Xavier Serra

Rajsuryan Singh, Pablo Zinemanas, Xavier Serra et al.

Most recent work in visual sound source localization relies on semantic audio-visual representations learned in a self-supervised manner, and by design excludes temporal information present in videos. While it proves to be effective for widely used benchmark datasets, the method falls short for challenging scenarios like urban traffic. This work introduces temporal context into the state-of-the-art methods for sound source localization in urban scenes using optical flow as a means to encode motion information. An analysis of the strengths and weaknesses of our methods helps us better understand the problem of visual sound source localization and sheds light on open challenges for audio-visual scene understanding.

Jose J. Valero-Mas, Antonio Javier Gallego, Pablo Alonso-Jiménez et al.

Prototype Generation (PG) methods are typically considered for improving the efficiency of the $k$-Nearest Neighbour ($k$NN) classifier when tackling high-size corpora. Such approaches aim at generating a reduced version of the corpus without decreasing the classification performance when compared to the initial set. Despite their large application in multiclass scenarios, very few works have addressed the proposal of PG methods for the multilabel space. In this regard, this work presents the novel adaptation of four multiclass PG strategies to the multilabel case. These proposals are evaluated with three multilabel $k$NN-based classifiers, 12 corpora comprising a varied range of domains and corpus sizes, and different noise scenarios artificially induced in the data. The results obtained show that the proposed adaptations are capable of significantly improving -- both in terms of efficiency and classification performance -- the only reference multilabel PG work in the literature as well as the case in which no PG method is applied, also presenting a statistically superior robustness in noisy scenarios. Moreover, these novel PG strategies allow prioritising either the efficiency or efficacy criteria through its configuration depending on the target scenario, hence covering a wide area in the solution space not previously filled by other works.

Roser Batlle-Roca, Wei-Hsiang Liao, Xavier Serra et al.

Recent advancements in music generation are raising multiple concerns about the implications of AI in creative music processes, current business models and impacts related to intellectual property management. A relevant discussion and related technical challenge is the potential replication and plagiarism of the training set in AI-generated music, which could lead to misuse of data and intellectual property rights violations. To tackle this issue, we present the Music Replication Assessment (MiRA) tool: a model-independent open evaluation method based on diverse audio music similarity metrics to assess data replication. We evaluate the ability of five metrics to identify exact replication by conducting a controlled replication experiment in different music genres using synthetic samples. Our results show that the proposed methodology can estimate exact data replication with a proportion higher than 10%. By introducing the MiRA tool, we intend to encourage the open evaluation of music-generative models by researchers, developers, and users concerning data replication, highlighting the importance of the ethical, social, legal, and economic consequences. Code and examples are available for reproducibility purposes.

Benno Weck, Xavier Serra

The recent progress in text-based audio retrieval was largely propelled by the release of suitable datasets. Since the manual creation of such datasets is a laborious task, obtaining data from online resources can be a cheap solution to create large-scale datasets. We study the recently proposed SoundDesc benchmark dataset, which was automatically sourced from the BBC Sound Effects web page. In our analysis, we find that SoundDesc contains several duplicates that cause leakage of training data to the evaluation data. This data leakage ultimately leads to overly optimistic retrieval performance estimates in previous benchmarks. We propose new training, validation, and testing splits for the dataset that we make available online. To avoid weak contamination of the test data, we pool audio files that share similar recording setups. In our experiments, we find that the new splits serve as a more challenging benchmark.

Benno Weck, Miguel Pérez Fernández, Holger Kirchhoff et al.

We present an analysis of large-scale pretrained deep learning models used for cross-modal (text-to-audio) retrieval. We use embeddings extracted by these models in a metric learning framework to connect matching pairs of audio and text. Shallow neural networks map the embeddings to a common dimensionality. Our system, which is an extension of our submission to the Language-based Audio Retrieval Task of the DCASE Challenge 2022, employs the RoBERTa foundation model as the text embedding extractor. A pretrained PANNs model extracts the audio embeddings. To improve the generalisation of our model, we investigate how pretraining with audio and associated noisy text collected from the online platform Freesound improves the performance of our method. Furthermore, our ablation study reveals that the proper choice of the loss function and fine-tuning the pretrained models are essential in training a competitive retrieval system.

Pedro Ramoneda, Emilia Parada-Cabaleiro, Benno Weck et al.

In this work, we explore the use and reliability of Large Language Models (LLMs) in musicology. From a discussion with experts and students, we assess the current acceptance and concerns regarding this, nowadays ubiquitous, technology. We aim to go one step further, proposing a semi-automatic method to create an initial benchmark using retrieval-augmented generation models and multiple-choice question generation, validated by human experts. Our evaluation on 400 human-validated questions shows that current vanilla LLMs are less reliable than retrieval augmented generation from music dictionaries. This paper suggests that the potential of LLMs in musicology requires musicology driven research that can specialized LLMs by including accurate and reliable domain knowledge.

Pedro Ramoneda, Vsevolod Eremenko, Alexandre D'Hooge et al.

