Probability Distributions.- Linear Models for Regression.- Linear Models for Classification.- Neural Networks.- Kernel Methods.- Sparse Kernel Machines.- Graphical Models.- Mixture Models and EM.- Approximate Inference.- Sampling Methods.- Continuous Latent Variables.- Sequential Data.- Combining Models.

Pattern Recognition and Machine Learning

This article describes research and development on car noise and vibration by the German car manufacturer Opel. Opel now has one of the world's largest anechoic chambers, 22m long by 16m wide. It can measure 'pass-by' noise in any weather, and has a powertrain test rig, which allows the whole powertrain and exhaust system to be examined away from the bodywork. By making vibrations visible, holography is helping to reduce the development time of a new car model by up to two months; it uses the interference principle to analyse structures without actual contact. Using holographic patterns, acoustic experts can work with design engineers on modifications to eliminate vibration. Psychoacoustics is perhaps the most exciting area of research. It applies the theory that sound can counter behaviour as well as influence it. If a driver is being aggressive or driving too fast, sound inside the car can be used to alter his driving style. By 'learning' a driver's personality, a car could take appropriate action to ensure safety and comfort by producing pleasant sounds in the car. However, these methods are still in the experimental stage. Opel is also conducting research on the stressful effects of noise and vibrations in cars.

The science of sound

When an audio—visual event is perceived in the natural environment, a physical delay will always occur between the arrival of the leading visual component and that of the trailing auditory component. This natural timing relationship suggests that the point of subjective simultaneity (PSS) should occur at an auditory delay greater than or equal to 0 msec. A review of the literature suggests that PSS estimates derived from a temporal order judgment (TOJ) task differ from those derived from a synchrony judgment (SJ) task, with (unnatural) auditory-leading PSS values reported mainly for the TOJ task. We report data from two stimulus types that differed in terms of complexity— namely, (1) a flash and a click and (2) a bouncing ball and an impact sound. The same participants judged the temporal order and synchrony of both stimulus types, using three experimental methods: (1) a TOJ task with two response categories (“audio first” or “video first”), (2) an SJ task with two response categories (“synchronous” or “asynchronous”; SJ2), and (3) an SJ task with three response categories (“audio first,” “synchronous,” or “video first”; SJ3). Both stimulus types produced correlated PSS estimates with the SJ tasks, but the estimates from the TOJ procedure were uncorrelated with those obtained from the SJ tasks. These results suggest that the SJ task should be preferred over the TOJ task when the primary interest is in perceived audio—visual synchrony.

/pdf/audiovisual-synchrony-and-temporal-order-judgments-effects-8fjt8ikxm9.pdf

Audiovisual synchrony and temporal order judgments: Effects of experimental method and stimulus type

Conventional speaker recognition systems perform poorly under noisy conditions. Inspired by auditory perception, computational auditory scene analysis (CASA) typically segregates speech by producing a binary time-frequency mask. We investigate CASA for robust speaker identification. We first introduce a novel speaker feature, gammatone frequency cepstral coefficient (GFCC), based on an auditory periphery model, and show that this feature captures speaker characteristics and performs substantially better than conventional speaker features under noisy conditions. To deal with noisy speech, we apply CASA separation and then either reconstruct or marginalize corrupted components indicated by a CASA mask. We find that both reconstruction and marginalization are effective. We further combine the two methods into a single system based on their complementary advantages, and this system achieves significant performance improvements over related systems under a wide range of signal-to-noise ratios.

/pdf/casa-based-robust-speaker-identification-2m24zduajs.pdf

CASA-Based Robust Speaker Identification

The temporal perception of simple auditory and visual stimuli can be modulated by exposure to asynchronous audiovisual speech. For instance, research using the temporal order judgment (TOJ) task has shown that exposure to temporally misaligned audiovisual speech signals can induce temporal adaptation that will influence the TOJs of other (simpler) audiovisual events (Navarra et al. (2005) Cognit Brain Res 25:499–507). Given that TOJ and simultaneity judgment (SJ) tasks appear to reflect different underlying mechanisms, we investigated whether adaptation to asynchronous speech inputs would also influence SJ task performance. Participants judged whether a light flash and a noise burst, presented at varying stimulus onset asynchronies, were simultaneous or not, or else they discriminated which of the two sensory events appeared to have occurred first. While performing these tasks, participants monitored a continuous speech stream for target words that were either presented in synchrony, or with the audio channel lagging 300 ms behind the video channel. We found that the sensitivity of participant’s TOJ and SJ responses was reduced when the background speech stream was desynchronized. A significant modulation of the point of subjective simultaneity (PSS) was also observed in the SJ task but, interestingly, not in the TOJ task, thus supporting previous claims that TOJ and SJ tasks may tap somewhat different aspects of temporal perception.

