on contrastive learning or masking techniques. In this study, we expand the scope of pretext tasks applied to music by investigating and comparing the performance of new self-supervised methods for music tagging. We open-source a simple ResNet model trained on a diverse catalog of millions of tracks. Our results demonstrate that, although most of these pre-training methods result in similar downstream results, contrastive learning consistently results in better downstream performance compared to other self-supervised pre-training methods. This holds true in a limited-data downstream context.

Joanne Affolter, Benjamin Martin, Elena V. Epure et al.

Music Cover Retrieval, also known as Version Identification, aims to recognize distinct renditions of the same underlying musical work, a task central to catalog management, copyright enforcement, and music retrieval. State-of-the-art approaches have largely focused on harmonic and melodic features, employing increasingly complex audio pipelines designed to be invariant to musical attributes that often vary widely across covers. While effective, these methods demand substantial training time and computational resources. By contrast, lyrics constitute a strong invariant across covers, though their use has been limited by the difficulty of extracting them accurately and efficiently from polyphonic audio. Early methods relied on simple frameworks that limited downstream performance, while more recent systems deliver stronger results but require large models integrated within complex multimodal architectures. We introduce LIVI (Lyrics-Informed Version Identification), an approach that seeks to balance retrieval accuracy with computational efficiency. First, LIVI leverages supervision from state-of-the-art transcription and text embedding models during training to achieve retrieval accuracy on par with--or superior to--harmonic-based systems. Second, LIVI remains lightweight and efficient by removing the transcription step at inference, challenging the dominance of complexity-heavy pipelines.

Markus Frohmann, Elena V. Epure, Gabriel Meseguer-Brocal et al.

The recent rise in capabilities of AI-based music generation tools has created an upheaval in the music industry, necessitating the creation of accurate methods to detect such AI-generated content. This can be done using audio-based detectors; however, it has been shown that they struggle to generalize to unseen generators or when the audio is perturbed. Furthermore, recent work used accurate and cleanly formatted lyrics sourced from a lyrics provider database to detect AI-generated music. However, in practice, such perfect lyrics are not available (only the audio is); this leaves a substantial gap in applicability in real-life use cases. In this work, we instead propose solving this gap by transcribing songs using general automatic speech recognition (ASR) models. We do this using several detectors. The results on diverse, multi-genre, and multi-lingual lyrics show generally strong detection performance across languages and genres, particularly for our best-performing model using Whisper large-v2 and LLM2Vec embeddings. In addition, we show that our method is more robust than state-of-the-art audio-based ones when the audio is perturbed in different ways and when evaluated on different music generators. Our code is available at https://github.com/deezer/robust-AI-lyrics-detection.

Markus Frohmann, Gabriel Meseguer-Brocal, Markus Schedl et al.

The rapid advancement of AI-based music generation tools is revolutionizing the music industry but also posing challenges to artists, copyright holders, and providers alike. This necessitates reliable methods for detecting such AI-generated content. However, existing detectors, relying on either audio or lyrics, face key practical limitations: audio-based detectors fail to generalize to new or unseen generators and are vulnerable to audio perturbations; lyrics-based methods require cleanly formatted and accurate lyrics, unavailable in practice. To overcome these limitations, we propose a novel, practically grounded approach: a multimodal, modular late-fusion pipeline that combines automatically transcribed sung lyrics and speech features capturing lyrics-related information within the audio. By relying on lyrical aspects directly from audio, our method enhances robustness, mitigates susceptibility to low-level artifacts, and enables practical applicability. Experiments show that our method, DE-detect, outperforms existing lyrics-based detectors while also being more robust to audio perturbations. Thus, it offers an effective, robust solution for detecting AI-generated music in real-world scenarios. Our code is available at https://github.com/deezer/robust-AI-lyrics-detection.

