Paper Archive

Recent neural-based synthesis models have achieved impressive results for musical instrument sound generation. In particular, the Differentiable Digital Signal Processing (DDSP) framework enables the usage of spectral modeling analysis and synthesis techniques in fully differentiable architectures. Yet currently, it has only been used for modeling monophonic instruments. Leveraging the interpretability and modularity of this framework, the present work introduces a polyphonic differentiable model for piano sound synthesis, conditioned on Musical Instrument Digital Interface (MIDI) inputs. The model architecture is motivated by high-level acoustic modeling knowledge of the instrument which, in tandem with the sound structure priors inherent to the DDSP components, makes for a lightweight, interpretable and realistic sounding piano model. The proposed model has been evaluated in a listening test, demonstrating improved sound quality compared to a benchmark neural-based piano model, with significantly less parameters and even with reduced training data. The same listening test indicates that physical-modeling-based models still achieve better quality, but the differentiability of our lightened approach encourages its usage in other musical tasks dealing with polyphonic audio and symbolic data.

Download

This paper introduces a novel data-driven strategy for synthesizing gramophone noise audio textures. A diffusion probabilistic model is applied to generate highly realistic quasiperiodic noises. The proposed model is designed to generate samples of length equal to one disk revolution, but a method to generate plausible periodic variations between revolutions is also proposed. A guided approach is also applied as a conditioning method, where an audio signal generated with manually-tuned signal processing is refined via reverse diffusion to improve realism. The method has been evaluated in a subjective listening test, in which the participants were often unable to recognize the synthesized signals from the real ones. The synthetic noises produced with the best proposed unconditional method are statistically indistinguishable from real noise recordings. This work shows the potential of diffusion models for highly realistic audio synthesis tasks.

Download

In this paper, we propose a new processor capable of directly changing the microdynamics of an audio signal primarily via a single dedicated user-facing parameter. The novelty of our processor is that it has built into it a measure of relative level, a short-term signal strength measurement which is robust to changes in signal macrodynamics. Consequent dynamic range processing is signal level-independent in its nature, and attempts to directly alter its observed relative level measurements. The inclusion of such a meter within our proposed processor also gives rise to a natural solution to the dynamics matching problem, where we attempt to transfer the microdynamic characteristics of one audio recording to another by means of estimating appropriate settings for the processor. We suggest a means of providing a reasonable initial guess for processor settings, followed by an efficient iterative algorithm to refine upon our estimates. Additionally, we implement the processor as a differentiable recurrent layer and show its effectiveness when wrapped around a gradient descent optimizer within a deep learning framework. Moreover, we illustrate that the proposed processor has more favorable gradient characteristics relative to a conventional dynamic range compressor. Throughout, we consider extensions of the processor, matching algorithm, and differentiable implementation for the multiband case.

Download

Recent advancements in deep learning have shown great potential for audio applications, improving the accuracy of previous solutions for tasks such as music transcription, beat detection, and real-time audio processing. In addition, the availability of increasingly powerful embedded computers has led many deep learning framework developers to devise software optimized to run pretrained models in resource-constrained contexts. As a result, the use of deep learning on embedded devices and audio plugins has become more widespread. However, confusion has been rising around deep learning inference engines, regarding which of these can run in real-time and which are less resource-hungry. In this paper, we present a comparison of four available deep learning inference engines for real-time audio classification on the CPU of an embedded single-board computer: TensorFlow Lite, TorchScript, ONNX Runtime, and RTNeural. Results show that all inference engines can execute neural network models in real-time with appropriate code practices, but execution time varies between engines and models. Most importantly, we found that most of the less-specialized engines offer great flexibility and can be used effectively for real-time audio classification, with slightly better results than a real-time-specific approach. In contrast, more specialized solutions can offer a lightweight and minimalist alternative where less flexibility is needed.

