Video magnification (VM) provides an alternative health monitoring solution by enabling contactless and remote measurement of vital signs such as heart rate (DR).DR is a crucial biomarker for assessing the overall health of individuals, and the increased need to replace traditional wearable devices and avoid complex applications of sensors makes VM essential for in-home health assessment. VM can measure HR by detecting subtle changes of skin color due to blood flow, however, external factors such as subject or camera motion, variable lighting and skin tone can greatly affect performance. Previous work implemented methods that improved the overall accuracy of VM-based HR predictions but did not indicate if a given HR prediction was likely to be correct or incorrect. This work assesses the confidence of the predicted HR using a set of features derived during the VM process. Several machine learning models were evaluated to analyze the confidence parameters from three different VM processing methods. The results show an accuracy of92.1% for an SVM model that classifies if VM-based HR is correct or incorrect. The high classification accuracy indicates the effectiveness of the confidence metrics for discriminating between correct and incorrect HR predictions using VM.

The global COVID-19 pandemic increased the interest in automatic analysis and classification of cough sounds. However, developing such a system requires a large dataset of expert-labelled cough sounds, which remains elusive. Data augmentation techniques are often employed to train machine learning models with limited data, while ensuring model robustness to real-world variations. We have recently proposed the use of “natural” spectral data augmentation methods, including noise and reverberation [1]. In this paper, we further investigate the method by studying the alignment between training and testing environments. We augment training sound data with varying levels of reverberation and Gaussian noise, and evaluate augmented convolutional neural network models across a range of test environments on two audio classification tasks: speech command recognition and cough sound classification. Results demonstrate the broad robustness of the data-augmented deep learning model across various test environments. Furthermore, the augmented model's performance on cough classification is compared to 4 expert annotations on a large cough dataset and is seen to be near-human-level accuracy. This augmented model is recommended for classifying cough sounds in natural settings beyond laboratory conditions.

Cough diagnosis is important for the elderly population, since cough is a key symptom of many respiratory illnesses and conditions. This paper introduces a Transformer-based feature learning approach for the analysis of cough recordings. A Transformer network leveraging feature learning on a big data set is investigated from a feature engineering perspective, in order to find dedicated classification models that can improve overall performance. The latter was achieved through adopting AutoML post-processing techniques on different data sets, driven by the feature engineering process based on both feature selection and feature generation via nonlinear methods. It was found that this approach led to substantial improvements (in the order of 17% from 0.818 to 0.956 of accuracy) on practically all metrics of classification performance, with respect t o t hose obtained with standalone Transformers. Moreover, AutoML models using reduced number of features, either selected or generated, resulted in higher quality models. In particular, a model working only with 1.2 % of the features (nonlinearly generated from the 768 produced by the Transformer), outperformed the model using all of them. These results highlight that big data-derived machine learning models, when post-processed, can play an important role in adapting to small-data scenarios.

The potential for pressure sensitive mats to provide a means for ambient well-being sensing and assessment when placed within beds or chairs has been widely studied although many mats have limitations in capability and cost. This paper presents the results for a novel pressure sensitive mat that combines an Infrared (IR) based emitter/detector proximity sensor device with a flexible gel material (Hexyoo Scientific Inc. Opus Gel). The performance results for a bench prototype sensor that combines layers of gel that have differing colour and structure properties with the sensors is shown. Specifically, the step response of the prototype is analyzed for step-up and step-down loads representing a pressure range typical for human body pressures. The results show that the frequency response for the two cases are both low-pass in nature and that the frequency response for the step-down response is a loss of 10dB at 6Hz while the step-up response is narrower with a 10dB loss at 2Hz. This bandwidth is sufficient to allow assessment of human motions with the sensor, while also using significantly simpler and lower cost technology compared to previously reported fibre optic mats.

The ability to understand and visualize big data sets is of increasing interest to caregivers and clinicians as ambient home sensing can provide massive amounts of data related to the activities of residents. However, this data is only useful if it can be effectively and simply visualized for review and analysis. This paper presents the visualization of longitudinal data sets from ambient well-being sensors deployed in 3 residences that have a spousal pair dyad where 1 resident has been diagnosed with Mild Cognitive Impairment or Dementia and the spousal partner is acting as a caregiver. The paper presents the differences in activity and behaviour that can be observed in the 3 residences by comparing two 30-day periods prior to and one 30-day period during COVID-19 social isolation precautions. The work shows the potential for this circle plot based visualization technique to summarize resident activity and also to convey external factors such as the variation in solar day that can itself influence behaviour.

