Nonlinear audio system identification generally relies on Gaussianity, whiteness and stationarity hypothesis on the input signal, although audio signals are non-Gaussian, highly correlated and non-stationary. However, since the physical behavior of nonlinear audio systems is input-dependent, they should be identified using natural audio signals (speech or music) as input, instead of artificial signals (sweeps or noise) as usually done. We propose an identification scheme that conditions audio signals to fit the desired properties for an efficient identification. The identification system consists in (1) a Gaussianization step that makes the signal near-Gaussian under a perceptual constraint; (2) a predictor filterbank that whitens the signal; (3) an orthonormalization step that enhances the statistical properties of the input vector of the last step, under a Gaussianity hypothesis; (4) an adaptive nonlinear model. The proposed scheme enhances the convergence rate of the identification and reduces the steady state identification error, compared to other schemes, for example the classical adaptive nonlinear identification.

/pdf/nonlinear-audio-systems-identification-through-audio-input-23q61bl740.pdf

Nonlinear Audio Systems Identification Through Audio Input Gaussianization

Identification and compensation purposes of nonlinear systems are of interest for many audio processing applications. The analysis of systems under test must be done through realistic audio inputs in order to capture different aspects of the nonlinearity. However, the Gaussianity of the tested signal, is a desirable factor because it guarantees easy implementation and good performances for the nonlinearity identification process. In this paper, we show at a first stage, the importance of input Gaussianity for the identification of memoryless nonlinear systems. At a second stage, we propose an algorithm that makes the speech signals Gaussian. The proposed “Gaussianization” algorithm is based on the embedding of an imperceptible signal in the speech signal, to force it to be Gaussian. As expected, the performances of the optimal identification of a polynomial nonlinearity are much better with the Gaussianized input than with the original one. Moreover, these performances exhibit a robustness similar to the Gaussian input case.

/pdf/gaussianization-method-for-identification-of-memoryless-1rhjogk527.pdf

”Gaussianization” Method for identification of memoryless nonlinear audio systems

In this paper, a new digital audio watermarking scheme is discussed to increase the information transmitted in a communication context [7]. It is based on a new approach to embed/extract the watermark by combining the Empirical Mode Decomposition (EMD) [4] and the Blind Source Separation (BSS) based on the Independent Subspaces Analysis (ISA) method [2]. In order to increase the embedded binary data flow, the Intrinsic Mode Functions (IMFs) of the audio host signal are watermarked using a specific mixture that ensures the inaudibility of the watermark. The separation between the IMFs and the watermarks is performed by a new proposed under-determined ISA method, (UISA). The performances and the robustness of the proposed watermarking system are evaluated with different types of audio signals. The obtained results permit to assess the achieved efficiency in terms of bit error rate (BER).

Hybrid embedding strategy for a blind audio watermarking system using EMD and ISA techniques

Method of cancelling acoustic echo in a first signal comprising an echo signal of a second signal, comprising; inserting, in an inaudible manner, into the second signal a pseudo-random sequence whose circular autocorrelation comprises a unit impulse and a continuous component; characterizing, in the first signal, by means of the inserted sequence, an acoustic channel followed by the echo signal; estimating the echo signal in the first signal by means of the characterization of the acoustic channel; and cancelling the echo signal by means of the estimation obtained.

Method and device for cancelling acoustic echo by audio watermarking

In this paper, we propose the concept of “doping watermarking”, whose principle is to add an imperceptible noise to an host signal in order to improve its properties. Especially, our aim is to reduce the spectral support of the probability density function (PDF) of an audio signal in order to match the conditions of the quantization theorem. In this context, we develop a specific audiowatermarking algorithm and test its performance on real audio signals. This watermark allows to recover the PDF of a digital signal from a sub-quantized version of the signal, with very low error.

Revisiting quantization theorem through audiowatermarking

Audio watermarking, or embedding information in a host signal was originally used for digital copyright protection purposes. As audio coding, watermarking is progressively brought in audio processing applications.

Speech processing in the watermarked domain: Application in adaptive Acoustic Echo Cancellation

The non-stationary RLS algorithm is used for decision feedback equalization of time varying channels. We consider Markovian time variations of the impulse response of the channel as in radio mobile communications. The non-stationary RLS algorithm has the capacity to learn in real time about time varying model of the channel, so it presents a better tracking capacity than the classical ones. Hence, it is able to improve the bit error rate especially for severe propagation conditions.

Blind decision feedback equalizer with tracking ability in severe propagation conditions

Qualitative and quantitative performances of adaptive filters can significantly vary when they are implemented in finite precision arithmetic. Analyzing theoretically such performances is very complex. In this paper, we present a new approach which simplifies both transient and steady state analysis of implemented adaptive filters. Indeed, the finite precision adaptive identification scheme in a stationary context is replaced by an equivalent simple infinite precision scheme in a specific non stationary context; where the channel to be identified has stochastic time-variations.

M. Turki-Hadj Alouane

Papers

”Gaussianization” Method for identification of memoryless nonlinear audio systems

Speech processing in the watermarked domain: Application in adaptive Acoustic Echo Cancellation

Blind decision feedback equalizer with tracking ability in severe propagation conditions

A non stationary/infinite precision system analogy for fixed-point digital adaptive filter analysis