György Fazekas

Ilaria Manco, Benno Weck, SeungHeon Doh et al. · bytedance

We introduce the Song Describer dataset (SDD), a new crowdsourced corpus of high-quality audio-caption pairs, designed for the evaluation of music-and-language models. The dataset consists of 1.1k human-written natural language descriptions of 706 music recordings, all publicly accessible and released under Creative Common licenses. To showcase the use of our dataset, we benchmark popular models on three key music-and-language tasks (music captioning, text-to-music generation and music-language retrieval). Our experiments highlight the importance of cross-dataset evaluation and offer insights into how researchers can use SDD to gain a broader understanding of model performance.

Jincheng Zhang, György Fazekas, Charalampos Saitis

Emerging Denoising Diffusion Probabilistic Models (DDPM) have become increasingly utilised because of promising results they have achieved in diverse generative tasks with continuous data, such as image and sound synthesis. Nonetheless, the success of diffusion models has not been fully extended to discrete symbolic music. We propose to combine a vector quantized variational autoencoder (VQ-VAE) and discrete diffusion models for the generation of symbolic music with desired composer styles. The trained VQ-VAE can represent symbolic music as a sequence of indexes that correspond to specific entries in a learned codebook. Subsequently, a discrete diffusion model is used to model the VQ-VAE's discrete latent space. The diffusion model is trained to generate intermediate music sequences consisting of codebook indexes, which are then decoded to symbolic music using the VQ-VAE's decoder. The evaluation results demonstrate our model can generate symbolic music with target composer styles that meet the given conditions with a high accuracy of 72.36%. Our code is available at https://github.com/jinchengzhanggg/VQVAE-Diffusion.

Ilaria Manco, Emmanouil Benetos, Elio Quinton et al.

As one of the most intuitive interfaces known to humans, natural language has the potential to mediate many tasks that involve human-computer interaction, especially in application-focused fields like Music Information Retrieval. In this work, we explore cross-modal learning in an attempt to bridge audio and language in the music domain. To this end, we propose MusCALL, a framework for Music Contrastive Audio-Language Learning. Our approach consists of a dual-encoder architecture that learns the alignment between pairs of music audio and descriptive sentences, producing multimodal embeddings that can be used for text-to-audio and audio-to-text retrieval out-of-the-box. Thanks to this property, MusCALL can be transferred to virtually any task that can be cast as text-based retrieval. Our experiments show that our method performs significantly better than the baselines at retrieving audio that matches a textual description and, conversely, text that matches an audio query. We also demonstrate that the multimodal alignment capability of our model can be successfully extended to the zero-shot transfer scenario for genre classification and auto-tagging on two public datasets.

Yinghao Ma, Anders Øland, Anton Ragni et al.

In recent years, foundation models (FMs) such as large language models (LLMs) and latent diffusion models (LDMs) have profoundly impacted diverse sectors, including music. This comprehensive review examines state-of-the-art (SOTA) pre-trained models and foundation models in music, spanning from representation learning, generative learning and multimodal learning. We first contextualise the significance of music in various industries and trace the evolution of AI in music. By delineating the modalities targeted by foundation models, we discover many of the music representations are underexplored in FM development. Then, emphasis is placed on the lack of versatility of previous methods on diverse music applications, along with the potential of FMs in music understanding, generation and medical application. By comprehensively exploring the details of the model pre-training paradigm, architectural choices, tokenisation, finetuning methodologies and controllability, we emphasise the important topics that should have been well explored, like instruction tuning and in-context learning, scaling law and emergent ability, as well as long-sequence modelling etc. A dedicated section presents insights into music agents, accompanied by a thorough analysis of datasets and evaluations essential for pre-training and downstream tasks. Finally, by underscoring the vital importance of ethical considerations, we advocate that following research on FM for music should focus more on such issues as interpretability, transparency, human responsibility, and copyright issues. The paper offers insights into future challenges and trends on FMs for music, aiming to shape the trajectory of human-AI collaboration in the music realm.

Rodrigo Diaz, Ben Hayes, Charalampos Saitis et al.