Estimating music piece difficulty is important for organizing educational music collections. This process could be partially automatized to facilitate the educator's role. Nevertheless, the decisions performed by prevalent deep-learning models are hardly understandable, which may impair the acceptance of such a technology in music education curricula. Our work employs explainable descriptors for difficulty estimation in symbolic music representations. Furthermore, through a novel parameter-efficient white-box model, we outperform previous efforts while delivering interpretable results. These comprehensible outcomes emulate the functionality of a rubric, a tool widely used in music education. Our approach, evaluated in piano repertoire categorized in 9 classes, achieved 41.4% accuracy independently, with a mean squared error (MSE) of 1.7, showing precise difficulty estimation. Through our baseline, we illustrate how building on top of past research can offer alternatives for music difficulty assessment which are explainable and interpretable. With this, we aim to promote a more effective communication between the Music Information Retrieval (MIR) community and the music education one.

Laura Ibáñez-Martínez, Chukwuemeka Nkama, Andrea Poltronieri et al.

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.

Genís Plaja-Roglans, Yun-Ning Hung, Xavier Serra et al.

Separating the individual elements in a musical mixture is an essential process for music analysis and practice. While this is generally addressed using neural networks optimized to mask or transform the time-frequency representation of a mixture to extract the target sources, the flexibility and generalization capabilities of generative diffusion models are giving rise to a novel class of solutions for this complicated task. In this work, we explore singing voice separation from real music recordings using a diffusion model which is trained to generate the solo vocals conditioned on the corresponding mixture. Our approach improves upon prior generative systems and achieves competitive objective scores against non-generative baselines when trained with supplementary data. The iterative nature of diffusion sampling enables the user to control the quality-efficiency trade-off, and also refine the output when needed. We present an ablation study of the sampling algorithm, highlighting the effects of the user-configurable parameters.

António Ramires, Gilberto Bernardes, Matthew E. P. Davies et al.

In this paper, we present TIV.lib, an open-source library for the content-based tonal description of musical audio signals. Its main novelty relies on the perceptually-inspired Tonal Interval Vector space based on the Discrete Fourier transform, from which multiple instantaneous and global representations, descriptors and metrics are computed - e.g., harmonic change, dissonance, diatonicity, and musical key. The library is cross-platform, implemented in Python and the graphical programming language Pure Data, and can be used in both online and offline scenarios. Of note is its potential for enhanced Music Information Retrieval, where tonal descriptors sit at the core of numerous methods and applications.

Pablo Alonso-Jiménez, Dmitry Bogdanov, Jordi Pons et al.

Essentia is a reference open-source C++/Python library for audio and music analysis. In this work, we present a set of algorithms that employ TensorFlow in Essentia, allow predictions with pre-trained deep learning models, and are designed to offer flexibility of use, easy extensibility, and real-time inference. To show the potential of this new interface with TensorFlow, we provide a number of pre-trained state-of-the-art music tagging and classification CNN models. We run an extensive evaluation of the developed models. In particular, we assess the generalization capabilities in a cross-collection evaluation utilizing both external tag datasets as well as manual annotations tailored to the taxonomies of our models.

Jordi Pons, Xavier Serra

Pronounced as "musician", the musicnn library contains a set of pre-trained musically motivated convolutional neural networks for music audio tagging: https://github.com/jordipons/musicnn. This repository also includes some pre-trained vgg-like baselines. These models can be used as out-of-the-box music audio taggers, as music feature extractors, or as pre-trained models for transfer learning. We also provide the code to train the aforementioned models: https://github.com/jordipons/musicnn-training. This framework also allows implementing novel models. For example, a musically motivated convolutional neural network with an attention-based output layer (instead of the temporal pooling layer) can achieve state-of-the-art results for music audio tagging: 90.77 ROC-AUC / 38.61 PR-AUC on the MagnaTagATune dataset --- and 88.81 ROC-AUC / 31.51 PR-AUC on the Million Song Dataset.

Pablo Alonso-Jiménez, Leonardo Pepino, Roser Batlle-Roca et al.

We present PECMAE, an interpretable model for music audio classification based on prototype learning. Our model is based on a previous method, APNet, which jointly learns an autoencoder and a prototypical network. Instead, we propose to decouple both training processes. This enables us to leverage existing self-supervised autoencoders pre-trained on much larger data (EnCodecMAE), providing representations with better generalization. APNet allows prototypes' reconstruction to waveforms for interpretability relying on the nearest training data samples. In contrast, we explore using a diffusion decoder that allows reconstruction without such dependency. We evaluate our method on datasets for music instrument classification (Medley-Solos-DB) and genre recognition (GTZAN and a larger in-house dataset), the latter being a more challenging task not addressed with prototypical networks before. We find that the prototype-based models preserve most of the performance achieved with the autoencoder embeddings, while the sonification of prototypes benefits understanding the behavior of the classifier.

Benno Weck, Holger Kirchhoff, Peter Grosche et al.

Multi-modal deep learning techniques for matching free-form text with music have shown promising results in the field of Music Information Retrieval (MIR). Prior work is often based on large proprietary data while publicly available datasets are few and small in size. In this study, we present WikiMuTe, a new and open dataset containing rich semantic descriptions of music. The data is sourced from Wikipedia's rich catalogue of articles covering musical works. Using a dedicated text-mining pipeline, we extract both long and short-form descriptions covering a wide range of topics related to music content such as genre, style, mood, instrumentation, and tempo. To show the use of this data, we train a model that jointly learns text and audio representations and performs cross-modal retrieval. The model is evaluated on two tasks: tag-based music retrieval and music auto-tagging. The results show that while our approach has state-of-the-art performance on multiple tasks, but still observe a difference in performance depending on the data used for training.