Audiovisual temporal adaptation of speech: temporal order versus simultaneity judgments.

In this study, we present a binaural scene analyzer that is able to simultaneously localize, detect and identify a known number of target speakers in the presence of spatially positioned noise sources and reverberation. In contrast to many other binaural cocktail party processors, the proposed system does not require a priori knowledge about the azimuth position of the target speakers. The proposed system consists of three main building blocks: binaural localization, speech source detection, and automatic speaker identification. First, a binaural front-end is used to robustly localize relevant sound source activity. Second, a speech detection module based on missing data classification is employed to determine whether detected sound source activity corresponds to a speaker or to an interfering noise source using a binary mask that is based on spatial evidence supplied by the binaural front-end. Third, a second missing data classifier is used to recognize the speaker identities of all detected speech sources. The proposed system is systematically evaluated in simulated adverse acoustic scenarios. Compared to state-of-the art MFCC recognizers, the proposed model achieves significant speaker recognition accuracy improvements.

https://orbit.dtu.dk/files/53464044/06178270.pdf

A Binaural Scene Analyzer for Joint Localization and Recognition of Speakers in the Presence of Interfering Noise Sources and Reverberation

Although the field of automatic speaker recognition (ASR) has been the subject of extensive research over the past decades, the lack of robustness against background noise has remained a major challenge. This paper describes a noise-robust speaker recognition system that combines missing data (MD) recognition with the adaptation of speaker models using a universal background model (UBM). For MD recognition, the identification of reliable and unreliable feature components is required. For this purpose, the signal-to-noise ratio (SNR) based mask estimation performance of various state-of-the art noise estimation techniques and noise reduction schemes is compared. Speaker recognition experiments show that the usage of a UBM in combination with missing data recognition yields substantial improvements in recognition performance, especially in the presence of highly non-stationary background noise at low SNRs.

/pdf/noise-robust-speaker-recognition-combining-missing-data-2v0p5oc0fu.pdf

Noise-Robust Speaker Recognition Combining Missing Data Techniques and Universal Background Modeling

A novel method for the estimation of the distance of a sound source from binaural speech signals is proposed. The method relies on several statistical features extracted from such signals and their binaural cues. Firstly, the standard deviation of the difference of the magnitude spectra of the left and right binaural signals is used as a feature for this method. In addition, an extended set of additional statistical features that can improve distance detection is extracted from an auditory front-end which models the peripheral processing of the human auditory system. The method incorporates the above features into two classification frameworks based on Gaussian mixture models and Support Vector Machines and the relative merits of those frameworks are evaluated. The proposed method achieves distance detection when tested in various acoustical environments and performs well in unknown environments. Its performance is also compared to an existing binaural distance detection method.

Sound Source Distance Estimation in Rooms based on Statistical Properties of Binaural Signals

In this paper, methods for improved parametric coding of transients are presented. We propose a signal model for coding of transients consisting of a sum of sinusoids each being amplitude-modulated by a different gamma envelope. These envelopes are characterized by an onset time, an attack and a decay parameter. An efficient method for estimating these parameters is presented. Further, methods are proposed that combine this transient model with a constant-amplitude sinusoidal model in order to achieve efficient coding of both stationary and transient signal parts. By rate-distortion optimization using a perceptual distortion measure, we combine variable rate bit allocation and segmentation in an optimal way. Formal, as well as informal, listening tests show that significant improvements can be achieved with the proposed model as compared to a state-of-the-art sinusoidal coder by the combination of optimal segmentation and amplitude modulated sinusoidal audio coding

/pdf/efficient-parametric-coding-of-transients-4quuqnpv2g.pdf

Efficient parametric coding of transients

This paper presents a rate-distortion optimization approach to hybrid sound coding. The advantages of sinusoidal and transform coding are combined by a rate-distortion optimization mechanism, using a perceptually relevant distortion measure based on spectral auditory masking. As a result, the coder can adapt to the input signal and to constraints such as bit rate. Listening test results show improved performance of the hybrid coder compared to the individual coding paradigms. There is a good correlation between the improved performance as reported by the listeners and the differences in distortion resulting from the perceptually relevant distortion measure. This confirms that the distortion measure used in the optimization is useful; moreover, it shows the feasibility of the rate-distortion optimization approach for hybrid sound coding.

S.L.J.D.E. van de Par

Papers

A Binaural Scene Analyzer for Joint Localization and Recognition of Speakers in the Presence of Interfering Noise Sources and Reverberation

Noise-Robust Speaker Recognition Combining Missing Data Techniques and Universal Background Modeling

Sound Source Distance Estimation in Rooms based on Statistical Properties of Binaural Signals

Efficient parametric coding of transients

Rate-distortion optimized hybrid sound coding