Darius Afchar, Gabriel Meseguer-Brocal, Romain Hennequin

In the face of a new era of generative models, the detection of artificially generated content has become a matter of utmost importance. The ability to create credible minute-long music deepfakes in a few seconds on user-friendly platforms poses a real threat of fraud on streaming services and unfair competition to human artists. This paper demonstrates the possibility (and surprising ease) of training classifiers on datasets comprising real audio and fake reconstructions, achieving a convincing accuracy of 99.8%. To our knowledge, this marks the first publication of a music deepfake detector, a tool that will help in the regulation of music forgery. Nevertheless, informed by decades of literature on forgery detection in other fields, we stress that a good test score is not the end of the story. We step back from the straightforward ML framework and expose many facets that could be problematic with such a deployed detector: calibration, robustness to audio manipulation, generalisation to unseen models, interpretability and possibility for recourse. This second part acts as a position for future research steps in the field and a caveat to a flourishing market of fake content checkers.

Yuexuan Kong, Gabriel Meseguer-Brocal, Vincent Lostanlen et al.

In music information retrieval (MIR), contrastive self-supervised learning for general-purpose representation models is effective for global tasks such as automatic tagging. However, for local tasks such as chord estimation, it is widely assumed that contrastively trained general-purpose self-supervised models are inadequate and that more sophisticated SSL is necessary; e.g., masked modeling. Our paper challenges this assumption by revealing the potential of contrastive SSL paired with a transformer in local MIR tasks. We consider a lightweight vision transformer with one-dimensional patches in the time--frequency domain (ViT-1D) and train it with simple contrastive SSL through normalized temperature-scaled cross-entropy loss (NT-Xent). Although NT-Xent operates only over the class token, we observe that, potentially thanks to weight sharing, informative musical properties emerge in ViT-1D's sequence tokens. On global tasks, the temporal average of class and sequence tokens offers a performance increase compared to the class token alone, showing useful properties in the sequence tokens. On local tasks, sequence tokens perform unexpectedly well, despite not being specifically trained for. Furthermore, high-level musical features such as onsets emerge from layer-wise attention maps and self-similarity matrices show different layers capture different musical dimensions. Our paper does not focus on improving performance but advances the musical interpretation of transformers and sheds light on some overlooked abilities of contrastive SSL paired with transformers for sequence modeling in MIR.

Bernardo Torres, Manuel Moussallam, Gabriel Meseguer-Brocal

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.

Yanis Labrak, Markus Frohmann, Gabriel Meseguer-Brocal et al.

In recent years, the use of large language models (LLMs) to generate music content, particularly lyrics, has gained in popularity. These advances provide valuable tools for artists and enhance their creative processes, but they also raise concerns about copyright violations, consumer satisfaction, and content spamming. Previous research has explored content detection in various domains. However, no work has focused on the text modality, lyrics, in music. To address this gap, we curated a diverse dataset of real and synthetic lyrics from multiple languages, music genres, and artists. The generation pipeline was validated using both humans and automated methods. We performed a thorough evaluation of existing synthetic text detection approaches on lyrics, a previously unexplored data type. We also investigated methods to adapt the best-performing features to lyrics through unsupervised domain adaptation. Following both music and industrial constraints, we examined how well these approaches generalize across languages, scale with data availability, handle multilingual language content, and perform on novel genres in few-shot settings. Our findings show promising results that could inform policy decisions around AI-generated music and enhance transparency for users.

Rachel M. Bittner, Katherine Pasalo, Juan José Bosch et al.

To compliment the existing set of datasets, we present a small dataset entitled vocadito, consisting of 40 short excerpts of monophonic singing, sung in 7 different languages by singers with varying of levels of training, and recorded on a variety of devices. We provide several types of annotations, including $f_0$, lyrics, and two different note annotations. All annotations were created by musicians. We provide an analysis of the differences between the two note annotations, and see that the agreement level is low, which has implications for evaluating vocal note estimation algorithms. We also analyze the relation between the $f_0$ and note annotations, and show that quantizing $f_0$ values in frequency does not provide a reasonable note estimate, reinforcing the difficulty of the note estimation task for singing voice. Finally, we provide baseline results from recent algorithms on vocadito for note and $f_0$ transcription. Vocadito is made freely available for public use.

Gabriel Meseguer-Brocal

This dissertation proposes the study of multimodal learning in the context of musical signals. Throughout, we focus on the interaction between audio signals and text information. Among the many text sources related to music that can be used (e.g. reviews, metadata, or social network feedback), we concentrate on lyrics. The singing voice directly connects the audio signal and the text information in a unique way, combining melody and lyrics where a linguistic dimension complements the abstraction of musical instruments. Our study focuses on the audio and lyrics interaction for targeting source separation and informed content estimation.

