Paper Archive

This paper presents an examination of State Space Models (SSM) and Koopman-based deep learning methods for modelling the dynamics of both linear and non-linear stiff strings. Through experiments with datasets generated under different initial conditions and sample rates, we assess the capacity of these models to accurately model the complex behaviours observed in string dynamics. Our findings indicate that our proposed Koopman-based model performs as well as or better than other existing approaches in nonlinear cases for long-sequence modelling. We inform the design of these architectures with the structure of the problems at hand. Although challenges remain in extending model predictions beyond the training horizon (i.e., extrapolation), the focus of our investigation lies in the models’ ability to generalise across different initial conditions within the training time interval. This research contributes insights into the physical modelling of dynamical systems (in particular those addressing musical acoustics) by offering a comparative overview of these and previous methods and introducing innovative strategies for model improvement. Our results highlight the efficacy of these models in simulating non-linear dynamics and emphasise their wide-ranging applicability in accurately modelling dynamical systems over extended sequences.

Download

Automatic blind upmixing aims to convert audio from a smaller channel format (e.g. mono or stereo) into a multichannel format using estimates of direct and diffuse spatial statistics within the signal. Current approaches rely on primary-ambient extraction (PAE) algorithms, which lack real-world context through limited processing windows. Deep learning music source separation (MSS) models have been applied in voice-primary-ambient extraction (VPA) upmixing systems for handling direct components, but still rely on DSP methods of surround channel generation. This work further investigates utilizing source separation within VPA upmixing, focusing specifically on the task of stereo decorrelation and ambience extraction for 5.1 surround. We also release PAEDB (Primary–Ambient Extraction Dataset), a high-quality music dataset derived from MUSDB18-HQ and MoisesDB, comprising 1,809 primary–ambient stem pairs totaling over 550 hours of audio. The performance of selected DNNs trained on PAEDB is then evaluated using signal metrics and a listening study. Our findings indicate that DNNs can effectively model the behavior of PAE algorithms, establishing PAEDB as a strong foundation for ML upmixing systems and underscoring the need for higher-quality multichannel data to advance beyond conventional methods.

Download

Deep State Space Models (SSMs) have shown remarkable performance in long-sequence reasoning tasks, such as raw audio classification, and audio generation. This paper introduces PianoSSM, an end-to-end deep SSM neural network architecture designed to synthesize raw piano audio directly from MIDI input. The network requires no intermediate representations or domainspecific expert knowledge, simplifying training and improving accessibility. Quantitative evaluations on the MAESTRO dataset show that Piano-SSM achieves a Multi-Scale Spectral Loss (MSSL) of 7.02 at 16kHz, outperforming DDSP-Piano v1 with a MSSL of 7.09. At 24kHz, Piano-SSM maintains competitive performance with an MSSL of 6.75, closely matching DDSP-Piano v2’s result of 6.58. Evaluations on the MAPS dataset achieve an MSSL score of 8.23, which demonstrates the generalization capability even when training with very limited data. Further analysis highlights Piano-SSM’s ability to train on high sampling-rate audio while synthesizing audio at lower sampling rates, explicitly linking performance loss to aliasing effects. Additionally, the proposed model facilitates real-time causal inference through a custom C++17 header-only implementation. Using an Intel Core i712700 processor at 4.5GHz, with single core inference, allows synthesizing one second of audio at 44.1kHz in 0.44s with a workload of 23.1GFLOPS/s and an 10.1µs input/output delay with the largest network. While the smallest network at 16kHz only needs 0.04s with 2.3GFLOP/s and 2.6µs input/output delay. These results underscore Piano-SSM’s practical utility and efficiency in real-time audio synthesis applications.

Download

Recently, i-vector features have entered the field of Music Information Retrieval (MIR), exhibiting highly promising performance in important tasks such as music artist recognition or music similarity estimation. The i-vector modelling approach relies on a complex processing chain that limits by the use of engineered features such as MFCCs. The goal of the present paper is to make an important step towards a truly end-to-end modelling system inspired by the i-vector pipeline, to exploit the power of Deep Neural Networks1 (DNNs) to learn optimized feature spaces and transformations. Several authors have already tried to combine the power of DNNs with i-vector features, where DNNs were used for feature extraction, scoring or classification. In this paper, we try to use neural networks for the important step of i-vector post-processing and classification for the task of music artist recognition. Specifically, we propose a novel neural network for i-vector features with a cosine-distance loss function, optimized with stochastic gradient decent (SGD). We first show that current networks do not perform well with unprocessed i-vector features, and that post-processing methods such as Within-Class Covariance Normalization (WCCN) and Linear Discriminant Analysis (LDA) are crucially important to improve the i-vector representation. We further demonstrate that these linear projections (WCCN and LDA) can not be learned using general objective functions usually used in neural networks. We examine our network on a 50-class music artist recognition dataset using i-vectors extracted from frame-level timbre features. Our experiments suggest that using our network with fully unprocessed i-vectors, we can achieve the performance of the i-vector pipeline which uses i-vector post processing methods such as LDA and WCCN.