An emerging urban challenge will be the proliferation of Remotely Piloted Aircraft Systems (RPAS) or drones, as their usage grows and drones fill the urban skies. Urban airspace will include many more Uninhabited Aerial Vehicles (UAVs), and related accidents and mishaps will also increase. Applications of UAVs in urban environments include photography and film-making, security monitoring, real estate, construction, property and infrastructure inspections, leisure, public safety (fires, natural disasters, investigations), traffic, and much more. Emerging UAV applications include last-mile drone delivery services. This paper discusses urban scenarios where drones are more ubiquitous. Monitoring and related safety of these UAVs will be increasingly important. Though multi-modal purpose-built drone tracking and monitoring systems will be the most effective solution for detection and tracking of these RPAS, during the transition to more regular drone use in urban areas and in everyday urban applications, a more rapid-deployment and agile detection system is needed that does not require installation of hardware. In this paper, sensory substitution is presented as an approach to use ambient and pre-existing microphones and AI techniques to detect the presence of RPAS. Preliminary results show promise for this agile IoT method as a key solution to this emerging urban challenge.

Internet of Things (IoT) continues proliferation. Hospitals, energy industry, power grid, food and water, transportation, etc., are critical asset resources. Critical systems must remain steadfast and reliable. This IoT evolution, sometimes termed, industrial IoT, is significant for critical infrastructure. In harsh climates, human lives can depend daily on these systems functioning well. Better situational awareness can improve resilience of these critical infrastructures and make them increasingly robust. In this paper, the method of agile IoT for critical infrastructure resilience is presented, and an agile IoT model is presented to use signal processing middleware to enable existing sensors commonly deployed in critical infrastructure to be agilely repurposed to: 1) add parallel communication methods and/or 2) sense additional mechanical/physical parameters using existing hardware. The results of the proposed method are applied to use common temperature sensors to detect nontemperature parameters: fluid flow in a pipe (manufacturing or water), ice buildup (transportation or energy infrastructure), and of mechanical door state (health venues of life-saving medicine stored in hospital refrigerators).

Sensors and data analytics have a tremendous potential to improve the resilience of the electricity system. Improving the analytics in a smart grid, and providing information to operators so that problems can be resolved quickly, may serve to improve the resiliency of the electricity system. Effective use of sensors and analytics will enable more timely a response, and more efficient use of personnel, specialty equipment, and site location specifics because of better information for deployment teams to address damage. Detection of physical characteristics such as vibration, ice build up, hot spots, aging and deterioration of assets/equipment, metal fatigue and other considerations could prevent disruptions. In cold climates, ice storms can cause outages, and extreme weather events are constantly a threat to electricity towers. A system was developed to collect information from sensors, as well as relevant analytics to detect abnormalities to address damage, and for operator visualizations screens.

In Internet of Things (IoT), there are many future applications in smart environments. This paper first presents a possible scenario of a family making use of the IoT in their daily living, and then work leading to integration of unobtrusive sensors in the smart environment is discussed. In the applications discussed, Bluetooth connections linked and connected the sensors that were used in the wireless sensor network as an intuitive sensor communication platform. The paper presents results of a methodology and experimentation for human identification using sensors and microcontrollers and discusses how this can lead to integration and increased personalization within an IoT smart environment using only unobtrusive sensors.

This paper presents robust adaptive information content system design using contiguous buffering scheme buffers for high data rate real-time applications like e-health, telemedicine and 3-D video using IP network. The design improves the resiliency of the realtime systems under adverse network conditions like congestion and/or temporary loss of connectivity and ensures that critical application data is not lost. The proposed adaptive information content management design for a high data rate real-time application shows how information content can be adaptively managed using contiguous sender buffer to reduce the data transfer rate when underlying network capacity changes over a time period, improving the survivability interval of the system. There is a tradeoff between loss of features, data rate reduction, criticality of the lost information and survivability time of the systems.

