Convolutional Noise Research Articles

A comparative study of noise estimation algorithms for nonlinear compensation in robust speech recognition

Nonlinear compensation models make use of a nonlinear mismatch function, which characterizes the joint effects of additive and convolutional noise, to realize noise-robust speech recognition. Representative compensation models consist of vector Taylor series (VTS), data-driven parallel model combination (DPMC), and unscented transform (UT). The noise parameters of the compensation models, often estimated in the maximum likelihood (ML) sense, are known to play an important role on the system performance in noisy conditions. In this paper, we conduct a systematic comparison between two popular approaches for estimating the noise parameters. The first approach employs the Gauss-Newton method in a generalized EM framework to iteratively maximizing the EM auxiliary function. The second approach views the compensation models from a generative perspective, giving rise to an EM algorithm, analogous to the ML estimation for factor analysis (EM-FA). We demonstrate a close connection between these two approaches: they belong to the family of gradient-based methods except with different convergence rates. Note that the convergence property can be crucial to the noise estimation since model compensation may be frequently carried out in changing noisy environments for retaining desired performance. Furthermore, we present an in-depth discussion on the advantages and limitations of the two approaches, and illustrate how to extend these approaches to allow for adaptive training. The investigated noise estimation approaches are evaluated on several tasks. The first is to fit a GMM model to artificially corrupted samples, and then speech recognition are performed on the Aurora 2 and Aurora 4 tasks.

Texton Noise

AbstractDesigning realistic noise patterns from scratch is hard. To solve this problem, recent contributions have proposed involved spectral analysis algorithms that enable procedural noise models to faithfully reproduce some class of textures. The aim of this paper is to propose the simplest and most efficient noise model that allows for the reproduction of any Gaussian texture. Texton noise is a simple sparse convolution noise that sums randomly scattered copies of a small bilinear texture called texton. We introduce an automatic algorithm to compute the texton associated with an input texture image that concentrates the input frequency content into the desired texton support. One of the main features of texton noise is that its evaluation only consists to sum 30 texture fetches on average. Consequently, texton noise generates Gaussian textures with an unprecedented evaluation speed for noise by example. A second main feature of texton noise is that it allows for high‐quality on‐the‐fly anisotropic filtering by simply invoking existing GPU hardware solutions for texture fetches. In addition, we demonstrate that texton noise can be applied on any surface using parameterization‐free surface noise and that it allows for noise mixing.

结合最小噪声分离变换和卷积神经网络的高分辨影像分类方法

结合最小噪声分离变换和卷积神经网络的高分辨影像分类方法

Feedback-Driven Sensory Mapping Adaptation for Robust Speech Activity Detection.

Parsing natural acoustic scenes using computational methodologies poses many challenges. Given the rich and complex nature of the acoustic environment, data mismatch between train and test conditions is a major hurdle in data-driven audio processing systems. In contrast, the brain exhibits a remarkable ability at segmenting acoustic scenes with relative ease. When tackling challenging listening conditions that are often faced in everyday life, the biological system relies on a number of principles that allow it to effortlessly parse its rich soundscape. In the current study, we leverage a key principle employed by the auditory system: its ability to adapt the neural representation of its sensory input in a high-dimensional space. We propose a framework that mimics this process in a computational model for robust speech activity detection. The system employs a 2-D Gabor filter bank whose parameters are retuned offline to improve the separability between the feature representation of speech and nonspeech sounds. This retuning process, driven by feedback from statistical models of speech and nonspeech classes, attempts to minimize the misclassification risk of mismatched data, with respect to the original statistical models. We hypothesize that this risk minimization procedure results in an emphasis of unique speech and nonspeech modulations in the high-dimensional space. We show that such an adapted system is indeed robust to other novel conditions, with a marked reduction in equal error rates for a variety of databases with additive and convolutive noise distortions. We discuss the lessons learned from biology with regard to adapting to an ever-changing acoustic environment and the impact on building truly intelligent audio processing systems.