Physical models of rigid bodies are used for sound synthesis in applications from virtual environments to music production. Traditional methods such as modal synthesis often rely on computationally expensive numerical solvers, while recent deep learning approaches are limited by post-processing of their results. In this work we present a novel end-to-end framework for training a deep neural network to generate modal resonators for a given 2D shape and material, using a bank of differentiable IIR filters. We demonstrate our method on a dataset of synthetic objects, but train our model using an audio-domain objective, paving the way for physically-informed synthesisers to be learned directly from recordings of real-world objects.

Cyrus Vahidi, Han Han, Changhong Wang et al.

Computer musicians refer to mesostructures as the intermediate levels of articulation between the microstructure of waveshapes and the macrostructure of musical forms. Examples of mesostructures include melody, arpeggios, syncopation, polyphonic grouping, and textural contrast. Despite their central role in musical expression, they have received limited attention in deep learning. Currently, autoencoders and neural audio synthesizers are only trained and evaluated at the scale of microstructure: i.e., local amplitude variations up to 100 milliseconds or so. In this paper, we formulate and address the problem of mesostructural audio modeling via a composition of a differentiable arpeggiator and time-frequency scattering. We empirically demonstrate that time--frequency scattering serves as a differentiable model of similarity between synthesis parameters that govern mesostructure. By exposing the sensitivity of short-time spectral distances to time alignment, we motivate the need for a time-invariant and multiscale differentiable time--frequency model of similarity at the level of both local spectra and spectrotemporal modulations.

Ben Hayes, Charalampos Saitis, György Fazekas

We discuss the discontinuities that arise when mapping unordered objects to neural network outputs of fixed permutation, referred to as the responsibility problem. Prior work has proved the existence of the issue by identifying a single discontinuity. Here, we show that discontinuities under such models are uncountably infinite, motivating further research into neural networks for unordered data.

Soumya Sai Vanka, Maryam Safi, Jean-Baptiste Rolland et al.

Effective music mixing requires technical and creative finesse, but clear communication with the client is crucial. The mixing engineer must grasp the client's expectations, and preferences, and collaborate to achieve the desired sound. The tacit agreement for the desired sound of the mix is often established using guides like reference songs and demo mixes exchanged between the artist and the engineer and sometimes verbalised using semantic terms. This paper presents the findings of a two-phased exploratory study aimed at understanding how professional mixing engineers interact with clients and use their feedback to guide the mixing process. For phase one, semi-structured interviews were conducted with five mixing engineers with the aim of gathering insights about their communication strategies, creative processes, and decision-making criteria. Based on the inferences from these interviews, an online questionnaire was designed and administered to a larger group of 22 mixing engineers during the second phase. The results of this study shed light on the importance of collaboration, empathy, and intention in the mixing process, and can inform the development of smart multi-track mixing systems that better support these practices. By highlighting the significance of these findings, this paper contributes to the growing body of research on the collaborative nature of music production and provides actionable recommendations for the design and implementation of innovative mixing tools.

Jincheng Zhang, György Fazekas, Charalampos Saitis

Diffusion models have shown promising results for a wide range of generative tasks with continuous data, such as image and audio synthesis. However, little progress has been made on using diffusion models to generate discrete symbolic music because this new class of generative models are not well suited for discrete data while its iterative sampling process is computationally expensive. In this work, we propose a diffusion model combined with a Generative Adversarial Network, aiming to (i) alleviate one of the remaining challenges in algorithmic music generation which is the control of generation towards a target emotion, and (ii) mitigate the slow sampling drawback of diffusion models applied to symbolic music generation. We first used a trained Variational Autoencoder to obtain embeddings of a symbolic music dataset with emotion labels and then used those to train a diffusion model. Our results demonstrate the successful control of our diffusion model to generate symbolic music with a desired emotion. Our model achieves several orders of magnitude improvement in computational cost, requiring merely four time steps to denoise while the steps required by current state-of-the-art diffusion models for symbolic music generation is in the order of thousands.

Chin-Yun Yu, Christopher Mitcheltree, Alistair Carson et al.