The performance of a typical linear echo canceller is limited by changes that occur in the acoustic echo path along with nonlinear echoes arising from electrodynamic loudspeaker distortion. This paper presents a subband structure based on proportionately adapted second-order Volterra filters for acoustic echo cancellation under these adverse conditions. The proposed structure allows the time and frequency domain nature of echo path changes to be exploited, resulting in faster convergence and tracking compared to an equivalent fullband structure while requiring significantly less computational complexity. Experimental results, based on measured data from a practical hands-free environment under changing conditions, demonstrate the improved echo cancellation performance of the proposed structure.

This paper presents a particle filter approach to spectral amplitude speech enhancement. Spectral amplitudes are known to exhibit inter-frame dependencies and non-Gaussian statistics; however, incorporating these properties makes closed-form solutions intractable. Using the particle filter framework allows the presented algorithm to model the speech spectral amplitudes as an autoregressive process with Laplace distributed excitation. Two variants of the standard algorithm are also presented: one that uses an interacting multiple model approach to account for transitions between active speech and silence intervals, and one that allows for phase differences between the clean speech and noise complex Fourier transform coefficients. All of the particle sampling distributions are constrained to take the measurement into account, improving sampling efficiency. In experiments using wideband speech and real recorded noise the proposed algorithm variants are shown to offer natural-sounding output speech, with objective evaluation results that compare favorably to existing particle filter speech enhancement algorithms. The multiple model variant is found to improve inter-speech noise reduction, while the phase variant improves performance when the signal-to-noise ratio is low.

This paper proposes a simple modification to adaptive echo cancellers to prevent rapid divergence due to doubletalk conditions. The proposed structure introduces a delay only into the adaptation algorithm, which compensates for doubletalk detector response time at the onset of near-end speech. The structure is described and analyzed for the NLMS and cross-correlation-based doubletalk detector algorithms. The method introduces no delay into the return path, and no additional constraints on adaptation step size arise. Simulations with ITU-T Q. 168 tests confirm that the proposed structure prevents divergence at the onset of doubletalk for near-end to echo signal ratios of-26 to -6 dB.

Acoustic echo paths can change frequently in response to minor changes in the near-end room. Using echo path measurements obtained using commercial wideband hands-free terminals, it is demonstrated that the effects of these changes are not uniform across time and frequency. A subband echo canceler combined with a proportionate step-size adaptation algorithm is proposed to exploit time and frequency response characteristics of echo path changes to achieve faster tracking and higher echo attenuation. Improvements of 510 dB compared to fullband structures are observed for real changing echo environments with wideband speech inputs.

This paper proposes a new robust structure for combining acoustic echo cancellation and microphone array beamforming. The new structure employs short adaptive filters in the acoustic echo cancellers that precede the beamformer, allowing for quick tracking of time variations within the hands-free environment. The beamformer is followed by a single echo canceller to further suppress echo. Simulations show that by taking into consideration the dynamic behaviour of the hands-free environment, the proposed structure reconverges faster compared to standard combined structures during changing acoustical environment conditions. An improvement in ERLE of up to 10 dB is observed during reconvergence.

Traditionally held beliefs assert that the minimum mean-square error (MMSE) of subband adaptive filters is higher than that of the equivalent fullband filter. In this paper we show that this is not always the case. Oversampled subband adaptive filters with white and speech-like excitation are observed to produce MMSE levels well below those of the equivalent fullband filter and even below the linear and time-invariant (LTI) Wiener filter. This observation is attributed to non-Wiener behavior, which has previously been observed in adaptive filters with narrowband inputs.

With the advent of voice over internet protocol (VoIP) service, assessing its voice quality is an area of intense research interest. Due to time-consuming and expensive natures of widely accepted subjective mean opinion score (MOS) test, objective methods are often used as an alternative. This paper presents applicability and accuracy analysis of several leading objective methods in quality testing. Particularly, the paper focuses on packet loss, which is one of major impairments in VoIP. A speech database covering typical packet loss conditions is designed. The subjective MOS test is conducted and the results are evaluated with objective MOS. The database is also used to verify a non-intrusive VoIP speech quality assessor the authors developed (El-Hennawey et al., 2006). The results show that perceptual evaluation of speech quality (PESQ) based algorithms are generally acceptable for quantifying the effects of packet loss. In addition, we also find that their performance is limited for codec G.711 without packet loss concealment (PLC).