Enric Gusó, Joanna Luberadzka, Umut Sayin et al.

We investigate the effects of four strategies for improving the ecological validity of synthetic room impulse response (RIR) datasets for monoaural Speech Enhancement (SE). We implement three features on top of the traditional image source method-based (ISM) shoebox RIRs: multiband absorption coefficients, source directivity and receiver directivity. We additionally consider mesh-based RIRs from the SoundSpaces dataset. We then train a DeepFilternet3 model for each RIR dataset and evaluate the performance on a test set of real RIRs both objectively and subjectively. We find that RIRs which use frequency-dependent acoustic absorption coefficients (MB-RIRs) can obtain +0.51dB of SDR and a +8.9 MUSHRA score when evaluated on real RIRs. The MB-RIRs dataset is publicly available for free download.

Roser Batlle-Roca, Laura Ibáñez-Martínez, Xavier Serra et al.

Since 2023, generative AI has rapidly advanced in the music domain. Despite significant technological advancements, music-generative models raise critical ethical challenges, including a lack of transparency and accountability, along with risks such as the replication of artists' works, which highlights the importance of fostering openness. With upcoming regulations such as the EU AI Act encouraging open models, many generative models are being released labelled as 'open'. However, the definition of an open model remains widely debated. In this article, we adapt a recently proposed evidence-based framework for assessing openness in LLMs to the music domain. Using feedback from a survey of 110 participants from the Music Information Retrieval (MIR) community, we refine the framework into MusGO (Music-Generative Open AI), which comprises 13 openness categories: 8 essential and 5 desirable. We evaluate 16 state-of-the-art generative models and provide an openness leaderboard that is fully open to public scrutiny and community contributions. Through this work, we aim to clarify the concept of openness in music-generative AI and promote its transparent and responsible development.

Genís Plaja-Roglans, Yun-Ning Hung, Xavier Serra et al.

Extracting individual elements from music mixtures is a valuable tool for music production and practice. While neural networks optimized to mask or transform mixture spectrograms into the individual source(s) have been the leading approach, the source overlap and correlation in music signals poses an inherent challenge. Also, accessing all sources in the mixture is crucial to train these systems, while complicated. Attempts to address these challenges in a generative fashion exist, however, the separation performance and inference efficiency remain limited. In this work, we study the potential of diffusion models to advance toward bridging this gap, focusing on generative singing voice separation relying only on corresponding pairs of isolated vocals and mixtures for training. To align with creative workflows, we leverage latent diffusion: the system generates samples encoded in a compact latent space, and subsequently decodes these into audio. This enables efficient optimization and faster inference. Our system is trained using only open data. We outperform existing generative separation systems, and level the compared non-generative systems on a list of signal quality measures and on interference removal. We provide a noise robustness study on the latent encoder, providing insights on its potential for the task. We release a modular toolkit for further research on the topic.

Andrea Poltronieri, Xavier Serra, Martín Rocamora

Audio Chord Estimation (ACE) holds a pivotal role in music information research, having garnered attention for over two decades due to its relevance for music transcription and analysis. Despite notable advancements, challenges persist in the task, particularly concerning unique characteristics of harmonic content, which have resulted in existing systems' performances reaching a glass ceiling. These challenges include annotator subjectivity, where varying interpretations among annotators lead to inconsistencies, and class imbalance within chord datasets, where certain chord classes are over-represented compared to others, posing difficulties in model training and evaluation. As a first contribution, this paper presents an evaluation of inter-annotator agreement in chord annotations, using metrics that extend beyond traditional binary measures. In addition, we propose a consonance-informed distance metric that reflects the perceptual similarity between harmonic annotations. Our analysis suggests that consonance-based distance metrics more effectively capture musically meaningful agreement between annotations. Expanding on these findings, we introduce a novel ACE conformer-based model that integrates consonance concepts into the model through consonance-based label smoothing. The proposed model also addresses class imbalance by separately estimating root, bass, and all note activations, enabling the reconstruction of chord labels from decomposed outputs.

Minz Won, Justin Salamon, Nicholas J. Bryan et al.

Content creators often use music to enhance their stories, as it can be a powerful tool to convey emotion. In this paper, our goal is to help creators find music to match the emotion of their story. We focus on text-based stories that can be auralized (e.g., books), use multiple sentences as input queries, and automatically retrieve matching music. We formalize this task as a cross-modal text-to-music retrieval problem. Both the music and text domains have existing datasets with emotion labels, but mismatched emotion vocabularies prevent us from using mood or emotion annotations directly for matching. To address this challenge, we propose and investigate several emotion embedding spaces, both manually defined (e.g., valence/arousal) and data-driven (e.g., Word2Vec and metric learning) to bridge this gap. Our experiments show that by leveraging these embedding spaces, we are able to successfully bridge the gap between modalities to facilitate cross modal retrieval. We show that our method can leverage the well established valence-arousal space, but that it can also achieve our goal via data-driven embedding spaces. By leveraging data-driven embeddings, our approach has the potential of being generalized to other retrieval tasks that require broader or completely different vocabularies.

