Paper Archive

Controlling the variations of sound effects using neural audio synthesis models has been a challenging task. Differentiable digital signal processing (DDSP) provides a lightweight solution that achieves high-quality sound synthesis while enabling deterministic acoustic attribute control by incorporating pre-processed audio features and digital synthesizers. In this research, we introduce DDSP-SFX, a model based on the DDSP architecture capable of synthesizing high-quality sound effects while enabling users to control the timbre variations easily. We integrate a transient modelling algorithm in DDSP that achieves higher objective evaluation scores and subjective ratings over impulsive signals (footsteps, gunshots). We propose a novel method that achieves frame-level timbre variation control while also allowing deterministic attribute control. We further qualitatively show the timbre transfer performance using voice as the guiding sound.

Download

Music mixing is compositional — experts combine multiple audio processors to achieve a cohesive mix from dry source tracks. We propose a method to reverse engineer this process from the input and output audio. First, we create a mixing console that applies all available processors to every chain. Then, after the initial console parameter optimization, we alternate between removing redundant processors and fine-tuning. We achieve this through differentiable implementation of both processors and pruning. Consequently, we find a sparse mixing graph that achieves nearly identical matching quality of the full mixing console. We apply this procedure to drymix pairs from various datasets and collect graphs that also can be used to train neural networks for music mixing applications.

Download

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-toend 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 reexpressing 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 feed-forward compressor. We make our code and audio samples available and provide the trained audio effect and synth models in a VST plugin1 .

Download

Controllable timbre synthesis has been a subject of research for several decades, and deep neural networks have been the most successful in this area. Deep generative models such as Variational Autoencoders (VAEs) have the ability to generate a high-level representation of audio while providing a structured latent space. Despite their advantages, the interpretability of these latent spaces in terms of human perception is often limited. To address this limitation and enhance the control over timbre generation, we propose a regularized VAE-based latent space that incorporates timbre descriptors. Moreover, we suggest a more concise representation of sound by utilizing its harmonic content, in order to minimize the dimensionality of the latent space.

Download

Modal methods for simulating vibrations of strings, membranes, and plates are widely used in acoustics and physically informed audio synthesis. However, traditional implementations, particularly for non-linear models like the von Kármán plate, are computationally demanding and lack differentiability, limiting inverse modelling and real-time applications. We introduce a fast, differentiable, GPU-accelerated modal framework built with the JAX library, providing efficient simulations and enabling gradientbased inverse modelling. Benchmarks show that our approach significantly outperforms CPU and GPU-based implementations, particularly for simulations with many modes. Inverse modelling experiments demonstrate that our approach can recover physical parameters, including tension, stiffness, and geometry, from both synthetic and experimental data. Although fitting physical parameters is more sensitive to initialisation compared to methods that fit abstract spectral parameters, it provides greater interpretability and more compact parameterisation. The code is released as open source to support future research and applications in differentiable physical modelling and sound synthesis.

Download

We present a non-supervised approach to optimize and evaluate the synthesis of non-speech audio effects from a speech production model. We use the Pink Trombone synthesizer as a case study of a simplified production model of the vocal tract to target nonspeech human audio signals –yawnings. We selected and optimized the control parameters of the synthesizer to minimize the difference between real and generated audio. We validated the most common optimization techniques reported in the literature and a specifically designed neural network. We evaluated several popular quality metrics as error functions. These include both objective quality metrics and subjective-equivalent metrics. We compared the results in terms of total error and computational demand. Results show that genetic and swarm optimizers outperform least squares algorithms at the cost of executing slower and that specific combinations of optimizers and audio representations offer significantly different results. The proposed methodology could be used in benchmarking other physical models and audio types.

Download

This paper presents a novel approach to neural instrument sound synthesis using a two-stage semi-supervised learning framework capable of generating pitch-accurate, high-quality music samples from an expressive timbre latent space. Existing approaches that achieve sufficient quality for music production often rely on highdimensional latent representations that are difficult to navigate and provide unintuitive user experiences. We address this limitation through a two-stage training paradigm: first, we train a pitchtimbre disentangled 2D representation of audio samples using a Variational Autoencoder; second, we use this representation as conditioning input for a Transformer-based generative model. The learned 2D latent space serves as an intuitive interface for navigating and exploring the sound landscape. We demonstrate that the proposed method effectively learns a disentangled timbre space, enabling expressive and controllable audio generation with reliable pitch conditioning. Experimental results show the model’s ability to capture subtle variations in timbre while maintaining a high degree of pitch accuracy. The usability of our method is demonstrated in an interactive web application, highlighting its potential as a step towards future music production environments that are both intuitive and creatively empowering: https://pgesam.faresschulz.com/.

Download

Audio effects and sound synthesizers are widely used processors in popular music. Their parameters control the quality of the output sound. Multiple combinations of parameters can lead to the same sound. While recent approaches have been proposed to estimate these parameters given only the output sound, those are deterministic, i.e. they only estimate a single solution among the many possible parameter configurations. In this work, we propose to model the parameters as probability distributions instead of deterministic values. To learn the distributions, we optimize two objectives: (1) we minimize the reconstruction error between the ground truth output sound and the one generated using the estimated parameters, asisit usuallydone, but also(2)we maximize the parameter diversity, using entropy. We evaluate our approach through two numerical audio experiments to show its effectiveness. These results show how our approach effectively outputs multiple combinations of parameters to match one sound.

Download

Given a target sound, finding the synthesizer preset that best reproduces it remains a core problem in sound design. Existing methods treat synthesis parameters as flat vectors, discarding the signal routing and parameter interactions that produce audio. We make two contributions. First, to learn a representation of parameters including their signal routing, we design a graph neural network whose message passing structure imitates FM signal processing. Second, we adapt the multimodal objective from SLAP to learn joint embeddings of audio and FM synthesizer parameters, enabling preset retrieval from a gallery. We focus on the Yamaha DX7, where six identical sinusoid operators interact according to one of 32 routing topologies. Our graph encoder's message passing weights are shared across all nodes and layers, enabling processing of arbitrary topologies of any size. When every topology is seen during training, the DX7-GNN and two baselines achieve strong audio-to-preset retrieval. When some topologies are held out for testing, the DX7-GNN substantially outperforms both baselines despite having the fewest parameters. Our ablations further support the claim that imitating FM signal flow in a parameter encoder improves generalization to unseen topologies.

Download

Current datasets for automatic drum transcription (ADT) are small and limited due to the tedious task of annotating onset events. While some of these datasets contain large vocabularies of percussive instrument classes (e.g. ~20 classes), many of these classes occur very infrequently in the data. This paucity of data makes it difficult to train models that support such large vocabularies. Therefore, data-driven drum transcription models often focus on a small number of percussive instrument classes (e.g. 3 classes). In this paper, we propose to support large-vocabulary drum transcription by generating a large synthetic dataset (210,000 eight second examples) of audio examples for which we have groundtruth transcriptions. Using this synthetic dataset along with existing drum transcription datasets, we train convolutional-recurrent neural networks (CRNNs) in a multi-task framework to support large-vocabulary ADT. We find that training on both the synthetic and real music drum transcription datasets together improves performance on not only large-vocabulary ADT, but also beat / downbeat detection small-vocabulary ADT.

Download