Infinite impulse response filters are an essential building block of many time-varying audio systems, such as audio effects and synthesisers. However, their recursive structure impedes end-to-end training of these systems using automatic differentiation. Although non-recursive filter approximations like frequency sampling and frame-based processing have been proposed and widely used in previous works, they cannot accurately reflect the gradient of the original system. We alleviate this difficulty by re-expressing a time-varying all-pole filter to backpropagate the gradients through itself, so the filter implementation is not bound to the technical limitations of automatic differentiation frameworks. This implementation can be employed within audio systems containing filters with poles for efficient gradient evaluation. We demonstrate its training efficiency and expressive capabilities for modelling real-world dynamic audio systems on a phaser, time-varying subtractive synthesiser, and compressor. We make our code and audio samples available and provide the trained audio effect and synth models in a VST plugin at https://diffapf.github.io/web/.

Ben Hayes, Charalampos Saitis, György Fazekas

Many audio synthesizers can produce the same signal given different parameter configurations, meaning the inversion from sound to parameters is an inherently ill-posed problem. We show that this is largely due to intrinsic symmetries of the synthesizer, and focus in particular on permutation invariance. First, we demonstrate on a synthetic task that regressing point estimates under permutation symmetry degrades performance, even when using a permutation-invariant loss function or symmetry-breaking heuristics. Then, viewing equivalent solutions as modes of a probability distribution, we show that a conditional generative model substantially improves performance. Further, acknowledging the invariance of the implicit parameter distribution, we find that performance is further improved by using a permutation equivariant continuous normalizing flow. To accommodate intricate symmetries in real synthesizers, we also propose a relaxed equivariance strategy that adaptively discovers relevant symmetries from data. Applying our method to Surge XT, a full-featured open source synthesizer used in real world audio production, we find our method outperforms regression and generative baselines across audio reconstruction metrics.

Jincheng Zhang, György Fazekas, Charalampos Saitis

The recent surge in the popularity of diffusion models for image synthesis has attracted new attention to their potential for generation tasks in other domains. However, their applications to symbolic music generation remain largely under-explored because symbolic music is typically represented as sequences of discrete events and standard diffusion models are not well-suited for discrete data. We represent symbolic music as image-like pianorolls, facilitating the use of diffusion models for the generation of symbolic music. Moreover, this study introduces a novel diffusion model that incorporates our proposed Transformer-Mamba block and learnable wavelet transform. Classifier-free guidance is utilised to generate symbolic music with target chords. Our evaluation shows that our method achieves compelling results in terms of music quality and controllability, outperforming the strong baseline in pianoroll generation. Our code is available at https://github.com/jinchengzhanggg/proffusion.

Elizabeth Wilson, György Fazekas, Geraint Wiggins

This paper presents Tidal-MerzA, a novel system designed for collaborative performances between humans and a machine agent in the context of live coding, specifically focusing on the generation of musical patterns. Tidal-MerzA fuses two foundational models: ALCAA (Affective Live Coding Autonomous Agent) and Tidal Fuzz, a computational framework. By integrating affective modelling with computational generation, this system leverages reinforcement learning techniques to dynamically adapt music composition parameters within the TidalCycles framework, ensuring both affective qualities to the patterns and syntactical correctness. The development of Tidal-MerzA introduces two distinct agents: one focusing on the generation of mini-notation strings for musical expression, and another on the alignment of music with targeted affective states through reinforcement learning. This approach enhances the adaptability and creative potential of live coding practices and allows exploration of human-machine creative interactions. Tidal-MerzA advances the field of computational music generation, presenting a novel methodology for incorporating artificial intelligence into artistic practices.

Eleanor Row, György Fazekas

We present a study that explores the role of user-centred design in developing Generative AI (GenAI) tools for music composition. Through semi-structured interviews with professional composers, we gathered insights on a novel generative model for creating variations, highlighting concerns around trust, transparency, and ethical design. The findings helped form a feedback loop, guiding improvements to the model that emphasised traceability, transparency and explainability. They also revealed new areas for innovation, including novel features for controllability and research questions on the ethical and practical implementation of GenAI models.