Download

Real-time applications of Music Information Retrieval (MIR) have been gaining interest as of recently. However, as deep learning becomes more and more ubiquitous for music analysis tasks, several challenges and limitations need to be overcome to deliver accurate and quick real-time MIR systems. In addition, modern embedded computers offer great potential for compact systems that use MIR algorithms, such as digital musical instruments. However, embedded computing hardware is generally resource constrained, posing additional limitations. In this paper, we identify and discuss the challenges and limitations of embedded real-time MIR. Furthermore, we discuss potential solutions to these challenges, and demonstrate their validity by presenting an embedded real-time classifier of expressive acoustic guitar techniques. The classifier achieved 99.2% accuracy in distinguishing pitched and percussive techniques and a 99.1% average accuracy in distinguishing four distinct percussive techniques with a fifth class for pitched sounds. The full classification task is a considerably more complex learning problem, with our preliminary results reaching only 56.5% accuracy. The results were produced with an average latency of 30.7 ms.

Download

This work aims to implement a novel deep learning architecture to perform audio processing in the context of matched equalization. Most existing methods for automatic and matched equalization show effective performance and their goal is to find a respective transfer function given a frequency response. Nevertheless, these procedures require a prior knowledge of the type of filters to be modeled. In addition, fixed filter bank architectures are required in automatic mixing contexts. Based on end-to-end convolutional neural networks, we introduce a general purpose architecture for equalization matching. Thus, by using an end-toend learning approach, the model approximates the equalization target as a content-based transformation without directly finding the transfer function. The network learns how to process the audio directly in order to match the equalized target audio. We train the network through unsupervised and supervised learning procedures. We analyze what the model is actually learning and how the given task is accomplished. We show the model performing matched equalization for shelving, peaking, lowpass and highpass IIR and FIR equalizers.

Download

This paper investigates the synthesis of audio textures and the structure of generative latent spaces using Variational Autoencoders (VAEs) within two paradigms of neural audio synthesis: DSP-inspired and data-driven approaches. For each paradigm, we propose VAE-based frameworks that allow fine-grained temporal control. We introduce datasets across three categories of environmental sounds to support our investigations. We evaluate and compare the models’ reconstruction performance using objective metrics, and investigate their generative capabilities and latent space structure through latent space interpolations.

Download

The success recently encountered by physically-based modeling (or model-based approaches) for music should not mask the deep challenges that remain in this area. This article first proposes an overview of the various goals that researchers and musicians, respectively operating from scientific and end-user perspectives, may pursue. Among these goals, those recently proposed or particularly critical for the coming years of research are highlighted. The article then introduces ten criteria that summarize the main features an optimal physically-based modeling scheme or language should present. With respect to these, it proposes an evaluation of the major approaches to physically-based modeling. Key words: goals of the physically-based approach to sound synthesis and music creation, languages and schemes, enduser needs, perception, evaluation criteria, bibliographic overview.

Download

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.

Download

In this paper, we propose an approach to real-time singing voice conversion and outline its development as a plug-in suitable for streaming use in a digital audio workstation. In order to simultaneously ensure pitch preservation and reduce the computational complexity of the overall system, we adopt a source-filter methodology and consider a vocoder-free paradigm for modeling the conversion task. In this case, the source is extracted and altered using more traditional DSP techniques, while the filter is determined using a deep neural network. The latter can be trained in an end-toend fashion and additionally uses adversarial training to improve system fidelity. Careful design allows the system to scale naturally to sampling rates higher than the neural filter model sampling rate, outputting full-band signals while avoiding the need for resampling. Accordingly, the resulting system, when operating at 44.1 kHz, incurs under 60 ms of latency and operates 20 times faster than real-time on a standard laptop CPU.

Download