Paper Archive

We address the issue of extracting automatically high-level musical descriptors out of their raw audio signal. This work focuses on the extraction of the perceived intensity of music titles, that evaluates how energic the music is perceived by listeners. We present here first the perceptive tests that we have conducted, in order to evaluate the relevance and the universality of the perceived intensity descriptor. Then we present several methods used to extract relevant features used to build automatic intensity extractors: usual Mpeg7 low level features, empirical method, and features automatically found using our Extractor Discovery System (EDS), and compare the final performances of their extractors.

Download

Estimations of panning attributes are an important feature to extract from a piece of recorded music, with downstream uses such as classification, quality assessment, and listening enhancement. While several algorithms exist in the literature, there is currently no comparison between them and no studies to suggest which one is most suitable for any particular task. This paper compares four algorithms for extracting amplitude panning features with respect to their suitability for unsupervised learning. It finds synchronicities between them and analyses their results on a small set of commercial music excerpts chosen for their distinct panning features. The ability of each algorithm to differentiate between the tracks is analysed. The results can be used in future work to either select the most appropriate panning feature algorithm or create a version customized for a particular task.

Download

In this paper we present a generic system for audio indexing (classification/ segmentation) and apply it to two usual problems: speech/ music segmentation and music genre recognition. We first present some requirements for the design of a generic system. The training part of it is based on a succession of four steps: feature extraction, feature selection, feature space transform and statistical modeling. We then propose several approaches for the indexing part depending of the local/ global characteristics of the indexes to be found. In particular we propose the use of segment-statistical models. The system is then applied to two usual problems. The first one is the speech/ music segmentation of a radio stream. The application is developed in a real industrial framework using real world categories and data. The performances obtained for the pure speech/ music classes problem are good. However when considering also the non-pure categories (mixed, bed) the performances of the system drop. The second problem is the music genre recognition. Since the indexes to be found are global, “segment-statistical models” are used leading to results close to the state of the art.

Download

Neural networks have found application within the Wave Digital Filters (WDFs) framework as data-driven input-output blocks for modeling single one-port or multi-port nonlinear devices in circuit systems. However, traditional neural networks lack predictable bounds for their output derivatives, essential to ensure convergence when simulating circuits with multiple nonlinear elements using fixed-point iterative methods, e.g., the Scattering Iterative Method (SIM). In this study, we address such issue by employing Lipschitz-bounded neural networks for regressing nonlinear WD scattering relations of one-port nonlinearities.

Download

In this paper we present beat histogram features for multiple level rhythm description and evaluate them in a musical genre classification task. Audio features pertaining to various musical content categories and their related novelty functions are extracted as a basis for the creation of beat histograms. The proposed features capture not only amplitude, but also tonal and general spectral changes in the signal, aiming to represent as much rhythmic information as possible. The most and least informative features are identified through feature selection methods and are then tested using Support Vector Machines on five genre datasets concerning classification accuracy against a baseline feature set. Results show that the presented features provide comparable classification accuracy with respect to other genre classification approaches using periodicity histograms and display a performance close to that of much more elaborate up-to-date approaches for rhythm description. The use of bar boundary annotations for the texture frames has provided an improvement for the dance-oriented Ballroom dataset. The comparably small number of descriptors and the possibility of evaluating the influence of specific signal components to the general rhythmic content encourage the further use of the method in rhythm description tasks.

Download

Recognition of musical instruments in multi-instrumental, polyphonic music is a difficult challenge which is yet far from being solved. Successful instrument recognition techniques in solos (monophonic or polyphonic recordings of single instruments) can help to deal with this task. We introduce an instrument recognition process in solo recordings of a set of instruments (bassoon, clarinet, flute, guitar, piano, cello and violin), which yields a high recognition rate. A large and very diverse solo database (108 different solos, all by different performers) is used in order to encompass the different sound possibilities of each instrument and evaluate the generalization abilities of the classification process. First we bring classification results using a very extensive collection of features (62 different feature types), and then use our GDE feature selection algorithm to select a smaller feature set with a relatively short computation time, which allows us to perform instrument recognition in solos in real-time, with only a slight decrease in recognition rate. We demonstrate that our real-time solo classifier can also be useful for instrument recognition in duet performances, and improved using simple “source reduction”.

Download

Automatic detection of musical patterns is an important task in the field of Music Information Retrieval due to its usage in multiple applications such as automatic music transcription, genre or instrument identification, music classification, and music recommendation. A significant sub-task in pattern detection is the realtime pattern detection in music due to its relevance in application domains such as the Internet of Musical Things. In this study, we present a method to identify the occurrence of known patterns in symbolic monophonic music streams in real-time. We introduce a matrix-based representation to denote musical notes using its pitch, pitch-bend, amplitude, and duration. We propose an algorithm based on an independent similarity index for each note attribute. We also introduce the Match Measure, which is a numerical value signifying the degree of the match between a pattern and a sequence of notes. We have tested the proposed algorithm against three datasets: a human recorded dataset, a synthetically designed dataset, and the JKUPDD dataset. Overall, a detection rate of 95% was achieved. The low computational load and minimal running time demonstrate the suitability of the method for real-world, real-time implementations on embedded systems.

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

Auditory displays for mobile service robots are developed. The design of digital-audio symbols, such as directional sounds and additional sounds for robot states, as well as the design of more complicated robot sound tracks are explained. Basic musical elements and robot movement sounds are combined. Two experimental studies, on the understandability of the directional sounds and on the auditory perception of intended robot trajectories in a simulated supermarket scenario, are described.

Download

Accurately estimating nonlinear audio effects without access to paired input-output signals remains a challenging problem. This work studies unsupervised probabilistic approaches for solving this task. We introduce a method, novel for this application, based on diffusion generative models for blind system identification, enabling the estimation of unknown nonlinear effects using blackand gray-box models. This study compares this method with a previously proposed adversarial approach, analyzing the performance of both methods under different parameterizations of the effect operator and varying lengths of available effected recordings. Through experiments on guitar distortion effects, we show that the diffusion-based approach provides more stable results and is less sensitive to data availability, while the adversarial approach is superior at estimating more pronounced distortion effects. Our findings contribute to the robust unsupervised blind estimation of audio effects, demonstrating the potential of diffusion models for system identification in music technology.

Download