Download

Many electronic music (EM) genres are composed through the activation of short audio recordings of instruments designed for seamless repetition—or loops. In this work, loops of key structural groups such as bass, percussive or melodic elements are labelled by the role they occupy in a piece of music through the task of automatic instrumentation role classification (AIRC). Such labels assist EM producers in the identification of compatible loops in large unstructured audio databases. While human annotation is often laborious, automatic classification allows for fast and scalable generation of these labels. We experiment with several deeplearning architectures and propose a data augmentation method for improving multi-label representation to balance classes within the Freesound Loop Dataset. To improve the classification accuracy of the architectures, we also evaluate different pooling operations. Results indicate that in combination with the data augmentation and pooling strategies, the proposed system achieves state-of-theart performance for AIRC. Additionally, we demonstrate how our proposed AIRC method is useful for analysing the structure of EM compositions through loop activation transcription.

Download

The integration of additional side information to improve music source separation has been investigated numerous times, e.g., by adding features to the input or by adding learning targets in a multi-task learning scenario. These approaches, however, require additional annotations such as musical scores, instrument labels, etc. in training and possibly during inference. The available datasets for source separation do not usually provide these additional annotations. In this work, we explore transfer learning strategies to incorporate VGGish features with a state-of-the-art source separation model; VGGish features are known to be a very condensed representation of audio content and have been successfully used in many music information retrieval tasks. We introduce three approaches to incorporate the features, including two latent space regularization methods and one naive concatenation method. Our preliminary results show that our proposed approaches could improve some evaluation metrics for music source separation. In this work, we also include a discussion of our proposed approaches, such as the pros and cons of each approach, and the potential extension/improvement.

Download

Joint time–frequency scattering (JTFS) is a convolutional operator in the time–frequency domain which extracts spectrotemporal modulations at various rates and scales. It offers an idealized model of spectrotemporal receptive fields (STRF) in the primary auditory cortex, and thus may serve as a biological plausible surrogate for human perceptual judgments at the scale of isolated audio events. Yet, prior implementations of JTFS and STRF have remained outside of the standard toolkit of perceptual similarity measures and evaluation methods for audio generation. We trace this issue down to three limitations: differentiability, speed, and flexibility. In this paper, we present an implementation of time–frequency scattering in Python. Unlike prior implementations, ours accommodates NumPy, PyTorch, and TensorFlow as backends and is thus portable on both CPU and GPU. We demonstrate the usefulness of JTFS via three applications: unsupervised manifold learning of spectrotemporal modulations, supervised classification of musical instruments, and texture resynthesis of bioacoustic sounds.

Download

Deep learning models applied to raw audio are rapidly gaining relevance in modeling audio analog devices. This paper investigates the use of different deep architectures for modeling audio optical compression. The models use as input and produce as output raw audio samples at audio rate, and it works with noor small-input buffers allowing a theoretical real-time and lowlatency implementation. In this study, two compressor parameters, the ratio, and threshold have been included in the modeling process aiming to condition the inference of the trained network. Deep learning architectures are compared to model an all-tube optical mono compressor including feed-forward, recurrent, and encoder-decoder models. The results of this study show that feedforward and long short-term memory architectures present limitations in modeling the triggering phase of the compressor, performing well only on the sustained phase. On the other hand, encoderdecoder models outperform other architectures in replicating the overall compression process, but they overpredict the energy of high-frequency components.

Download

Neural music instrument cloning is an application of deep neural networks for imitating the timbre of a particular music instrument recording with a trained neural network. One can create such clones using an approach such as DDSP [1], which has been shown to achieve good synthesis quality for several instrument types [2]. However, this approach needs about ten minutes of audio data from the instrument of interest (target recording audio). In this work, we modify the DDSP architecture and apply transfer learning techniques used in speech voice cloning [3] to significantly reduce the amount of target recording audio required. We compare various cloning approaches and architectures across durations of target recording audio, ranging from four to 256 seconds. We demonstrate editing of loudness and pitch as well as timbre transfer from only 16 seconds of target recording audio. Our code is available online1 as well as many audio examples.2

Download

This paper explores the digital emulation of analog dynamic range compressors, proposing a grey-box model that uses a combination of traditional signal processing techniques and machine learning. The main idea is to use the structure of a traditional digital compressor in a machine learning framework, so it can be trained end-to-end to create a virtual analog model of a compressor from data. The complexity of the model can be adjusted, allowing a trade-off between the model accuracy and computational cost. The proposed model has interpretable components, so its behaviour can be controlled more readily after training in comparison to a black-box model. The result is a model that achieves similar accuracy to a black-box baseline, whilst requiring less than 10% of the number of operations per sample at runtime.

Download