Eleanor Row, Ivan Shanin, György Fazekas

This paper explores transformer-based models for music overpainting, focusing on jazz piano variations. Music overpainting generates new variations while preserving the melodic and harmonic structure of the input. Existing approaches are limited by small datasets, restricting scalability and diversity. We introduce VAR4000, a subset of a larger dataset for jazz piano performances, consisting of 4,352 training pairs. Using a semi-automatic pipeline, we evaluate two transformer configurations on VAR4000, comparing their performance with the smaller JAZZVAR dataset. Preliminary results show promising improvements in generalisation and performance with the larger dataset configuration, highlighting the potential of transformer models to scale effectively for music overpainting on larger and more diverse datasets.

Ilaria Manco, Emmanouil Benetos, Elio Quinton et al.

Audio representations for music information retrieval are typically learned via supervised learning in a task-specific fashion. Although effective at producing state-of-the-art results, this scheme lacks flexibility with respect to the range of applications a model can have and requires extensively annotated datasets. In this work, we pose the question of whether it may be possible to exploit weakly aligned text as the only supervisory signal to learn general-purpose music audio representations. To address this question, we design a multimodal architecture for music and language pre-training (MuLaP) optimised via a set of proxy tasks. Weak supervision is provided in the form of noisy natural language descriptions conveying the overall musical content of the track. After pre-training, we transfer the audio backbone of the model to a set of music audio classification and regression tasks. We demonstrate the usefulness of our approach by comparing the performance of audio representations produced by the same audio backbone with different training strategies and show that our pre-training method consistently achieves comparable or higher scores on all tasks and datasets considered. Our experiments also confirm that MuLaP effectively leverages audio-caption pairs to learn representations that are competitive with audio-only and cross-modal self-supervised methods in the literature.

Syed Rifat Mahmud Rafee, Gyorgy Fazekas, Geraint A. ~Wiggins

Music Performers have their own idiosyncratic way of interpreting a musical piece. A group of skilled performers playing the same piece of music would likely to inject their unique artistic styles in their performances. The variations of the tempo, timing, dynamics, articulation etc. from the actual notated music are what make the performers unique in their performances. This study presents a dataset consisting of four movements of Schubert's ``Sonata in B-flat major, D.960" performed by nine virtuoso pianists individually. We proposed and extracted a set of expressive features that are able to capture the characteristics of an individual performer's style. We then present a performer identification method based on the similarity of feature distribution, given a set of piano performances. The identification is done considering each feature individually as well as a fusion of the features. Results show that the proposed method achieved a precision of 0.903 using fusion features. Moreover, the onset time deviation feature shows promising result when considered individually.

Elona Shatri, György Fazekas

The main challenges of Optical Music Recognition (OMR) come from the nature of written music, its complexity and the difficulty of finding an appropriate data representation. This paper provides a first look at DoReMi, an OMR dataset that addresses these challenges, and a baseline object detection model to assess its utility. Researchers often approach OMR following a set of small stages, given that existing data often do not satisfy broader research. We examine the possibility of changing this tendency by presenting more metadata. Our approach complements existing research; hence DoReMi allows harmonisation with two existing datasets, DeepScores and MUSCIMA++. DoReMi was generated using a music notation software and includes over 6400 printed sheet music images with accompanying metadata useful in OMR research. Our dataset provides OMR metadata, MIDI, MEI, MusicXML and PNG files, each aiding a different stage of OMR. We obtain 64% mean average precision (mAP) in object detection using half of the data. Further work includes re-iterating through the creation process to satisfy custom OMR models. While we do not assume to have solved the main challenges in OMR, this dataset opens a new course of discussions that would ultimately aid that goal.