Doubletalk detection is an important part of a practical echo canceller implementation, but a difficult problem is calibrating the doubletalk detector for arbitrary environments and input signals. In this paper, it is shown that a statistical model of a doubletalk detection variable's probability density function (PDF) can be used to obtain an optimal detection threshold and expected detection performance curves. In particular, a statistical analysis of a recently proposed cross-correlation-based doubletalk detector is presented. The doubletalk detection variable is modeled in terms of its constituent parameter estimators, resulting in conditional PDFs in the absence and presence of doubletalk. These are used to obtain a signal-adaptive detection threshold for calibration, and to provide expected doubletalk detection probability. Simulations are presented comparing the theoretical and measured detection probability compared to a fixed detection threshold for speech input and doubletalk signals. The results indicate a close agreement with the proposed model for moderate-to-high levels of doubletalk

This paper investigates postfiltering for residual echo suppression in networks employing low-bit-rate speech compression in the echo path. Simulations show that the residual echo from nonlinear vocoder distortion with ITU G.729 is proportional to the input signal LPC spectrum. An algorithm is proposed to estimate the residual echo power spectrum using a frequency-dependent scaling factor. The algorithm is incorporated into a psychoacoustic postfilter for residual echo suppression and compared to an existing estimator with a fixed scaling factor. Experiments with speech input and near-end signals show an average 0.85 dB lower spectral distortion and 0.4 higher estimated mean opinion score

This paper addresses the problem of doubletalk detector calibration for acoustic echo cancellers in hands-free environments such as videoconferencing. A statistical model of a recently proposed doubletalk detector is used to show that optimal detection thresholds are dependent on the input signal and the adaptive filter error. A signal-adaptive algorithm is proposed for calculating an optimal threshold for arbitrary input signals and echo path environments. Simulation results verify the improvement in detection probability offered by the proposed algorithm compared to simple empirical calibration methods

In this paper, we propose a new system for classifying audio segments as speech or music. The proposed system improves classification accuracy, particularly in low signal-to-noise ratio (SNR) environments. The system selects the features with the highest classification accuracy that corresponds to the SNR value. The value of this features are compared to certain thresholds, which are also adapted to the SNR. Multi-expert method of combining the features to improve classification accuracy is implemented. A new feature, termed the variance of low-band energy ratio, is also introduced. This feature produces large improvements in classification accuracy at low SNR. Performance of the proposed system is evaluated for different SNR using a library of speech and music audio segments. Using one-second segments it is shown that the proposed system can enhance the classification accuracy by 22% at SNR = -15 dB, and obtain classification accuracy of 90.3% at SNR = 0 dB.

In this paper, standard echo canceller performance measures are evaluated in terms of psychoacoustic aspects of human hearing. The focus is on wideband speech communications systems with long round-trip delays of 200 ms and up present in the transmission path. The results of a simple acoustic echo cancellation experiment are analyzed with a standard psychoacoustic model, revealing that steady-state echo return loss enhancement and mean square error cannot be used to determine whether residual echo is perceivable in the presence of background noise. In addition, a simple modification to the normalized least mean square (NLMS) algorithm is introduced by adding a perceptual preemphasis filter. Simulation results and listening tests show that it is possible to improve the perceived performance of an echo canceller during convergence by placing greater emphasis on frequencies at which the human auditory system is most sensitive.

A masking threshold constrained Kalman filter for speech enhancement is derived in the paper. A key step in a traditional Kalman filter requires minimizing an estimation error variance between a clean signal and its estimation. Our new method is to minimize the estimation error variance under the constraint that the energy of the estimation error is smaller than a masking threshold, computed from both time-domain forward masking and frequency-domain simultaneous masking properties of human auditory systems. The new Kalman filter provides a theoretical base for the application of the masking properties in Kalman filtering for speech enhancement. Due to the high computation cost of the proposed perceptually constrained Kalman filter, a perceptual post-filter concatenated with a standard Kalman filter is also proposed as a heuristic alternative for real-time implementation. The post-filter is constructed to make the estimation error obtained from the Kalman filter lower than the masking threshold. A wavelet Kalman filter with post-filtering is introduced to further reduce the computational load. Experimental results with colored noise show that the new constrained Kalman filter method produces the best performance when compared with other recent methods, and that the proposed heuristics with post-filtering can also produce a significant performance gain over other recent methods.