Gabriel Meseguer-Brocal, Geoffroy Peeters

Informed source separation has recently gained renewed interest with the introduction of neural networks and the availability of large multitrack datasets containing both the mixture and the separated sources. These approaches use prior information about the target source to improve separation. Historically, Music Information Retrieval researchers have focused primarily on score-informed source separation, but more recent approaches explore lyrics-informed source separation. However, because of the lack of multitrack datasets with time-aligned lyrics, models use weak conditioning with non-aligned lyrics. In this paper, we present a multimodal multitrack dataset with lyrics aligned in time at the word level with phonetic information as well as explore strong conditioning using the aligned phonemes. Our model follows a U-Net architecture and takes as input both the magnitude spectrogram of a musical mixture and a matrix with aligned phonetic information. The phoneme matrix is embedded to obtain the parameters that control Feature-wise Linear Modulation (FiLM) layers. These layers condition the U-Net feature maps to adapt the separation process to the presence of different phonemes via affine transformations. We show that phoneme conditioning can be successfully applied to improve singing voice source separation.

Gabriel Meseguer-Brocal, Rachel Bittner, Simon Durand et al.

Data cleansing is a well studied strategy for cleaning erroneous labels in datasets, which has not yet been widely adopted in Music Information Retrieval. Previously proposed data cleansing models do not consider structured (e.g. time varying) labels, such as those common to music data. We propose a novel data cleansing model for time-varying, structured labels which exploits the local structure of the labels, and demonstrate its usefulness for vocal note event annotations in music. %Our model is trained in a contrastive learning manner by automatically creating local deformations of likely correct labels. Our model is trained in a contrastive learning manner by automatically contrasting likely correct labels pairs against local deformations of them. We demonstrate that the accuracy of a transcription model improves greatly when trained using our proposed strategy compared with the accuracy when trained using the original dataset. Additionally we use our model to estimate the annotation error rates in the DALI dataset, and highlight other potential uses for this type of model.

Gabriel Meseguer-Brocal, Geoffroy Peeters

Data-driven models for audio source separation such as U-Net or Wave-U-Net are usually models dedicated to and specifically trained for a single task, e.g. a particular instrument isolation. Training them for various tasks at once commonly results in worse performances than training them for a single specialized task. In this work, we introduce the Conditioned-U-Net (C-U-Net) which adds a control mechanism to the standard U-Net. The control mechanism allows us to train a unique and generic U-Net to perform the separation of various instruments. The C-U-Net decides the instrument to isolate according to a one-hot-encoding input vector. The input vector is embedded to obtain the parameters that control Feature-wise Linear Modulation (FiLM) layers. FiLM layers modify the U-Net feature maps in order to separate the desired instrument via affine transformations. The C-U-Net performs different instrument separations, all with a single model achieving the same performances as the dedicated ones at a lower cost.

Gabriel Meseguer-Brocal, Alice Cohen-Hadria, Geoffroy Peeters

The goal of this paper is twofold. First, we introduce DALI, a large and rich multimodal dataset containing 5358 audio tracks with their time-aligned vocal melody notes and lyrics at four levels of granularity. The second goal is to explain our methodology where dataset creation and learning models interact using a teacher-student machine learning paradigm that benefits each other. We start with a set of manual annotations of draft time-aligned lyrics and notes made by non-expert users of Karaoke games. This set comes without audio. Therefore, we need to find the corresponding audio and adapt the annotations to it. To that end, we retrieve audio candidates from the Web. Each candidate is then turned into a singing-voice probability over time using a teacher, a deep convolutional neural network singing-voice detection system (SVD), trained on cleaned data. Comparing the time-aligned lyrics and the singing-voice probability, we detect matches and update the time-alignment lyrics accordingly. From this, we obtain new audio sets. They are then used to train new SVD students used to perform again the above comparison. The process could be repeated iteratively. We show that this allows to progressively improve the performances of our SVD and get better audio-matching and alignment.