Ludovick Lepauloux

Acoustic context recognition using local binary pattern codebooks

Automatic context recognition enables mobile devices to adapt their configuration to different environments and situations. This paper investigates the use of acoustic cues as a means of recognising context. The majority of existing approaches exploit Mel-scaled cep-stral coefficients (MFCCs) developed for the analysis of speech signals. The hypothesis in this paper is that new features are needed in order to capture complex acoustic structure. The paper introduces the use of local binary pattern (LBP) analysis which is used to complement MFCCs with acoustic texture information. The second contribution relates to a bag-of-features extension which clusters LBPs into a small number of codewords. Both approaches outperform the current state of the art and the latter is particularly appealing for embedded applications in which computational efficiency is paramount.

The open-set problem in acoustic scene classification

Acoustic scene classification (ASC) has attracted growing research interest in recent years. Whereas the previous work has investigated closed-set classification scenarios, the predominant ASC application is open-set in nature. The contributions of the paper are (i) the first investigation of ASC in an open-set scenario, (ii) the formulation of open-set ASC as a detection problem, (iii) a classifier tailored to the open-set scenario and (iv) a new assessment protocol and metric. Experiments show that, despite the challenge of open-set ASC, reliable performance is achieved with the support vector data description classifier for varying levels of openness.

Acoustic context recognition for mobile devices using a reduced complexity SVM

Automatic context recognition enables mobile devices to react to changes in the environment and different situations. While many different sensors can be used for context recognition, the use of acoustic cues is among the most popular and successful. Current approaches to acoustic context recognition (ACR) are too costly in terms of computation and memory requirements to support an always-listening mode. This paper describes our work to develop a reduced complexity, efficient approach to ACR involving support vector machine classifiers. The principal hypothesis is that a significant fraction of training data contains information redundant to classification. Through clustering, training data can thus be selectively decimated in order to reduce the number of support vectors needed to represent discriminative hyperplanes. This represents a significant saving in terms of computational and memory efficiency, with only modest degradations in classification accuracy.

A comparison of different loudspeaker models to empirically estimated non-linearities

This paper investigates and questions the suitability of modelling non-linear loudspeaker distortion with scalar diagonal (SD) Volterra series. This approach, popular in studies of non-linear acoustic echo cancellation (NAEC), is compared to an alternative non-scalar diagonal (NSD) model. The new model is estimated empirically but based on the theoretical underpinnings of non-linear convolution. Using common, real-speech test signals, the loudspeaker outputs synthesised by each model are evaluated objectively through their comparison to real loudspeaker outputs measured in controlled conditions. Results show that non-linear distortion estimated with the NSD model better reflects that measured empirically. We also show that NAEC experiments conducted with SD loudspeaker models have the potential to over-exaggerate performance, whereas those conducted with an NSD model better reflect practical performance.

An experimental framework for the derivation of perceptually-optimal noise suppression functions

This paper presents a novel experimental framework designed to derive, through subjective testings, noise suppression functions which are perceptually optimal under specific experimental conditions. Noisy speech sequences are continuously processed according to a gain curve function of the a priori SNR that listeners are required to adjust two points at a time with respect to specified perceptual criteria. An experiment based on this framework is reported testing one specific combination of speech and noise signals. The specified perceptual criterion was the suitability for a phone conversation. The resulting mean experimental gain function shows a statistically significant deviation from an ideal Wiener filter. Experiments based on this framework are repeatable, suit untrained listeners and are considerably faster than conventional subjective testing methods, without the necessity to place restrictive assumptions on the assessed noise suppression function.