This paper proposes a new algorithm for classifying an audio segment as speech or non-speech. The proposed algorithm is capable of handling reverberation and low signal-to-noise environments; therefore, it is suitable for hands-free applications. The algorithm divides an audio segment into frames, estimates the presence of pitch in each frame, and calculates a pitch ratio parameter. This parameter is then used to classify the audio segment. The threshold used in calculating this parameter is adapted to accommodate different environments. The performance of the proposed algorithm is evaluated for different signal-to-noise ratios and different segment sizes using a library of audio segments. The library includes speech segments and nonspeech segments such as fan noise and cocktail noise. Using 0.4 second segments it is shown that the proposed algorithm can achieve a correct decision for 95.7% of the speech segments and 96.7% of the nonspeech segments under reverberant conditions.

The paper proposes a new algorithm to estimate and suppress highly non-stationary background noise from speech. The algorithm consists of two spectral detectors. The first one uses strict criteria and is based on the Fourier transform of the AMDF (average magnitude difference function). The second one uses loose criteria and is based on variable start minima search. By combining the two detectors, the algorithm instantaneously detects and tracks a sudden change of noise energy level. The proposed algorithm is then merged to conventional MMSE-STSA (MMSE short time spectral amplitude) to suppress non-stationary noise in speech. Simulation results are given to show the superiority of our proposed algorithm.

A wavelet Kalman filter combined with masking properties of the human auditory system is proposed for enhancing speech degraded by colored noise. A wavelet domain speech state-space model is first constructed, with a corresponding Kalman filter. A post-filter, considering both time and frequency masking properties, is then applied. From the calculated masking threshold, the noisy speech spectrum is further enhanced. Simulation results show that the proposed approach has the best performance compared with either classic or recently introduced methods, evaluated using ITU-T P.862 PESQ scores. The proposed approach also has a reduced complexity compared to a time-domain implementation of the Kalman filter with a perceptual post-filter.

This paper investigates performance measures of adaptive echo cancellers for packet-based telephony. It is shown that steady-state echo return loss enhancement (ERLE) does not accurately reflect perceived echo canceller convergence when background noise is present. An upper bound is derived for the maximum perceivable ERLE achievable in practice, and an algorithm is introduced for calculating ERLE that incorporates these masking effects based on a perceptual hearing model. Simulation and informal listening test results show a clear correspondence between the new performance measure and the perceptual upper bound induced by background noise

This paper presents a novel acoustic echo canceller structure based on combining the filtered-X LMS algorithm with an LPC-based speech coder for use in videoconferencing and VoIP. The algorithm updates coefficients using filtered versions of the input and error signals obtained by directly tapping the short-term excitation signal from the speech decoder, and by filtering the error signal with a bank of FIR decorrelation filters constructed from the LPC synthesis filter coefficients. The proposed algorithm was implemented using ITU G.729, and simulation results with 2000-tap room impulse responses show a faster and more constant rate of convergence than NLMS using speech input signals and an average 10 dB greater ERLE observed during convergence.

This paper proposes a generalized proportionate variable step-size algorithm based on affine projection. It controls the step size for each tap individually based on the gradient approximated by the difference between the current coefficient and an averaged filter coefficient with delay. This step-size control is specifically effective for tracking fast changing acoustic environments. It is shown that the P-NLMS family and the ES algorithm are special cases of the proposed algorithm in terms of step-size control. Simulation results in the context of echo cancellation in artificial and real environments demonstrate that this step-size control combined with the affine projection algorithm (APA) reduces the echo by up to 5 dB compared to the standard APA. The best tracking performance, without a priori knowledge on the acoustic environment, among the same family of algorithms is obtained.

A new method for speech enhancement in colored noise is proposed. A Kalman filter concatenated with a post-filter based on the masking properties of the human auditory system is proposed for the problem. A recursive approach to compute the noise covariance matrix is used for estimating the colored noise statistics. In the post-filter, both time domain masking properties and frequency domain masking properties are taken into account. From the calculated masking level, the noisy speech spectrum is adjusted accordingly. Simulation results show that the proposed approach has the best performance compared with other recent methods, evaluated with PESQ scores.