Minz Won, Keunwoo Choi, Xavier Serra

We present Music Tagging Transformer that is trained with a semi-supervised approach. The proposed model captures local acoustic characteristics in shallow convolutional layers, then temporally summarizes the sequence of the extracted features using stacked self-attention layers. Through a careful model assessment, we first show that the proposed architecture outperforms the previous state-of-the-art music tagging models that are based on convolutional neural networks under a supervised scheme. The Music Tagging Transformer is further improved by noisy student training, a semi-supervised approach that leverages both labeled and unlabeled data combined with data augmentation. To our best knowledge, this is the first attempt to utilize the entire audio of the million song dataset.

Benno Weck, Xavier Favory, Konstantinos Drossos et al.

Automated audio captioning (AAC) is the task of automatically generating textual descriptions for general audio signals. A captioning system has to identify various information from the input signal and express it with natural language. Existing works mainly focus on investigating new methods and try to improve their performance measured on existing datasets. Having attracted attention only recently, very few works on AAC study the performance of existing pre-trained audio and natural language processing resources. In this paper, we evaluate the performance of off-the-shelf models with a Transformer-based captioning approach. We utilize the freely available Clotho dataset to compare four different pre-trained machine listening models, four word embedding models, and their combinations in many different settings. Our evaluation suggests that YAMNet combined with BERT embeddings produces the best captions. Moreover, in general, fine-tuning pre-trained word embeddings can lead to better performance. Finally, we show that sequences of audio embeddings can be processed using a Transformer encoder to produce higher-quality captions.

Magdalena Fuentes, Justin Salamon, Pablo Zinemanas et al.

Soundata is a Python library for loading and working with audio datasets in a standardized way, removing the need for writing custom loaders in every project, and improving reproducibility by providing tools to validate data against a canonical version. It speeds up research pipelines by allowing users to quickly download a dataset, load it into memory in a standardized and reproducible way, validate that the dataset is complete and correct, and more. Soundata is based and inspired on mirdata and design to complement mirdata by working with environmental sound, bioacoustic and speech datasets, among others. Soundata was created to be easy to use, easy to contribute to, and to increase reproducibility and standardize usage of sound datasets in a flexible way.

Eduardo Fonseca, Andres Ferraro, Xavier Serra

Recent studies have put into question the commonly assumed shift invariance property of convolutional networks, showing that small shifts in the input can affect the output predictions substantially. In this paper, we analyze the benefits of addressing lack of shift invariance in CNN-based sound event classification. Specifically, we evaluate two pooling methods to improve shift invariance in CNNs, based on low-pass filtering and adaptive sampling of incoming feature maps. These methods are implemented via small architectural modifications inserted into the pooling layers of CNNs. We evaluate the effect of these architectural changes on the FSD50K dataset using models of different capacity and in presence of strong regularization. We show that these modifications consistently improve sound event classification in all cases considered. We also demonstrate empirically that the proposed pooling methods increase shift invariance in the network, making it more robust against time/frequency shifts in input spectrograms. This is achieved by adding a negligible amount of trainable parameters, which makes these methods an appealing alternative to conventional pooling layers. The outcome is a new state-of-the-art mAP of 0.541 on the FSD50K classification benchmark.

Andres Ferraro, Xavier Serra, Christine Bauer

Music streaming platforms are currently among the main sources of music consumption, and the embedded recommender systems significantly influence what the users consume. There is an increasing interest to ensure that those platforms and systems are fair. Yet, we first need to understand what fairness means in such a context. Although artists are the main content providers for music platforms, there is a research gap concerning the artists' perspective. To fill this gap, we conducted interviews with music artists to understand how they are affected by current platforms and what improvements they deem necessary. Using a Qualitative Content Analysis, we identify the aspects that the artists consider relevant for fair platforms. In this paper, we discuss the following aspects derived from the interviews: fragmented presentation, reaching an audience, transparency, influencing users' listening behavior, popularity bias, artists' repertoire size, quotas for local music, gender balance, and new music. For some topics, our findings do not indicate a clear direction about the best way how music platforms should act and function; for other topics, though, there is a clear consensus among our interviewees: for these, the artists have a clear idea of the actions that should be taken so that music platforms will be fair also for the artists.

Pritish Chandna, António Ramires, Xavier Serra et al.

Loops, seamlessly repeatable musical segments, are a cornerstone of modern music production. Contemporary artists often mix and match various sampled or pre-recorded loops based on musical criteria such as rhythm, harmony and timbral texture to create compositions. Taking such criteria into account, we present LoopNet, a feed-forward generative model for creating loops conditioned on intuitive parameters. We leverage Music Information Retrieval (MIR) models as well as a large collection of public loop samples in our study and use the Wave-U-Net architecture to map control parameters to audio. We also evaluate the quality of the generated audio and propose intuitive controls for composers to map the ideas in their minds to an audio loop.

Eduardo Fonseca, Aren Jansen, Daniel P. W. Ellis et al.