Noise Robust Speaker Identification Using RASTA–MFCC Feature with Quadrilateral Filter Bank Structure

This paper motivates the use of Relative Spectra---Mel Frequency Cepstral Coefficients (RASTA---MFCC) feature extracted from the newly designed Quadrilateral filter bank structure and Gaussian Mixture Model---Universal Background Model (GMM---UBM) for improved text independent speaker identification under noisy environment. Unlike neural network model which requires retraining of entire database when a new sample is added to it, GMM---UBM model does not require retraining of entire database which leads to easier and faster processing. RASTA---MFCC is found to be more robust to noisy environment compared with traditional MFCC method. MFCC is an efficient feature for identifying the speaker as it has speaker specific information capturing ability. RASTA processing of speech improves the performance of recognizer in the presence of convolution and additive noise. This work combines the better of these two processes to yield RASTA---MFCC feature which is robust to noise and also proposes a new Quadrilateral filter bank structure which approximates the response of cochlear membrane of human ear to effectively capture the feature vectors. The proposed Quadrilateral filter bank structure with RASTA---MFCC feature and GMM---UBM modeling for speaker identification demonstrates supremacy over triangular and Gaussian filter banks and offers a speaker identification accuracy of 97.67 % for the MEPCO noisy speech database with 50 speakers.

Operator differential-algebraic equations with noise arising in fluid dynamics

We study linear semi-explicit stochastic operator differential algebraic equations (DAEs) for which the constraint equation is given in an explicit form. In particular, this includes the Stokes equations arising in fluid dynamics. We combine a white noise polynomial chaos expansion approach to include stochastic perturbations with deterministic regularization techniques. With this, we are able to include Gaussian noise and stochastic convolution terms as perturbations in the differential as well as in the constraint equation. By the application of the polynomial chaos expansion method, we reduce the stochastic operator DAE to an infinite system of deterministic operator DAEs for the stochastic coefficients. Since the obtained system is very sensitive to perturbations in the constraint equation, we analyze a regularized version of the system. This then allows to prove the existence and uniqueness of the solution of the initial stochastic operator DAE in a certain weighted space of stochastic processes.

NOISE IMMUNITY CODES DISPLAY IN RELIABILITY CORDINATES

Background. Noise immunity codes are widely used in the modern telecommunication systems. The methodology of properties count of convolutional codes, offered by authors before, to the equivalent parameters of block codes gave the opportunity to use the known and new methodologies of estimation of noise immunity block codes for the corresponding estimations of convolutional codes.Objective. The purpose of this paper is the estimation of maximum possibilities of block and convolutional noise immunity codes in the reliability co-ordinates on the basis of accordance of their equivalent parameters.Methods. Based on Plotkin’s and Varshamov-Gilbert’s boundaries usage the lines of theoretic error-correcting boundary of block and convolutional codes are built in reliability coordinates. The lines allow finding maximum channel probability of error for which error-correcting code exists that can ensure necessary reliability.Results. Due to search methodic of convolutional code parameters that are equivalent by correcting abilities to block code parameters, complex assessment of error-correcting, informational and forming complexity properties is made on different kinds of error-correcting codes.Conclusions. For each probability of symbol errors in a communication channel it is possible to estimate the expedience of the use of block and convolutional codes from the standpoint of the necessary reliability provision, informative efficiency and complication of encoding. Investigational limits allow defining maximally possible probability of error, at which there is a code, able to provide necessary authenticity on the output of decoder.Key words: noise immunity codes; block and convolutional codes; maximum possibilities of error-correcting; Plotkin’s andVarshamov-Gilbert’s boundaries; error of symbols; necessary reliability.

Eigennoise Speech Recovery in Adverse Environments with Joint Compensation of Additive and Convolutive Noise

The learning-based speech recovery approach using statistical spectral conversion has been used for some kind of distorted speech as alaryngeal speech and body-conducted speech (or bone-conducted speech). This approach attempts to recover clean speech (undistorted speech) from noisy speech (distorted speech) by converting the statistical models of noisy speech into that of clean speech without the prior knowledge on characteristics and distributions of noise source. Presently, this approach has still not attracted many researchers to apply in general noisy speech enhancement because of some major problems: those are the difficulties of noise adaptation and the lack of noise robust synthesizable features in different noisy environments. In this paper, we adopted the methods of state-of-the-art voice conversions and speaker adaptation in speech recognition to the proposed speech recovery approach applied in different kinds of noisy environment, especially in adverse environments with joint compensation of additive and convolutive noises. We proposed to use the decorrelated wavelet packet coefficients as a low-dimensional robust synthesizable feature under noisy environments. We also proposed a noise adaptation for speech recovery with the eigennoise similar to the eigenvoice in voice conversion. The experimental results showed that the proposed approach highly outperformed traditional nonlearning-based approaches.