This paper proposes a robust near-field adaptive beamformer for microphone array applications in small rooms. Robustness against location errors is crucial for near-field adaptive beamforming due to the difficulty in estimating near-field signal locations especially the radial distances. A near-field regionally constrained adaptive beamformer is proposed to design a set of linear constraints by filtering on a low rank subspace of the near-field signal over a spatial region and frequency band such that the beamformer response over the designed spatial-temporal region can be accurately controlled by a small number of linear constraint vectors. The proposed constraint design method is a systematic approach which guarantees real arithmetic implementation and direct time domain algorithms for broadband beamforming. It improves the robustness against large errors in distance and directions of arrival, and achieves good distance discrimination simultaneously. We show with a nine-element uniform linear array that the proposed near-field adaptive beamformer is robust against distance errors as large as /spl plusmn/32% of the presumed radial distance and angle errors up to /spl plusmn/20/spl deg/. It can suppress a far field interfering signal with the same angle of incidence as a near-field target by more than 20 dB with no loss of the array gain at the near-field target. The significant distance discrimination of the proposed near-field beamformer also helps to improve the dereverberation gain and reduce the desired signal cancellation in reverberant environments.

The paper investigates the effects of packet loss and delay jitter on speech quality in voice over Internet protocol (VoIP) scenarios. A new formula is proposed to quantify these effects and incorporated into ITU-T G.107, the E-model. In the simulation, codecs ITU-T G.723.1 and G.729 are used; random packet loss and Pareto distributed network delay are introduced. The prediction errors range between -0.20 and +0.12 MOS (mean opinion score). The formula extends the coverage of the current E-model, and is very useful in MOS prediction as well as network planning.

This paper proposes a new algorithm for predicting audio packet playout delay for voice conferencing applications that use silence suppression. The proposed algorithm uses a hidden Markov model (HMM) to predict the playout delay. Several existing algorithms are reviewed to show that the HMM technique is based on a combination of various desirable features of other algorithms. Voice over Internet protocol (VoIP) applications produce packets at a deterministic rate but various queuing delays are added to the packets by the network causing packet interarrival jitter. Playout delay prediction techniques schedule audio packets for playout and attempt to make a reasonable compromise between the number of lost packets, the one-way delay and the delay variation since these criteria cannot be optimized simultaneously. In particular, this paper will show that the proposed HMM technique makes a good compromise between the mean end-to-end delay, end-to-end delay standard deviation and average packet loss rate.

This paper discusses eigenvector constraint design for broadband adaptive array beam forming. The constraint equations are constructed using the real and imaginary parts of the steering vectors and the response vectors, thus perm its real arithmetic and guarantees real outputs. A simple and elegant structure is developed for implementing the designed beam former by a Generalized Sidelobe Canceler (GSC). This structure reduces the computational complexity from O (N2) to O (N), where N being the dimension of the weight vector.

This paper proposes an acoustic echo cancellation structure for hands-free synthetic surround sound applications, such as multiple participant conferencing, and virtual reality applications. Voice over Internet protocol (VoIP) and other virtual reality applications can benefit from the addition of 3D spatial audio generated by more than two loudspeakers. When full-duplex audio is present in a system, however, acoustic echo cancellation is required to eliminate the feedback echo path. The acoustic echo cancellation structure proposed by this paper is based on the acoustic echo canceller per spatial region allocation scheme previously introduced by the authors for two channel synthetic stereo. This paper shows that the spatial region allocation scheme is extensible to any number of channels which makes it extremely versatile and flexible, especially for surround sound audio. Microsoft Direct X 7, a commonly used application programmer interface (API), was used in our simulations to generate the 3D spatial audio on a PC.

