Adaptive beamforming and Noise cancellation

The paramount reason for going in for a scheme which has its basis on adaptive filtering and uses multiple microphones was the need for complete mobility. When a person is away from the microphone the voice captured by a stand alone microphone is extremely poor. This is because of the effect of reverberations due of enclosed rooms, coherent/incoherent noise e.g. Babble noise, fans, machines. All these eventually lead to a degradation in speech recognition performance. To improve on the same the following scheme was sought.

The idea is to use a pair of microphones for hands free speech recognition in noisy and reverberant environment. As microphones pick up everything around them, the aim is to increase their sensitivity in one direction i.e. the direction in which the speech signal is strongest and attenuate the noise coming in from the other directions. For achieving such a system, adaptive beamforming and noise cancellation techniques will be used using normalized LMS algorithm. Normally off the shelf products use around 8 to 10 microphones but my goal was to draw inspiration from biology, thus only two microphones will be used in accordance with two ears that biology provides us. These will be placed in the room place 30 cm apart from each other. The output of both these microphones will be used as inputs to the algorithm. In addition to this, adaptive noise cancellation will be used to reduce the noise captured via the microphone to have even better performance in a cocktail party like environment. The processed output of the beamformer and adaptive noise canceller will be used as an input to an HMM to check for the recognition accuracy.

Initially the coding will be done in matlab but the final goal will be to implement this algorithm in Java so that it becomes platform independent. Continuing on my work done for EEL 6825 on smart appliances, I am planning of recognizing a vocabulary of four words, namely

The links provided below will give a thorough and detailed description of the project and shall help you get the complete picture.

This is a power point presentation which will help you get an overview of the project. For details of the project, links above will be of help.

The main area of research will be to enhance the beamformer performance to achieve even higher accuracy rates under varying noise conditions. Also real time implementation of the same is a highly sought after goal.

Papers:

M. Omologo, "Environmental conditions and acoustic transduction in hands free speech recognition" Speech comm, 1998

M.Joho, S. Moschytz , "On the design of the target-signal filter in adaptive beamforming" IEEE trans, July 1999

B. D. Van, "Beamforming: A versatile approach to spatial filtering", IEEE ASSP Magazine, Apr 1988

S. Weiss, "An effecient scheme for broadband adaptive beamforming", IEEE trans, 1999

M. Zhang, M. H. Er, "Adaptive beamforming by microphone arrays" IEEE trans, 1995

M.W. Hoffman, "Desired-speech signal cancellation by microphone arrays in reverberant rooms" IEEE trans, 1991

K. Oteri " Adaptive antenna approach for GPS multipath mitigation" Standford university, Jun 2001

M. Godavarti, "Stochastic partial update LMS algorithm for adaptive arrays"

V. Ramasami,"Self orthogonalizing algorithms for adaptive beamformers"

J.V. Berghe, "An adaptive noise canceller for hearing aids using two nearby microphones", IEEE trans, 1998

M. Dahl, "Acoustic noise and echo cancellation with microphone arrays", IEEE trans, 1999

L. Godara, "Improved LMS algorithm for adaptive beamforming", IEEE trans, 1990

Code:

Complete source code for this project

Hidden Markov Model code, http://www-sig.enst.fr/~cappe/h2m/h2m.html

Source code:

All the Matlab code being propriety of the Department of Electrical and Computer Eng. and the Computational Neuro Engineering lab, is available on request.

Kaustubh R. Kale
Research assistant: Department of ECE
telephone: 352 392 2626
Mail: kkale@cnel.ufl.edu

Back to home page