Ideal binary masking for reducing convolutive noise

It is important to know the degree to which convolutive noise disrupts the perceptual aspects of speech and its intelligibility. This paper presents the ideal binary masking criterion for reducing the convolutive noise (reverberation) and to improve the quality and intelligibility of speech. The noise is suppressed using ideal binary time---frequency masking that is based on signal-to-reverberation ratio (SRR) of individual time---frequency channels. All T---F channels with the SRR greater than pre-selected threshold are retained while others are eliminated. The performance of algorithm is evaluated using IEEE sentences corrupted with different degrees of reverberation times (RT60) ranging from 0.3 to 2.0 s. The results indicate that with the increase of reverberation time, the intelligibility and perceptual aspects of speech decrease. Additional analyses indicated that ideal binary masking reduced the temporary envelope spreading effect introduced by the reverberation. The algorithm is evaluated with perceptual evaluation of speech quality, SNRLOSS, log-likelihood-ratio and frequency weighted segmental signal-to-noise ratio.

Advanced parallel combined Gaussian mixture model based feature compensation integrated with iterative channel estimation

This study proposes an effective feature compensation scheme to address severely adverse environments for speech recognition where background noise and channel distortion are simultaneously involved. In the proposed scheme, an iterative channel estimation method is integrated into the framework of our previously proposed Parallel Combined Gaussian Mixture Model (PCGMM) based feature compensation algorithm. A new speech corpus is developed which reflects both additive and convolutional noise corruption. The channel distortion effects are obtained from the NTIMIT and CTIMIT corpora. Evaluation based on objective measures including STNR, PESQ, and speech recognition shows that generated speech corpus includes highly challenging acoustic conditions for speech recognition. The proposed feature compensation method is evaluated over the developed speech corpus. The experimental results demonstrate that the proposed feature compensation scheme is effective at improving speech recognition performance in the presence of both background noise and channel distortion, employing the iterative channel estimation method. The proposed PCGMM-based feature compensation scheme employing the channel estimation method shows +3.58% and +11.61% relative improvements in averaged WER compared to the ETSI AFE algorithm for the developed speech corpora including NTIMIT and CTIMIT channel effects respectively. For real-life application, a voice activity detection technique is employed to estimate the noise model for PCGMM-based method without a priori knowledge of the non-speech locations of input speech. The proposed method is also evaluated on the CU-Move corpus which represents actual in-vehicle conditions, showing a +12.99% relative improvement compared to the ETSI AFE. This study confirms that the proposed PCGMM-based feature compensation method integrated with channel estimation is effective at increasing speech recognition accuracy in real-life severely adverse conditions.

Review: Audio Noise Reduction Using Filters and Discrete Wavelet Transformation

Audio noise reduction using filters and Discrete Wavelet Transformation” our applications include noise propagation problem in industrial air handling systems, noise in aircrafts and tonal noise from electric power, as well as isolation of vibration from which noise is one kind of sound that is unexpected or undesired . The noise related problem can be divided into non-additive noise and additive noise. The non-additive noise includes multiplier noise and convolution noise, which can be transformed into additive noise throughhomomorphism transform. The additive noise includes periodical noise, pulse noise, and broadband noise related problems. There are many kinds of broadband noise, which may include heat noise, wind noise, quantization noise, and all kinds of random noise such as white noise and pink noise. In acoustics applications, noise from the surrounding environment severely reduces the quality ofspeech and audio signals. Therefore, basic linear are used to denoise the audio signals and enhance speech and audio signal quality. Our main objective is to reduce noise from system which is heavily dependent on the specific context and application. As, we want to increase the intelligibility or improve the overall speech perception quality. Such as SNR, PSNR, MSE and the Time to reduce the noise for noisy signals for removing noise.

A Maximum Entropy inspired model for the convolutional noise PDF

In this paper we consider a blind adaptive deconvolution problem in which we observe the output of an unknown linear system (channel) from which we want to recover its input using an adaptive blind equalizer (adaptive linear filter). Since the channel coefficients are unknown, the optimal equalizer's coefficients are also unknown. Thus, the equalizer's coefficients used in the deconvolution process are only approximated values leading to an error signal in addition to the source signal at the output of the deconvolutional process. We define this error signal throughout the paper as the convolutional noise. It is well known that the convolutional noise probability density function (pdf) is not a Gaussian pdf at the early stages of the deconvolutional process and only at the latter stages of the deconvolutional process the convolutional noise pdf tends to be approximately Gaussian. Despite this knowledge, the convolutional noise pdf was modeled up to recently as a Gaussian pdf because it simplifies the Bayesian calculations when carrying out the conditional expectation of the source input given the equalized or deconvolutional output and since no other model was suggested for it. Recently, a new model was suggested by the same author for the convolutional noise pdf based on the Edgeworth expansion series. This new model leads to improved deconvolution performance for the 16 Quadrature Amplitude Modulation (QAM) input and for a signal to noise ratio (SNR) of 30 dB. Thus, the question that arose here was whether we may find another model for the convolutional noise pdf that will also lead the system with improved deconvolutional performance compared to the case when the Gaussian model is applied for the convolutional noise pdf. In this paper, we propose a new model for the convolutional noise pdf inspired by the Maximum Entropy density approximation technique. We derive the relevant Lagrange multipliers and obtain as a by-product new closed-form approximated expressions for the conditional expectation and mean square error (MSE). Simulation results indicate that improved system performance is obtained from the residual ISI point of view for the 16QAM input case with our new proposed model for the convolutional noise pdf compared to the case when the Gaussian model or Edgeworth expansion series are applied for the convolutional noise pdf. For two other chosen input sources, a faster convergence rate is observed with the algorithm using our new proposed model for the convolutional noise pdf compared to the Maximum Entropy and Godard's algorithm.