This paper presents a robust algorithm for voice activity detection (VAD) based on newly established properties of the higher order statistics (HOS) of speech. Analytical expressions for the third and fourth-order cumulants of the LPC residual of short-term speech are derived assuming a sinusoidal model. The flat spectral feature of this residual results in distinct characteristics for these cumulants in terms of phase, periodicity and harmonic content and yields closed-form expressions for the skewness and kurtosis. Important properties about these cumulants and their similarity with the autocorrelation function are revealed from this exploratory part. They show that the HOS of speech are sufficiently distinct from those of Gaussian noise and can be used as a basis for speech detection. Their immunity to Gaussian noise makes them particularly useful in algorithms designed for low SNR environments. The proposed VAD algorithm combines HOS metrics with second-order measures, such as SNR and LPC prediction error, to classify speech and noise frames. The variance of the HOS estimators is quantified and used to yield a likelihood measure for noise frames. Moreover, a voicing condition for speech frames is derived based on the relation between the skewness and kurtosis of voiced speech. The performance of the algorithm is compared to the ITU-T G.729B VAD in various noise conditions, and quantified using the probability of correct and false classifications. The results show that the proposed algorithm has an overall better performance than G.729B, with noticeable improvement in Gaussian-like noises, such as street and parking garage, and moderate to low SNR.

This paper proposes a novel acoustic echo cancellation structure intended for multiple participant, full-duplex, hands-free voice over Internet protocol (VoIP) conferencing. A synthetic stereo image is generated through the use of spatialization functions, which are used to assist the near-end user in distinguishing between far-end talkers. The proposed synthetic stereo structure, which uses a single echo canceler per spatial region, is compared to structures with a single echo canceler per channel and to stereophonic echo cancelers. The proposed structure does not suffer from correlation of the reference signals, which can cause system misconvergence in structures that use a single canceler per channel. Correlation of the reference signals is eliminated since monaural signals are transmitted at the far-end and static spatialization functions are used at the near-end to allocate each participant to a spatial region.

This paper proposes a synthetic stereo acoustic echo cancellation structure with microphone array beamforming for multiple participant voice over Internet protocol (VoIP) conferencing. The proposed acoustic front-end combines a synthetic stereo structure, acoustic echo canceler and microphone array beamformer. Synthetic stereo is used to add spatial information to signals from multiple participants that facilitates distinguishing between participants. Microphone array beamforming is then combined with the single canceler per spatial region echo cancellation structure to reduce noise and reverberation for near-end speech pickup and to cancel echo.

A noise-reduction (NR) enhancement to the generalized multi-delay frequency domain algorithm (GMDF) involving minimal additional computations is presented. Findings are presented on the ability of the algorithm to suppress street, office, air-conditioning and car noise from the microphone signal in hands-free telephony. For an additional eleven N/P real multiplies and eight N/P real additions per output sample, where N is the block size and P is the amount of output samples per GMDF iteration, results show an average increase in echo return loss enhancement (ERLE) of 9 dB at low SNR (-5 dB) and 6 dB at higher SNR (15 dB).

This paper describes the application of array optimization techniques in the near-field of a microphone array. It is shown that the optimum weights obtained for microphone spacing equal to a half wavelength are real, aside from the propagation delay, and inversely proportional to the distance from each microphone to the focal point. When the microphone spacing is less than a half wavelength, the optimum weights are super-directive requiring both an amplitude and phase adjustment in addition to delay-and-sum beamforming. In the super directive regime, the maximum array gain increases as the distance to the focal points decreases. The technique is suitable for applications where the source of interest is located in the near-field region but interfering sources are located farther from the array.

A new method for speech enhancement based on optimal filtering, subbands, and higher-order cumulants is proposed in this paper. The key idea is to use the 4th cumulant to estimate the parameters required for the enhancement fil-ters: It is shown that the kurtosis of noisy speech may be used to estimate the SNR and the probability of speech presence when speech is divided in narrow bands and mod-eled as a sinusoidal signal. The resulting algorithm is tested in typical mobile noise conditions and proves effective under such types as street, office and fan noises. Compared to the TIA IS-127 standard for noise reduction, the proposed algorithm is better at preserving the harmonic structure of the speech and results in overall more noise reduction in Gaussian-like conditions. However, this comes at the cost of slightly more noise artifact, mostly at very low SNR and non-Gaussian conditions. 1.