Real-world sound scenes consist of time-varying collections of sound sources, each generating characteristic sound events that are mixed together in audio recordings. The association of these constituent sound events with their mixture and each other is semantically constrained: the sound scene contains the union of source classes and not all classes naturally co-occur. With this motivation, this paper explores the use of unsupervised automatic sound separation to decompose unlabeled sound scenes into multiple semantically-linked views for use in self-supervised contrastive learning. We find that learning to associate input mixtures with their automatically separated outputs yields stronger representations than past approaches that use the mixtures alone. Further, we discover that optimal source separation is not required for successful contrastive learning by demonstrating that a range of separation system convergence states all lead to useful and often complementary example transformations. Our best system incorporates these unsupervised separation models into a single augmentation front-end and jointly optimizes similarity maximization and coincidence prediction objectives across the views. The result is an unsupervised audio representation that rivals state-of-the-art alternatives on the established shallow AudioSet classification benchmark.

Andres Ferraro, Yuntae Kim, Soohyeon Lee et al.

One of the main limitations in the field of audio signal processing is the lack of large public datasets with audio representations and high-quality annotations due to restrictions of copyrighted commercial music. We present Melon Playlist Dataset, a public dataset of mel-spectrograms for 649,091tracks and 148,826 associated playlists annotated by 30,652 different tags. All the data is gathered from Melon, a popular Korean streaming service. The dataset is suitable for music information retrieval tasks, in particular, auto-tagging and automatic playlist continuation. Even though the latter can be addressed by collaborative filtering approaches, audio provides opportunities for research on track suggestions and building systems resistant to the cold-start problem, for which we provide a baseline. Moreover, the playlists and the annotations included in the Melon Playlist Dataset make it suitable for metric learning and representation learning.

Eduardo Fonseca, Diego Ortego, Kevin McGuinness et al.

Self-supervised representation learning can mitigate the limitations in recognition tasks with few manually labeled data but abundant unlabeled data---a common scenario in sound event research. In this work, we explore unsupervised contrastive learning as a way to learn sound event representations. To this end, we propose to use the pretext task of contrasting differently augmented views of sound events. The views are computed primarily via mixing of training examples with unrelated backgrounds, followed by other data augmentations. We analyze the main components of our method via ablation experiments. We evaluate the learned representations using linear evaluation, and in two in-domain downstream sound event classification tasks, namely, using limited manually labeled data, and using noisy labeled data. Our results suggest that unsupervised contrastive pre-training can mitigate the impact of data scarcity and increase robustness against noisy labels, outperforming supervised baselines.

Minz Won, Sergio Oramas, Oriol Nieto et al.

Tag-based music retrieval is crucial to browse large-scale music libraries efficiently. Hence, automatic music tagging has been actively explored, mostly as a classification task, which has an inherent limitation: a fixed vocabulary. On the other hand, metric learning enables flexible vocabularies by using pretrained word embeddings as side information. Also, metric learning has already proven its suitability for cross-modal retrieval tasks in other domains (e.g., text-to-image) by jointly learning a multimodal embedding space. In this paper, we investigate three ideas to successfully introduce multimodal metric learning for tag-based music retrieval: elaborate triplet sampling, acoustic and cultural music information, and domain-specific word embeddings. Our experimental results show that the proposed ideas enhance the retrieval system quantitatively, and qualitatively. Furthermore, we release the MSD500, a subset of the Million Song Dataset (MSD) containing 500 cleaned tags, 7 manually annotated tag categories, and user taste profiles.

Xavier Favory, Konstantinos Drossos, Tuomas Virtanen et al.

Self-supervised audio representation learning offers an attractive alternative for obtaining generic audio embeddings, capable to be employed into various downstream tasks. Published approaches that consider both audio and words/tags associated with audio do not employ text processing models that are capable to generalize to tags unknown during training. In this work we propose a method for learning audio representations using an audio autoencoder (AAE), a general word embeddings model (WEM), and a multi-head self-attention (MHA) mechanism. MHA attends on the output of the WEM, providing a contextualized representation of the tags associated with the audio, and we align the output of MHA with the output of the encoder of AAE using a contrastive loss. We jointly optimize AAE and MHA and we evaluate the audio representations (i.e. the output of the encoder of AAE) by utilizing them in three different downstream tasks, namely sound, music genre, and music instrument classification. Our results show that employing multi-head self-attention with multiple heads in the tag-based network can induce better learned audio representations.

Eduardo Fonseca, Xavier Favory, Jordi Pons et al.

Most existing datasets for sound event recognition (SER) are relatively small and/or domain-specific, with the exception of AudioSet, based on over 2M tracks from YouTube videos and encompassing over 500 sound classes. However, AudioSet is not an open dataset as its official release consists of pre-computed audio features. Downloading the original audio tracks can be problematic due to YouTube videos gradually disappearing and usage rights issues. To provide an alternative benchmark dataset and thus foster SER research, we introduce FSD50K, an open dataset containing over 51k audio clips totalling over 100h of audio manually labeled using 200 classes drawn from the AudioSet Ontology. The audio clips are licensed under Creative Commons licenses, making the dataset freely distributable (including waveforms). We provide a detailed description of the FSD50K creation process, tailored to the particularities of Freesound data, including challenges encountered and solutions adopted. We include a comprehensive dataset characterization along with discussion of limitations and key factors to allow its audio-informed usage. Finally, we conduct sound event classification experiments to provide baseline systems as well as insight on the main factors to consider when splitting Freesound audio data for SER. Our goal is to develop a dataset to be widely adopted by the community as a new open benchmark for SER research.

