Man Mohan Sondhi

Advances in Network and Acoustic Echo Cancellation

This book brings together many advanced topics in network and acoustic echo cancellation aimed towards enhancing the echo cancellation performance of next-generation telecommunication systems. The resulting compendium provides a coherent treatment of such topics not found otherwise in journals or other books.

Maximum likelihood estimation for multivariate mixture observations of markov chains (Corresp.)

To use probabilistic functions of a Markov chain to model certain parameterizations of the speech signal, we extend an estimation technique of Liporace to the eases of multivariate mixtures, such as Gaussian sums, and products of mixtures. We also show how these problems relate to Liporaces original framework.

Cochlear macromechanics: time domain solutions.

In this paper we report on a new method of solving a previous derived, two-dimensional model, integral equation for basilar membrane (BM) motion. The method uses a recursive algorithm for the solution of an initial-value problem in the time domain, combined with a fast Fourier transform (FFT) convolution in the space domain at each time step. Thus, the method capitalizes on the high speed and accuracy of the FFT yet allows the BM to have nonlinear mechanical properties. Using the new method we compute (linear) solutions for various choices of model parameters and compare the results to the experimental measurements of Rhode. [J. Acoust. Soc. Am. 49, 1218-1231 (1971)]. We also demonstrate the effect of including longitudinal stiffness along the BM and conclude that it is useful in matching the high-frequency slope as measured by Rhode.

Recent developments in the application of hidden Markov models to speaker-independent isolated word recognition

In this paper we extend previous work on isolated word recognition based on hidden Markov models by replacing the discrete symbol representation of the speech signal by a continuous Gaussian mixture density. In this manner the inherent quantization error introduced by the discrete representation is essentially eliminated. The resulting recognizer was tested on a vocabulary of the 10 digits across a wide range of talkers and test conditions, and shown to have an error rate at least comparable to that of the best template recognizers and significantly lower than that of the discrete symbol hidden Markov model system. Several issues involved in the training of the continuous density models and in the implementation of the recognizer are discussed.

Stereophonic acoustic echo cancellation using nonlinear transformations and comb filtering

Stereophonic sound becomes more and more important in a growing number of applications (such as teleconferencing, multimedia workstations, televideo gaming, etc.) where spatial realism is demanded. Such hands-free systems need stereophonic acoustic echo cancelers (AECs) to reduce echos that result from coupling between loudspeakers and microphones in full-duplex communication. We propose a new stereo AEC based on two experimental observations: (a) the stereo effect is due mostly to sound energy below about 1 kHz and (b) comb filtering above 1 kHz does not degrade auditory localization. The principle of the proposed structure is to use one stereo AEC at low frequencies (e.g. below 1 kHz) with nonlinear transformations on the input signals and another stereo AEC at higher frequencies (e.g. above 1 kHz) with complementary comb filters on the input signals.

Acoustic beam forming with robust signal estimation

Audio signals from any array of microphones are individually filtered, delayed, and scaled in order to form an acoustic beam that focuses the array on a particular region. Nonlinear robust signal estimation processing is applied to the resulting set of audio signals to generate an output signal for the array. The nonlinear robust signal estimation processing may involve dropping or otherwise reducing the magnitude of one or more of the highest and lowest data in each set of values from the resulting audio signals and then selecting the median from or generating an average of the remaining values to produce a representative, central value for the output audio signal. The nonlinear robust signal estimation processing effectively discriminates against noise originating at an unknown location outside of the focal region of the acoustic beam.

General Derivation of Frequency-Domain Adaptive Filtering

Adaptive filters [60] play an important role in echo cancellation because we need to identify and track unknown and time-varying channels [24J . There are roughly two classes of adaptive algorithms. One class includes filters that are updated in the time domain, sample-by-sample in general, like the classical least mean square (LMS) [134] and recursive least-squares (RLS) [4], [66] algorithms. The other class contains filters that are updated in the frequency domain, block-by-block in general, using the fast Fourier transform (FFT) as an intermediary step. As a result of this block processing, the arithmetic complexity of the algorithms in the latter category is significantly reduced compared to time-domain adaptive algorithms. Use of the FFT is appropriate to the Toeplitz structure, which results from the time-shift properties of the filter input signal. Consequently, deriving a frequency-domain (FD) adaptive algorithm is just a matter of rewriting the time-domain error criterion in a way that Toeplitz and circulant matrices are explicitly shown.

Echo Cancellation for Speech Signals

The sections in this article are 1 Line Echoes 2 Adaptive Cancellation 3 Single-Channel Acoustic Echo Cancellation 4 Multichannel Acoustic Echo Cancellation 5 Concluding Remarks 6 Acknowledgments

Multichannel Acoustic Echo Cancellation

One may ask a legitimate question: why do we need multichannel sound for telecommunication? Let’s take the following example. When we are in a room with several people talking, laughing, or just communicating with each other, thanks to our binaural auditory system, we can concentrate on one particular talker (if several persons are talking at the same time), localize or identify a person who is talking, and somehow we are able to process a noisy or a reverberant speech signal in order to make it intelligible. On the other hand, with only one ear or, equivalently, if we record what happens in the room with one microphone and listen to this monophonic signal, it will likely make all of the above mentioned tasks more difficult. So, multichannel sound teleconferencing systems provide a realistic presence that mono-channel systems cannot offer.

A Family of Robust PNLMS-Like Algorithms for Network Echo Cancellation

In this chapter, we present a family of fast-converging algorithms that are extensions of the proportionate normalized least mean square (PNLMS) algorithm introduced by Duttweiler [38], [54] This new family of algorithms is based on the affine projection algorithm/normalized least mean square (APA/NLMS) algorithm family [106], [100], [55], [125], [66]. What differentiates the new algorithms from the NLMS and APA algorithms is that they inherit the proportional step-size idea from the PNLMS algorithm, i.e., individually assigned step-sizes to each filter coefficient, where the step sizes are calculated from the previous estimate of the echo path. Because of these individual step sizes, the algorithms achieve a higher convergence rate by using the fact that the active part of a network echo path is usually much smaller (4–8 ms) than the possible echo path range (64–128 ms) that has to be covered. A natural extension of the basic PNLMS algorithm is a proportionate affine projection algorithm (PAPA) [51]. This algorithm (family) combines the fast converging APA with the proportional step size technique of PNLMS.