Convolutional Noise Analysis via Large Deviation Technique

Due to non-ideal coefficients of the adaptive equalizer used in the system, a convolutional noise arises at the output of the deconvolutional process in addition to the source input. A higher convolutional noise may make the recovering process of the source signal more difficult or in other cases even impossible. In this paper we deal with the fluctuations of the arithmetic average (sample mean) of the real part of consecutive convolutional noises which deviate from the mean of order higher than the typical fluctuations. Typical fluctuations are those fluctuations that fluctuate near the mean, while the other fluctuations that deviate from the mean of order higher than the typical ones are considered as rare events. Via the large deviation theory, we obtain a closed-form approximated expression for the amount of deviation from the mean of those fluctuations considered as rare events as a function of the system’s parameters (step-size parameter, equalizer’s tap length, SNR, input signal statistics, characteristics of the chosen equalizer and channel power), for a pre-given probability that these events may occur.

Quantum white noise convolution operators with application to differential equations

Abstract In this paper we introduce a quantum white noise (QWN) convolution calculus over a nuclear algebra of operators. We use this calculus to discuss new solutions of some linear and non-linear differential equations.

A review on speech enhancement algorithms and why to combine with environment classification

Speech enhancement has been an intensive research for several decades to enhance the noisy speech that is corrupted by additive noise, multiplicative noise or convolutional noise. Even after decades of research it is still the most challenging problem, because most papers rely on estimating the noise during the nonspeech activity assuming that the background noise is uncorrelated (statistically independent of speech signal), nonstationary and slowly varying, so that the noise characteristics estimated in the absence of speech can be used subsequently in the presence of speech, whereas in a real time environment such assumptions do not hold for all the time. In this paper, we discuss the historical development of approaches that starts from the year 1970 to, the recent, 2013 for enhancing the noisy speech corrupted by additive background noise. Seeing the history, there are algorithms that enhance the noisy speech very well as long as a specific application is concerned such as the In-car noisy environments. It has to be observed that a speech enhancement algorithm performs well with a good estimation of the noise Power Spectral Density (PSD) from the noisy speech. Our idea pops up based on this observation, for online speech enhancement (i.e. in a real time environment) such as mobile phone applications, instead of estimating the noise from the noisy speech alone, the system should be able to monitor an environment continuously and classify it. Based on the current environment of the user, the system should adapt the algorithm (i.e. enhancement or estimation algorithm) for the current environment to enhance the noisy speech.

Joint Mixing Vector and Binaural Model Based Stereo Source Separation

In this paper the mixing vector (MV) in the statistical mixing model is compared to the binaural cues represented by interaural level and phase differences (ILD and IPD). It is shown that the MV distributions are quite distinct while binaural models overlap when the sources are close to each other. On the other hand, the binaural cues are more robust to high reverberation than MV models. According to this complementary behavior we introduce a new robust algorithm for stereo speech separation which considers both additive and convolutive noise signals to model the MV and binaural cues in parallel and estimate probabilistic time-frequency masks. The contribution of each cue to the final decision is also adjusted by weighting the log-likelihoods of the cues empirically. Furthermore, the permutation problem of the frequency domain blind source separation (BSS) is addressed by initializing the MVs based on binaural cues. Experiments are performed systematically on determined and underdetermined speech mixtures in five rooms with various acoustic properties including anechoic, highly reverberant, and spatially-diffuse noise conditions. The results in terms of signal-to-distortion-ratio (SDR) confirm the benefits of integrating the MV and binaural cues, as compared with two state-of-the-art baseline algorithms which only use MV or the binaural cues.