This article addresses the problem of instantaneous signal-to-noise ratio (SNR) estimation during speech activity for the purpose of improving the performance of speech enhancement algorithms. It is shown that the kurtosis of noisy speech may be used to individually estimate speech and noise energies when speech is divided into narrow bands. Based on this concept, a novel method is proposed to continuously estimate the SNR across the frequency bands without the need for a speech detector. The derivations are based on a sinusoidal model for speech and a Gaussian assumption about the noise. Experimental results using recorded speech and noise show that the model and the derivations are valid, though not entirely accurate across the whole spectrum; it is also found that many noise types encountered in mobile telephony are not far from Gaussianity as far as higher statistics are concerned, making this scheme quite effective.

With reduction of the block size (increasing the number of subfilters) regular generalized multi-delay frequency domain adaptive filter (GMDF/spl alpha/) can achieve low throughput delay at the expense of the system performance. In situations where zero delay is desirable, we propose a new method which is not dependent on the block size. In addition, by using an adaptive reconstruction filter, further performance gains can be achieved with minimal additional computation complexity. Results from experiments performed in a conference room show an increase in the average echo return loss enhancement (ERLE) of >2.5 dB for acoustic echo cancellation over the traditional moving average reconstruction filter.

This paper proposes a theoretical model and an ITU-T H.323 compliant empirical measurement method for the acoustic round trip delay (ARTD) of a voice over Internet protocol (VoIP) conference. The empirical measurement method is able to compute the complete acoustic round trip delay that the user perceives. The acoustic round trip delay is experimentally measured by slicing the real-time signal path and inserting test hooks on the Carleton University VoIP platform. Current VoIP systems cannot measure the acoustic round trip delay. An acoustic domain verification test is used to check the ARTD empirical measurements for accuracy.

This paper focuses on multilayer perceptron neural networks where the activation functions are adaptive and where each neuron synapses is modelled by a finite impulse response (FIR) filter. A simplified architecture consisting of a variable activation (VA) function which is sandwiched between two FIR synapses is studied. The VA function consists of, a mixed linear-tank sigmoid with a parameter which controls the linear region. The VA parameters and FIR synaptic weights are updated using a modified form of the instantaneous-cost (IC) temporal backpropagation algorithm. Simulations for identifying cascaded nonlinear transfer functions with internal memory and arbitrary activation functions illustrate the improved modelling performance over models with non-adaptive activation functions.

This paper describes the application of array optimization techniques to improving the near-field response of an arbitrary microphone array. The optimization exploits the differences in wavefront curvature between near-field and far-field sound sources and is suitable for reverberation reduction in small rooms. The optimum near-field beamformer provides increased array gain over that obtained from a uniformly weighted delay-and-sum beamformer.

In this paper a modified form of the partial conjugate gradient algorithm is presented for use in nonlinear filtering using neural networks. The algorithm is based on using a gradient average window to provide a trade-off between convergence rate and complexity which, depending on the choice of averaging window, is (in both complexity and speed of convergence) intermediate between the conventional backpropagation (BP) algorithm and the Newton methods. An additional simplification is introduced by replacing the calculated optimum step size /spl alpha//sub k/ by a normalized step size /spl alpha/~, in the same manner as the normalized LMS algorithm. This new algorithm is applied to a cascaded neural network/nonlinear least mean squares structure for the identification of a nonlinear system. This proposed algorithm demonstrates improved convergence rates with even small choices of window size.

One of the limitations of linear adaptive echo cancellers is nonlinearities which are generated mainly in the loudspeaker. The complete acoustic channel can be modelled as a nonlinear system convolved with a linear dispersive echo channel. Two new acoustic echo canceller models are developed to improve nonlinear performance. The first model consists of a time-delay feedforward neural network (TDNN) and the second model consists of a memoryless neural network followed by an adaptive normalized least mean square (NLMS) structure. Simulations demonstrate that both neural network based structures improve the echo return loss enhancement (ERLE) performance compared to a linear NLMS acoustic echo canceller. Experimental results using the TDNN improved the ERLE by 10 dB at low to medium loudspeaker volumes.

Due to computational and convergence speed restraints, acoustic echo cancellers used for desktop handsfree units typically use an adaptive filter significantly shorter in length than the length of the enclosure/room impulse response it is attempting to model. A simple measurement based on this impulse response is proposed that gives the maximum obtainable depth of cancellation for a given filter length. By comparing this measurement with actual Echo Return Loss Enhancement measurements for different adaptive filter lengths, the optimum filter length for a given handsfree enclosure and transducers can be determined.>