Antonio Ramires, Frederic Font, Dmitry Bogdanov et al.

Music loops are essential ingredients in electronic music production, and there is a high demand for pre-recorded loops in a variety of styles. Several commercial and community databases have been created to meet this demand, but most are not suitable for research due to their strict licensing. We present the Freesound Loop Dataset (FSLD), a new large-scale dataset of music loops annotated by experts. The loops originate from Freesound, a community database of audio recordings released under Creative Commons licenses, so the audio in our dataset may be redistributed. The annotations include instrument, tempo, meter, key and genre tags. We describe the methodology used to assemble and annotate the data, and report on the distribution of tags in the data and inter-annotator agreement. We also present to the community an online loop annotator tool that we developed. To illustrate the usefulness of FSLD, we present short case studies on using it to estimate tempo and key, generate music tracks, and evaluate a loop separation algorithm. We anticipate that the community will find yet more uses for the data, in applications from automatic loop characterisation to algorithmic composition.

Andres Ferraro, Dietmar Jannach, Xavier Serra

Session-based recommendation is a problem setting where the task of a recommender system is to make suitable item suggestions based only on a few observed user interactions in an ongoing session. The lack of long-term preference information about individual users in such settings usually results in a limited level of personalization, where a small set of popular items may be recommended to many users. This repeated exposure of such a subset of the items through the recommendations may in turn lead to a reinforcement effect over time, and to a system which is not able to help users discover new content anymore to the desirable extent. In this work, we investigate such potential longitudinal effects of session-based recommendations in a simulation-based approach. Specifically, we analyze to what extent algorithms of different types may lead to concentration effects over time. Our experiments in the music domain reveal that all investigated algorithms---both neural and heuristic ones---may lead to lower item coverage and to a higher concentration on a subset of the items. Additional simulation experiments however also indicate that relatively simple re-ranking strategies, e.g., by avoiding too many repeated recommendations in the music domain, may help to deal with this problem.

Xavier Favory, Konstantinos Drossos, Tuomas Virtanen et al.

Audio representation learning based on deep neural networks (DNNs) emerged as an alternative approach to hand-crafted features. For achieving high performance, DNNs often need a large amount of annotated data which can be difficult and costly to obtain. In this paper, we propose a method for learning audio representations, aligning the learned latent representations of audio and associated tags. Aligning is done by maximizing the agreement of the latent representations of audio and tags, using a contrastive loss. The result is an audio embedding model which reflects acoustic and semantic characteristics of sounds. We evaluate the quality of our embedding model, measuring its performance as a feature extractor on three different tasks (namely, sound event recognition, and music genre and musical instrument classification), and investigate what type of characteristics the model captures. Our results are promising, sometimes in par with the state-of-the-art in the considered tasks and the embeddings produced with our method are well correlated with some acoustic descriptors.

Minz Won, Andres Ferraro, Dmitry Bogdanov et al.

Recent advances in deep learning accelerated the development of content-based automatic music tagging systems. Music information retrieval (MIR) researchers proposed various architecture designs, mainly based on convolutional neural networks (CNNs), that achieve state-of-the-art results in this multi-label binary classification task. However, due to the differences in experimental setups followed by researchers, such as using different dataset splits and software versions for evaluation, it is difficult to compare the proposed architectures directly with each other. To facilitate further research, in this paper we conduct a consistent evaluation of different music tagging models on three datasets (MagnaTagATune, Million Song Dataset, and MTG-Jamendo) and provide reference results using common evaluation metrics (ROC-AUC and PR-AUC). Furthermore, all the models are evaluated with perturbed inputs to investigate the generalization capabilities concerning time stretch, pitch shift, dynamic range compression, and addition of white noise. For reproducibility, we provide the PyTorch implementations with the pre-trained models.

Eduardo Fonseca, Shawn Hershey, Manoj Plakal et al.

The study of label noise in sound event recognition has recently gained attention with the advent of larger and noisier datasets. This work addresses the problem of missing labels, one of the big weaknesses of large audio datasets, and one of the most conspicuous issues for AudioSet. We propose a simple and model-agnostic method based on a teacher-student framework with loss masking to first identify the most critical missing label candidates, and then ignore their contribution during the learning process. We find that a simple optimisation of the training label set improves recognition performance without additional computation. We discover that most of the improvement comes from ignoring a critical tiny portion of the missing labels. We also show that the damage done by missing labels is larger as the training set gets smaller, yet it can still be observed even when training with massive amounts of audio. We believe these insights can generalize to other large-scale datasets.

Xavier Favory, Frederic Font, Xavier Serra

The large size of nowadays' online multimedia databases makes retrieving their content a difficult and time-consuming task. Users of online sound collections typically submit search queries that express a broad intent, often making the system return large and unmanageable result sets. Search Result Clustering is a technique that organises search-result content into coherent groups, which allows users to identify useful subsets in their results. Obtaining coherent and distinctive clusters that can be explored with a suitable interface is crucial for making this technique a useful complement of traditional search engines. In our work, we propose a graph-based approach using audio features for clustering diverse sound collections obtained when querying large online databases. We propose an approach to assess the performance of different features at scale, by taking advantage of the metadata associated with each sound. This analysis is complemented with an evaluation using ground-truth labels from manually annotated datasets. We show that using a confidence measure for discarding inconsistent clusters improves the quality of the partitions. After identifying the most appropriate features for clustering, we conduct an experiment with users performing a sound design task, in order to evaluate our approach and its user interface. A qualitative analysis is carried out including usability questionnaires and semi-structured interviews. This provides us with valuable new insights regarding the features that promote efficient interaction with the clusters.

