Paper Archive

This paper reports on a new multi-label classification task for guitar effect recognition that is closer to the actual use case of guitar effect pedals. To generate the dataset, we used multiple clean guitar audio datasets and applied various combinations of 13 commonly used guitar effects. We compared four neural network structures: a simple Multi-Layer Perceptron as a baseline, ResNet models, a CRNN model, and a sample-level CNN model. The ResNet models achieved the best performance in terms of accuracy and robustness under various setups (with or without clean audio, seen or unseen dataset), with a micro F1 of 0.876 and Macro F1 of 0.906 in the hardest setup. An ablation study on the ResNet models further indicates the necessary model complexity for the task.

Download

In this paper we investigate the problem of singer identification on acapella recordings of isolated notes. Most of studies on singer identification describe the content of signals of singing voice with features related to the timbre (such as MFCC or LPC). These features aim to describe the behavior of frequencies at a given instant of time (local features). In this paper, we propose to describe sung tone with the temporal variations of the fundamental frequency (and its harmonics) of the note. The periodic and continuous variations of the frequency trajectories are analyzed on the whole note and the features obtained reflect expressive and intonative elements of singing such as vibrato, tremolo and portamento. The experiments, conducted on two distinct data-sets (lyric and pop-rock singers), prove that the new set of features capture a part of the singer identity. However, these features are less accurate than timbre-based features. We propose to increase the recognition rate of singer identification by combining information conveyed by local and global description of notes. The proposed method, that shows good results, can be adapted for classification problem involving a large number of classes, or to combine classifications with different levels of performance.

Download

Drums modelling is of special interest in musical source separation because of its widespread presence in western popular music. Current research has often focused on drums separation without specifically modelling the other sources present in the signal. This paper presents an extensive study of the use of regularizations and constraints to drive the factorization towards the separation between percussive and non-percussive music accompaniment. The proposed regularizations control the frequency smoothness of the basis components and the temporal sparseness of the gains. We also evaluated the use of temporal constraints on the gains to perform the separation, using both ground truth manual annotations (made publicly available) and automatically extracted transients. Objective evaluation of the results shows that, while optimal regularizations are highly dependent on the signal, drum event position contains enough information to achieve a high quality separation.

Download

This paper introduces the possibilities of optimizing neural network convolutional layers for modeling nonlinear audio systems and effects. Enhanced methods for real-time dilated convolutions are presented to achieve faster signal processing times than in previous work. Due to the improved implementation of convolutional layers, a significant decrease in computational requirements was observed and validated on different configurations of single layers with dilated convolutions and WaveNet-style feedforward neural network models. In most cases, equivalent signal processing times were achieved to those using recurrent neural networks with Long Short-Term Memory units and Gated Recurrent Units, which are considered state-of-the-art in the field of black-box virtual analog modeling.

Download

Deep neural networks have been recently applied to the task of automatic synthesizer programming, i.e., finding optimal values of sound synthesis parameters in order to reproduce a given input sound. This paper focuses on generative models, which can infer parameters as well as generate new sets of parameters or perform smooth morphing effects between sounds. We introduce new models to ensure scalability and to increase performance by using heterogeneous representations of parameters as numerical and categorical random variables. Moreover, a spectral variational autoencoder architecture with multi-channel input is proposed in order to improve inference of parameters related to the pitch and intensity of input sounds. Model performance was evaluated according to several criteria such as parameters estimation error and audio reconstruction accuracy. Training and evaluation were performed using a 30k presets dataset which is published with this paper. They demonstrate significant improvements in terms of parameter inference and audio accuracy and show that presented models can be used with subsets or full sets of synthesizer parameters.