An iterative longest matching segment approach to speech enhancement with additive noise and channel distortion

This paper presents a new approach to speech enhancement from single-channel measurements involving both noise and channel distortion (i.e., convolutional noise), and demonstrates its applications for robust speech recognition and for improving noisy speech quality. The approach is based on finding longest matching segments (LMS) from a corpus of clean, wideband speech. The approach adds three novel developments to our previous LMS research. First, we address the problem of channel distortion as well as additive noise. Second, we present an improved method for modeling noise for speech estimation. Third, we present an iterative algorithm which updates the noise and channel estimates of the corpus data model. In experiments using speech recognition as a test with the Aurora 4 database, the use of our enhancement approach as a preprocessor for feature extraction significantly improved the performance of a baseline recognition system. In another comparison against conventional enhancement algorithms, both the PESQ and the segmental SNR ratings of the LMS algorithm were superior to the other methods for noisy speech enhancement.

Feature enhancement by deep LSTM networks for ASR in reverberant multisource environments

This article investigates speech feature enhancement based on deep bidirectional recurrent neural networks. The Long Short-Term Memory (LSTM) architecture is used to exploit a self-learnt amount of temporal context in learning the correspondences of noisy and reverberant with undistorted speech features. The resulting networks are applied to feature enhancement in the context of the 2013 2nd Computational Hearing in Multisource Environments (CHiME) Challenge track 2 task, which consists of the Wall Street Journal (WSJ-0) corpus distorted by highly non-stationary, convolutive noise. In extensive test runs, different feature front-ends, network training targets, and network topologies are evaluated in terms of frame-wise regression error and speech recognition performance. Furthermore, we consider gradually refined speech recognition back-ends from baseline ‘out-of-the-box’ clean models to discriminatively trained multi-condition models adapted to the enhanced features. In the result, deep bidirectional LSTM networks processing log Mel filterbank outputs deliver best results with clean models, reaching down to 42% word error rate (WER) at signal-to-noise ratios ranging from −6 to 9dB (multi-condition CHiME Challenge baseline: 55% WER). Discriminative training of the back-end using LSTM enhanced features is shown to further decrease WER to 22%. To our knowledge, this is the best result reported for the 2nd CHiME Challenge WSJ-0 task yet.

Edgeworth Expansion Based Model for the Convolutional Noise pdf

Recently, the Edgeworth expansion up to order 4 was used to represent the convolutional noise probability density function (pdf) in the conditional expectation calculations where the source pdf was modeled with the maximum entropy density approximation technique. However, the applied Lagrange multipliers were not the appropriate ones for the chosen model for the convolutional noise pdf. In this paper we use the Edgeworth expansion up to order 4 and up to order 6 to model the convolutional noise pdf. We derive the appropriate Lagrange multipliers, thus obtaining new closed-form approximated expressions for the conditional expectation and mean square error (MSE) as a byproduct. Simulation results indicate hardly any equalization improvement with Edgeworth expansion up to order 4 when using optimal Lagrange multipliers over a nonoptimal set. In addition, there is no justification for using the Edgeworth expansion up to order 6 over the Edgeworth expansion up to order 4 for the 16QAM and easy channel case. However, Edgeworth expansion up to order 6 leads to improved equalization performance compared to the Edgeworth expansion up to order 4 for the 16QAM and hard channel case as well as for the case where the 64QAM is sent via an easy channel.

Residual ISI Obtained by Blind Adaptive Equalizers and Fractional Noise

Recently, a closed-form approximated expression was derived by the same author for the achievable residual intersymbol interference (ISI) case that depends on the step-size parameter, equalizer’s tap length, input signal statistics, signal-to-noise ratio (SNR), and channel power. But this expression was obtained by assuming that the input noise is a white Gaussian process where the Hurst exponent (H) is equal to 0.5. In this paper, we derive a closed-form approximated expression (or an upper limit) for the residual ISI obtained by blind adaptive equalizers valid for fractional Gaussian noise (fGn) input where the Hurst exponent is in the region of0.5≤H<1. Up to now, the statistical behaviour of the residual ISI was not investigated. Furthermore, the convolutional noise for the latter stages of the deconvolutional process was assumed to be a white Gaussian process (H=0.5). In this paper, we show that the Hurst exponent of the residual ISI is close to one, almost independent of the SNR or equalizer’s tap length but depends on the step-size parameter. In addition, the convolutional noise obtained in the steady state is a noise process having a Hurst exponent depending on the step-size parameter.