António Ramires, Pritish Chandna, Xavier Favory et al.

We present a deep neural network-based methodology for synthesising percussive sounds with control over high-level timbral characteristics of the sounds. This approach allows for intuitive control of a synthesizer, enabling the user to shape sounds without extensive knowledge of signal processing. We use a feedforward convolutional neural network-based architecture, which is able to map input parameters to the corresponding waveform. We propose two datasets to evaluate our approach on both a restrictive context, and in one covering a broader spectrum of sounds. The timbral features used as parameters are taken from recent literature in signal processing. We also use these features for evaluation and validation of the presented model, to ensure that changing the input parameters produces a congruent waveform with the desired characteristics. Finally, we evaluate the quality of the output sound using a subjective listening test. We provide sound examples and the system's source code for reproducibility.

Andres Ferraro, Dmitry Bogdanov, Xavier Serra et al.

Algorithms have an increasing influence on the music that we consume and understanding their behavior is fundamental to make sure they give a fair exposure to all artists across different styles. In this on-going work we contribute to this research direction analyzing the impact of collaborative filtering recommendations from the perspective of artist and music style exposure given by the system. We first analyze the distribution of the recommendations considering the exposure of different styles or genres and compare it to the users' listening behavior. This comparison suggests that the system is reinforcing the popularity of the items. Then, we simulate the effect of the system in the long term with a feedback loop. From this simulation we can see how the system gives less opportunity to the majority of artists, concentrating the users on fewer items. The results of our analysis demonstrate the need for a better evaluation methodology for current music recommendation algorithms, not only limited to user-focused relevance metrics.

Andres Ferraro, Dmitry Bogdanov, Xavier Serra et al.

Automatic tagging of music is an important research topic in Music Information Retrieval and audio analysis algorithms proposed for this task have achieved improvements with advances in deep learning. In particular, many state-of-the-art systems use Convolutional Neural Networks and operate on mel-spectrogram representations of the audio. In this paper, we compare commonly used mel-spectrogram representations and evaluate model performances that can be achieved by reducing the input size in terms of both lesser amount of frequency bands and larger frame rates. We use the MagnaTagaTune dataset for comprehensive performance comparisons and then compare selected configurations on the larger Million Song Dataset. The results of this study can serve researchers and practitioners in their trade-off decision between accuracy of the models, data storage size and training and inference times.

Minz Won, Sanghyuk Chun, Xavier Serra

Recently, we proposed a self-attention based music tagging model. Different from most of the conventional deep architectures in music information retrieval, which use stacked 3x3 filters by treating music spectrograms as images, the proposed self-attention based model attempted to regard music as a temporal sequence of individual audio events. Not only the performance, but it could also facilitate better interpretability. In this paper, we mainly focus on visualizing and understanding the proposed self-attention based music tagging model.

Eduardo Fonseca, Frederic Font, Xavier Serra

Label noise is emerging as a pressing issue in sound event classification. This arises as we move towards larger datasets that are difficult to annotate manually, but it is even more severe if datasets are collected automatically from online repositories, where labels are inferred through automated heuristics applied to the audio content or metadata. While learning from noisy labels has been an active area of research in computer vision, it has received little attention in sound event classification. Most recent computer vision approaches against label noise are relatively complex, requiring complex networks or extra data resources. In this work, we evaluate simple and efficient model-agnostic approaches to handling noisy labels when training sound event classifiers, namely label smoothing regularization, mixup and noise-robust loss functions. The main advantage of these methods is that they can be easily incorporated to existing deep learning pipelines without need for network modifications or extra resources. We report results from experiments conducted with the FSDnoisy18k dataset. We show that these simple methods can be effective in mitigating the effect of label noise, providing up to 2.5\% of accuracy boost when incorporated to two different CNNs, while requiring minimal intervention and computational overhead.

Andres Perez-Lopez, Eduardo Fonseca, Xavier Serra

This work describes and discusses an algorithm submitted to the Sound Event Localization and Detection Task of DCASE2019 Challenge. The proposed methodology relies on parametric spatial audio analysis for source localization and detection, combined with a deep learning-based monophonic event classifier. The evaluation of the proposed algorithm yields overall results comparable to the baseline system. The main highlight is a reduction of the localization error on the evaluation dataset by a factor of 2.6, compared with the baseline performance.