Sebastian Löbbers, Mathieu Barthet, György Fazekas

Sound synthesiser controls typically correspond to technical parameters of signal processing algorithms rather than intuitive sound descriptors that relate to human perception of sound. This makes it difficult to realise sound ideas in a straightforward way. Cross-modal mappings, for example between gestures and sound, have been suggested as a more intuitive control mechanism. A large body of research shows consistency in human associations between sounds and shapes. However, the use of drawings to drive sound synthesis has not been explored to its full extent. This paper presents an exploratory study that asked participants to sketch visual imagery of sounds with a monochromatic digital drawing interface, with the aim to identify different representational approaches and determine whether timbral sound characteristics can be communicated reliably through visual sketches. Results imply that the development of a synthesiser exploiting sound-shape associations is feasible, but a larger and more focused dataset is needed in followup studies.

Ben Hayes, Charalampos Saitis, György Fazekas

We present the Neural Waveshaping Unit (NEWT): a novel, lightweight, fully causal approach to neural audio synthesis which operates directly in the waveform domain, with an accompanying optimisation (FastNEWT) for efficient CPU inference. The NEWT uses time-distributed multilayer perceptrons with periodic activations to implicitly learn nonlinear transfer functions that encode the characteristics of a target timbre. Once trained, a NEWT can produce complex timbral evolutions by simple affine transformations of its input and output signals. We paired the NEWT with a differentiable noise synthesiser and reverb and found it capable of generating realistic musical instrument performances with only 260k total model parameters, conditioned on F0 and loudness features. We compared our method to state-of-the-art benchmarks with a multi-stimulus listening test and the Fréchet Audio Distance and found it performed competitively across the tested timbral domains. Our method significantly outperformed the benchmarks in terms of generation speed, and achieved real-time performance on a consumer CPU, both with and without FastNEWT, suggesting it is a viable basis for future creative sound design tools.

Cyrus Vahidi, Charalampos Saitis, György Fazekas

Convolutional Neural Networks have been extensively explored in the task of automatic music tagging. The problem can be approached by using either engineered time-frequency features or raw audio as input. Modulation filter bank representations that have been actively researched as a basis for timbre perception have the potential to facilitate the extraction of perceptually salient features. We explore end-to-end learned front-ends for audio representation learning, ModNet and SincModNet, that incorporate a temporal modulation processing block. The structure is effectively analogous to a modulation filter bank, where the FIR filter center frequencies are learned in a data-driven manner. The expectation is that a perceptually motivated filter bank can provide a useful representation for identifying music features. Our experimental results provide a fully visualisable and interpretable front-end temporal modulation decomposition of raw audio. We evaluate the performance of our model against the state-of-the-art of music tagging on the MagnaTagATune dataset. We analyse the impact on performance for particular tags when time-frequency bands are subsampled by the modulation filters at a progressively reduced rate. We demonstrate that modulation filtering provides promising results for music tagging and feature representation, without using extensive musical domain knowledge in the design of this front-end.

Ilaria Manco, Emmanouil Benetos, Elio Quinton et al.

Content-based music information retrieval has seen rapid progress with the adoption of deep learning. Current approaches to high-level music description typically make use of classification models, such as in auto-tagging or genre and mood classification. In this work, we propose to address music description via audio captioning, defined as the task of generating a natural language description of music audio content in a human-like manner. To this end, we present the first music audio captioning model, MusCaps, consisting of an encoder-decoder with temporal attention. Our method combines convolutional and recurrent neural network architectures to jointly process audio-text inputs through a multimodal encoder and leverages pre-training on audio data to obtain representations that effectively capture and summarise musical features in the input. Evaluation of the generated captions through automatic metrics shows that our method outperforms a baseline designed for non-music audio captioning. Through an ablation study, we unveil that this performance boost can be mainly attributed to pre-training of the audio encoder, while other design choices - modality fusion, decoding strategy and the use of attention - contribute only marginally. Our model represents a shift away from classification-based music description and combines tasks requiring both auditory and linguistic understanding to bridge the semantic gap in music information retrieval.

