Download Towards an Invertible Rhythm Representation This paper investigates the development of a rhythm representation of music audio signals, that (i) is able to tackle rhythm related tasks and, (ii) is invertible, i.e. is suitable to reconstruct audio from it with the corresponding rhythm content being preserved. A conventional front-end processing schema is applied to the audio signal to extract time varying characteristics (accent features) of the signal. Next, a periodicity analysis method is proposed that is capable of reconstructing the accent features. Afterwards, a network consisting of Restricted Boltzmann Machines is applied to the periodicity function to learn a latent representation. This latent representation is finally used to tackle two distinct rhythm tasks, namely dance style classification and meter estimation. The results are promising for both input signal reconstruction and rhythm classification performance. Moreover, the proposed method is extended to generate random samples from the corresponding classes.
Download Low-delay vector-quantized subband ADPCM coding Several modern applications require audio encoders featuring low data rate and lowest delays. In terms of delay, Adaptive Differential Pulse Code Modulation (ADPCM) encoders are advantageous compared to block-based codecs due to their instantaneous output and therefore preferred in time-critical applications. If the the audio signal transport is done block-wise anyways, as in Audio over IP (AoIP) scenarios, additional advantages can be expected from block-wise coding. In this study, a generalized subband ADPCM concept using vector quantization with multiple realizations and configurations is shown. Additionally, a way of optimizing the codec parameters is derived. The results show that for the cost of small algorithmic delays the data rate of ADPCM can be significantly reduced while obtaining a similar or slightly increased perceptual quality. The largest algorithmic delay of about 1 ms at 44.1 kHz is still smaller than the ones of well-known low-delay codecs.
Download Sparse Decomposition of Audio Signals Using a Perceptual Measure of Distortion. Application to Lossy Audio Coding State-of the art audio codecs use time-frequency transforms derived from cosine bases, followed by a quantification stage. The quantization steps are set according to perceptual considerations. In the last decade, several studies applied adaptive sparse time-frequency transforms to audio coding, e.g. on unions of cosine bases using a Matching-Pursuit-derived algorithm [1]. This was shown to significantly improve the coding efficiency. We propose another approach based on a variational algorithm, i.e. the optimization of a cost function taking into account both a perceptual distortion measure derived form a hearing model and a sparsity constraint, which favors the coding efficiency. In this early version, we show that, using a coding scheme without perceptual control of quantization, our method outperforms a codec from the literature with the same quantization scheme [1]. In future work, a more sophisticated quantization scheme would probably allow our method to challenge standard codecs e.g. AAC. Index Terms– Audio coding, Sparse approximation, Iterative thresholding algorithm, Perceptual model.
Download Approaches for constant audio latency on Android This paper discusses issues related to audio latency for realtime processing Android OS applications. We first introduce the problem, determining the difference between the concepts of low latency and constant latency. It is a well-known issue that programs written for this platform cannot implement low-latency audio. However, in some cases, while low latency is desirable, it is not crucial. In some of these cases, achieving a constant delay between control events and sound output is the necessary condition. The paper briefly outlines the audio architecture in the Android platform to tease out the difficulties. Following this, we proposed some approaches to deal with two basic situations, one where the audio callback system provided by the system software is isochronous, and one where it is not.
Download GstPEAQ – an Open Source Implementation of the PEAQ Algorithm In 1998, the ITU published a recommendation for an algorithm for objective measurement of audio quality, aiming to predict the outcome of listening tests. Despite the age, today only one implementation of that algorithm meeting the conformance requirements exists. Additionally, two open source implementations of the basic version of the algorithm are available which, however, do not meet the conformance requirements. In this paper, yet another non-conforming open source implementation, GstPEAQ, is presented. However, it improves upon the previous ones by coming closer to conformance and being computationally more efficient. Furthermore, it implements not only the basic, but also the advanced version of the algorithm. As is also shown, despite the nonconformance, the results obtained computationally still closely resemble those of listening tests.
Download Harmonic Mixing Based on Roughness and Pitch Commonality The practice of harmonic mixing is a technique used by DJs for the beat-synchronous and harmonic alignment of two or more pieces of music. In this paper, we present a new harmonic mixing method based on psychoacoustic principles. Unlike existing commercial DJ-mixing software which determine compatible matches between songs via key estimation and harmonic relationships in the circle of fifths, our approach is built around the measurement of musical consonance at the signal level. Given two tracks, we first extract a set of partials using a sinusoidal model and average this information over sixteenth note temporal frames. Then within each frame, we measure the consonance between all combinations of dyads according to psychoacoustic models of roughness and pitch commonality. By scaling the partials of one track over ± 6 semitones (in 1/8th semitone steps), we can determine the optimal pitch-shift which maximises the consonance of the resulting mix. Results of a listening test show that the most consonant alignments generated by our method were preferred to those suggested by an existing commercial DJ-mixing system.
Download Extraction of Metrical Structure from Music Recordings Rhythm is a fundamental aspect of music and metrical structure is an important rhythm-related element. Several mid-level features encoding metrical structure information have been proposed in the literature, although the explicit extraction of this information is rarely considered. In this paper, we present a method to extract the full metrical structure from music recordings without the need for any prior knowledge. The algorithm is evaluated against expert annotations of metrical structure for the GTZAN dataset, each track being annotated multiple times. Inter-annotator agreement and the resulting upper bound on algorithm performance are evaluated. The proposed system reaches 93% of this upper limit and largely outperforms the baseline method.
Download A set of audio features for the morphological description of vocal imitations In our current project, vocal signal has to be used to drive sound synthesis. In order to study the mapping between voice and synthesis parameters, the inverse problem is first studied. A set of reference synthesizer sounds have been created and each sound has been imitated by a large number of people. Each reference synthesizer sound belongs to one of the six following morphological categories: “up”, “down”, “up/down”, “impulse”, “repetition”, “stable”. The goal of this paper is to study the automatic estimation of these morphological categories from the vocal imitations. We propose three approaches for this. A base-line system is first introduced. It uses standard audio descriptors as inputs for a continuous Hidden Markov Model (HMM) and provides an accuracy of 55.1%. To improve this, we propose a set of slope descriptors which, converted into symbols, are used as input for a discrete HMM. This system reaches 70.8% accuracy. The recognition performance has been further increased by developing specific compact audio descriptors that directly highlight the morphological aspects of sounds instead of relying on HMM. This system allows reaching the highest accuracy: 83.6%.
Download On studying auditory distance perception in concert halls with multichannel auralizations Virtual acoustics and auralizations have been previously used to study the perceptual properties of concert hall acoustics in a descriptive profiling framework. The results have indicated that the apparent auditory distance to the orchestra might play a crucial role in enhancing the listening experience and the appraisal of hall acoustics. However, it is unknown how the acoustics of the hall influence auditory distance perception in such large spaces. Here, we present one step towards studying auditory distance perception in concert halls with virtual acoustics. The aims of this investigation were to evaluate the feasibility of the auralizations and the system to study perceived distances as well as to obtain first evidence on the effects of hall acoustics and the source materials to distance perception. Auralizations were made from measured spatial impulse responses in two concert halls at 14 and 22 meter distances from the center of a calibrated loudspeaker orchestra on stage. Anechoic source materials included symphonic music and pink noise as well as signals produced by concatenating random segments of anechoic instrument recordings. Forty naive test subjects were blindfolded before entering the listening room, where they verbally reported distances to sound sources in the auralizations. Despite the large variance in distance judgments between the individuals, the reported distances were on average in the same range as the actual distances. The results show significant main effects of halls, distances and signals, but also some unexpected effects associated with the presentation order of the stimuli.