António Ramires, Xavier Serra

Reusing recorded sounds (sampling) is a key component in Electronic Music Production (EMP), which has been present since its early days and is at the core of genres like hip-hop or jungle. Commercial and non-commercial services allow users to obtain collections of sounds (sample packs) to reuse in their compositions. Automatic classification of one-shot instrumental sounds allows automatically categorising the sounds contained in these collections, allowing easier navigation and better characterisation. Automatic instrument classification has mostly targeted the classification of unprocessed isolated instrumental sounds or detecting predominant instruments in mixed music tracks. For this classification to be useful in audio databases for EMP, it has to be robust to the audio effects applied to unprocessed sounds. In this paper we evaluate how a state of the art model trained with a large dataset of one-shot instrumental sounds performs when classifying instruments processed with audio effects. In order to evaluate the robustness of the model, we use data augmentation with audio effects and evaluate how each effect influences the classification accuracy.

Minz Won, Sanghyuk Chun, Xavier Serra

Self-attention is an attention mechanism that learns a representation by relating different positions in the sequence. The transformer, which is a sequence model solely based on self-attention, and its variants achieved state-of-the-art results in many natural language processing tasks. Since music composes its semantics based on the relations between components in sparse positions, adopting the self-attention mechanism to solve music information retrieval (MIR) problems can be beneficial. Hence, we propose a self-attention based deep sequence model for music tagging. The proposed architecture consists of shallow convolutional layers followed by stacked Transformer encoders. Compared to conventional approaches using fully convolutional or recurrent neural networks, our model is more interpretable while reporting competitive results. We validate the performance of our model with the MagnaTagATune and the Million Song Dataset. In addition, we demonstrate the interpretability of the proposed architecture with a heat map visualization.

Eduardo Fonseca, Manoj Plakal, Frederic Font et al.

This paper introduces Task 2 of the DCASE2019 Challenge, titled "Audio tagging with noisy labels and minimal supervision". This task was hosted on the Kaggle platform as "Freesound Audio Tagging 2019". The task evaluates systems for multi-label audio tagging using a large set of noisy-labeled data, and a much smaller set of manually-labeled data, under a large vocabulary setting of 80 everyday sound classes. In addition, the proposed dataset poses an acoustic mismatch problem between the noisy train set and the test set due to the fact that they come from different web audio sources. This can correspond to a realistic scenario given by the difficulty in gathering large amounts of manually labeled data. We present the task setup, the FSDKaggle2019 dataset prepared for this scientific evaluation, and a baseline system consisting of a convolutional neural network. All these resources are freely available.

Xavier Favory, Xavier Serra

Recent advancements in web-based audio systems have enabled sufficiently accurate timing control and real-time sound processing capabilities. Numerous specialized music tools, as well as digital audio workstations, are now accessible from browsers. Features such as the large accessibility of data and real-time communication between clients make the web attractive for collaborative data manipulation. However, this innovative field has yet to produce effective tools for multiple-user coordination on specialized music creation tasks. The Multi Web Audio Sequencer is a prototype of an application for segment-based sequencing of Freesound sound clips, with an emphasis on seamless remote collaboration. In this work we consider a fixed-grid step sequencer as a probe for understanding the necessary features of crowd-shared music creation sessions. This manuscript describes the sequencer and the functionalities and types of interactions required for effective and attractive collaboration of remote people during creative music creation activities.

Andrés Ferraro, Dmitry Bogdanov, Xavier Serra

The Spotify Sequential Skip Prediction Challenge focuses on predicting if a track in a session will be skipped by the user or not. In this paper, we describe our approach to this problem and the final system that was submitted to the challenge by our team from the Music Technology Group (MTG) under the name "aferraro". This system consists in combining the predictions of multiple boosting trees models trained with features extracted from the sessions and the tracks. The proposed approach achieves good overall performance (MAA of 0.554), with our model ranked 14th out of more than 600 submissions in the final leaderboard.

Andres Ferraro, Dmitry Bogdanov, Kyumin Choi et al.

There are many offline metrics that can be used as a reference for evaluation and optimization of the performance of recommender systems. Hybrid recommendation approaches are commonly used to improve some of those metrics by combining different systems. In this work we focus on music recommendation and propose a new way to improve recommendations, with respect to a desired metric of choice, by combining multiple systems for each user individually based on their expected performance. Essentially, our approach consists in predicting an expected error that each system will produce for each user based on their previous activity. To this end, we propose to train regression models for different metrics predicting the performance of each system based on a number of features characterizing previous user behavior in the system. We then use different fusion strategies to combine recommendations generated by each system. Following this approach one can optimize the final hybrid system with respect to the desired metric of choice. As a proof of concept, we conduct experiments combining two recommendation systems, a Matrix Factorization model and a popularity-based recommender. We use the data provided by Melon, a Korean music streaming service, to train and evaluate the performance of the systems.

Eduardo Fonseca, Manoj Plakal, Daniel P. W. Ellis et al.

As sound event classification moves towards larger datasets, issues of label noise become inevitable. Web sites can supply large volumes of user-contributed audio and metadata, but inferring labels from this metadata introduces errors due to unreliable inputs, and limitations in the mapping. There is, however, little research into the impact of these errors. To foster the investigation of label noise in sound event classification we present FSDnoisy18k, a dataset containing 42.5 hours of audio across 20 sound classes, including a small amount of manually-labeled data and a larger quantity of real-world noisy data. We characterize the label noise empirically, and provide a CNN baseline system. Experiments suggest that training with large amounts of noisy data can outperform training with smaller amounts of carefully-labeled data. We also show that noise-robust loss functions can be effective in improving performance in presence of corrupted labels.