Elona Shatri, György Fazekas

Optical Music Recognition (OMR) is concerned with transcribing sheet music into a machine-readable format. The transcribed copy should allow musicians to compose, play and edit music by taking a picture of a music sheet. Complete transcription of sheet music would also enable more efficient archival. OMR facilitates examining sheet music statistically or searching for patterns of notations, thus helping use cases in digital musicology too. Recently, there has been a shift in OMR from using conventional computer vision techniques towards a deep learning approach. In this paper, we review relevant works in OMR, including fundamental methods and significant outcomes, and highlight different stages of the OMR pipeline. These stages often lack standard input and output representation and standardised evaluation. Therefore, comparing different approaches and evaluating the impact of different processing methods can become rather complex. This paper provides recommendations for future work, addressing some of the highlighted issues and represents a position in furthering this important field of research.

Johan Pauwels, György Fazekas, Mark B. Sandler

In recent years, Markov logic networks (MLNs) have been proposed as a potentially useful paradigm for music signal analysis. Because all hidden Markov models can be reformulated as MLNs, the latter can provide an all-encompassing framework that reuses and extends previous work in the field. However, just because it is theoretically possible to reformulate previous work as MLNs, does not mean that it is advantageous. In this paper, we analyse some proposed examples of MLNs for musical analysis and consider their practical disadvantages when compared to formulating the same musical dependence relationships as (dynamic) Bayesian networks. We argue that a number of practical hurdles such as the lack of support for sequences and for arbitrary continuous probability distributions make MLNs less than ideal for the proposed musical applications, both in terms of easy of formulation and computational requirements due to their required inference algorithms. These conclusions are not specific to music, but apply to other fields as well, especially when sequential data with continuous observations is involved. Finally, we show that the ideas underlying the proposed examples can be expressed perfectly well in the more commonly used framework of (dynamic) Bayesian networks.

Di Sheng, György Fazekas

In this paper, a siamese DNN model is proposed to learn the characteristics of the audio dynamic range compressor (DRC). This facilitates an intelligent control system that uses audio examples to configure the DRC, a widely used non-linear audio signal conditioning technique in the areas of music production, speech communication and broadcasting. Several alternative siamese DNN architectures are proposed to learn feature embeddings that can characterise subtle effects due to dynamic range compression. These models are compared with each other as well as handcrafted features proposed in previous work. The evaluation of the relations between the hyperparameters of DNN and DRC parameters are also provided. The best model is able to produce a universal feature embedding that is capable of predicting multiple DRC parameters simultaneously, which is a significant improvement from our previous research. The feature embedding shows better performance than handcrafted audio features when predicting DRC parameters for both mono-instrument audio loops and polyphonic music pieces.

Keunwoo Choi, György Fazekas, Kyunghyun Cho et al.

Following their success in Computer Vision and other areas, deep learning techniques have recently become widely adopted in Music Information Retrieval (MIR) research. However, the majority of works aim to adopt and assess methods that have been shown to be effective in other domains, while there is still a great need for more original research focusing on music primarily and utilising musical knowledge and insight. The goal of this paper is to boost the interest of beginners by providing a comprehensive tutorial and reducing the barriers to entry into deep learning for MIR. We lay out the basic principles and review prominent works in this hard to navigate the field. We then outline the network structures that have been successful in MIR problems and facilitate the selection of building blocks for the problems at hand. Finally, guidelines for new tasks and some advanced topics in deep learning are discussed to stimulate new research in this fascinating field.

Keunwoo Choi, György Fazekas, Kyunghyun Cho et al.

In this paper, we empirically investigate the effect of audio preprocessing on music tagging with deep neural networks. We perform comprehensive experiments involving audio preprocessing using different time-frequency representations, logarithmic magnitude compression, frequency weighting, and scaling. We show that many commonly used input preprocessing techniques are redundant except magnitude compression.

Keunwoo Choi, György Fazekas, Mark Sandler et al.

In this paper, we present a transfer learning approach for music classification and regression tasks. We propose to use a pre-trained convnet feature, a concatenated feature vector using the activations of feature maps of multiple layers in a trained convolutional network. We show how this convnet feature can serve as general-purpose music representation. In the experiments, a convnet is trained for music tagging and then transferred to other music-related classification and regression tasks. The convnet feature outperforms the baseline MFCC feature in all the considered tasks and several previous approaches that are aggregating MFCCs as well as low- and high-level music features.