Download

Audio effects (AFXs) are essential tools in music production, frequently applied in chains to shape timbre and dynamics. The order of AFXs in a chain plays a crucial role in determining the final sound, particularly when non-linear (e.g., distortion) or timevariant (e.g., chorus) processors are involved. Despite its importance, most AFX-related studies have primarily focused on estimating effect types and their parameters from a wet signal. To address this gap, we formulate AFX chain recognition as the task of jointly estimating AFX types and their order from a wet signal. We propose a neural-network-based method that embeds wet signals into a hyperbolic space and classifies their AFX chains. Hyperbolic space can represent tree-structured data more efficiently than Euclidean space due to its exponential expansion property. Since AFX chains can be represented as trees, with AFXs as nodes and edges encoding effect order, hyperbolic space is well-suited for modeling the exponentially growing and non-commutative nature of ordered AFX combinations, where changes in effect order can result in different final sounds. Experiments using guitar sounds demonstrate that, with an appropriate curvature, the proposed method outperforms its Euclidean counterpart. Further analysis based on AFX type and chain length highlights the effectiveness of the proposed method in capturing AFX order.

Download

Active acoustics (AA) refers to an electroacoustic system that actively modifies the acoustics of a room. For common use cases, the number of transducers—loudspeakers and microphones—involved in the system is large, resulting in a large number of system parameters. To optimally blend the response of the system into the natural acoustics of the room, the parameters require careful tuning, which is a time-consuming process performed by an expert. In this paper, we present a differentiable AA framework, which allows multi-objective optimization without impairing architecture flexibility. The system is implemented in PyTorch to be easily translated into a machine-learning pipeline, thus automating the tuning process. The objective of the pipeline is to optimize the digital signal processor (DSP) component to evenly distribute the energy in the feedback loop across frequencies. We investigate the effectiveness of DSPs composed of finite impulse response filters, which are unconstrained during the optimization. We study the effect of multiple filter orders, number of transducers, and loss functions on the performance. Different loss functions behave similarly for systems with few transducers and low-order filters. Increasing the number of transducers and the order of the filters improves results and accentuates the difference in the performance of the loss functions.

Download

We explore the use of neural synthesis for acoustic guitar from string-wise MIDI input. We propose four different systems and compare them with both objective metrics and subjective evaluation against natural audio and a sample-based baseline. We iteratively develop these four systems by making various considerations on the architecture and intermediate tasks, such as predicting pitch and loudness control features. We find that formulating the control feature prediction task as a classification task rather than a regression task yields better results. Furthermore, we find that our simplest proposed system, which directly predicts synthesis parameters from MIDI input performs the best out of the four proposed systems. Audio examples and code are available.

Download

There is an increasing number of consumers of broadcast audio who suffer from a degree of hearing impairment. One of the methods developed for tackling this issue consists of creating customizable object-based audio mixes where users can attenuate parts of the mix using a simple complexity parameter. The method relies on the mixing engineer classifying audio objects in the mix according to their narrative importance. This paper focuses on automating this process. Individual tracks are classified based on their music, speech, or sound effect content. Then the decisions for assigning narrative importance to each segment of a radio drama mix are modelled using mixture distributions. Finally, the learned decisions and resultant mixes are evaluated using the Short Term Objective Intelligibility, with reference to the narrative importance selections made by the original producer. This approach has applications for providing customizable mixes for legacy content, or automatically generated media content where the engineer is not able to intervene.

Download

This paper addresses the task of lyrics-to-audio alignment, which involves synchronizing textual lyrics with corresponding music audio. Most publicly available datasets for this task provide annotations only at the line or word level. This poses a challenge for training lyrics-to-audio models due to the lack of frame-wise phoneme labels. However, we find that phoneme labels can be partially derived from word-level annotations: for single-phoneme words, all frames corresponding to the word can be labeled with the same phoneme; for multi-phoneme words, phoneme labels can be assigned at the first and last frames of the word. To leverage this partial information, we construct a mask for those frames and propose a masked frame-wise cross-entropy (CE) loss that considers only frames with known phoneme labels. As a baseline model, we adopt an autoencoder trained with a Connectionist Temporal Classification (CTC) loss and a reconstruction loss. We then enhance the training process by incorporating the proposed framewise masked CE loss. Experimental results show that incorporating the frame-wise masked CE loss improves alignment performance. In comparison to other state-of-the art models, our model provides a comparable Mean Absolute Error (MAE) of 0.216 seconds and a top Median Absolute Error (MedAE) of 0.041 seconds on the testing Jamendo dataset.

Download