21-06-2012, 04:09 PM
Echo Cancellation using Cepstral Technique
Introduction
Echo canceller is a vital part of any digital
synthesizer In order to be feasible, small chunks of data
(audio) is processed together, thus making it real time. There
will be a definite amount of delay in between input and
output. The technique works satisfactorily for lower
frequency and larger delay in echo. It is designed to be used
primarily with DSP processor or in conjunction with host
based applications, such as VoIP, pulse detection, speaker
verification and audio conferencing. It can be used for
network echo canceller, but is not optimized for such
applications as it will introduce a delay.
Principal
Voiced speech can be represented on a short-time
basis as the convolution of a pulse with a periodic impulse
train. The physical model is replete with signals that can be
represented by very similar models. For example, in many
areas of physical measurement and communication, signals
are transmitted or recorded in what may broadly be termed
as reverberant environment.
In this case, a signal is represented as the sum of
delayed replica, or echo, or some basic waveform. In case
where reverberation is viewed as distortion, we may wish to
recover the basic waveform.
Practical Implementation
Following is an example of this technique
implemented for a wave audio format of 11.025Khz
sampling frequency and 8-bits per sample. First we have a
wave file which has a component of echo in it. The first
figure below shows the original audio signal. We now
degrade this signal but adding a time delayed replica of the
same. Figure 2 shows the same.
Applications
The echo canceller using cepstrum technique finds
its advantages in applications like audio recording,
conference telephony, radar and sonar detection, seismic
measurements and electrophysiology.
Future Scope
In this application, we have only considered first
order delay. For good echo cancellation, we need to consider
higher order terms as well, that will make system more
complex.
Also, it is very difficult to model the echo
phenomenon in practice. Given the echoed signal, the
parameters and T are difficult to estimate in real time.