<< Chapter < Page Chapter >> Page >

Results for the deconvolution method

Comparison of three methods of finding impulse response

Each frequency response is taken in the OEDK classroom.

Observations from magnitude of the frequency response

  • The pseudo-dirac's frequency response is centered at higher frequencies that the magnitude of the responses given by the sine-sweep and the balloon pop.
  • The sine-sweep has the most defined peak at low frequencies: lower frequency sounds tend to reverberate longer in closed spaces.
  • The balloon pop has a sinc-like decaying envelope which could simply mean there is some noise involved.

Frequency response of acoustically desirable room

Interestingly, the phase response of the acoustically desirable classroom is very similar to the echoic OEDK classroom.

Comparing sine-sweep impulse responses

Compared to the acoustically desirable room, the OEDK classroom filtered out more high frequencies.

Deconvolution result

Devoncolution may have filtered out echoes, but the output of the deconvolution is so noisy that it is unintelligible even when noise is subtracted/filtered out.

Result of deconvolution on echoed male speech recorded in OEDK classroom.

Results for the adaptive filtering method

Sound files recorded in each room (one desired signal and two different input signals for each algorithm).
Signals before filtering
Signal Acoustically Desirable Room (Hanszen) Echoic Room (Wiess) OEDK Classroom
Impulse
Female Speech
Male Speech

For each of the three types of signals, an adaptive filter was trained with the recorded desirable signal (in Table 1 above). Then, for each algorithm, the filter was used to clean the echoed signal recorded in one of the two classrooms (echoic and OEDK). The resulting signals can be found below:

Echoic Room (Wiess) adaptive filter output sound files.
Signal LMS NLMS BLMS
Impulse
Female Speech
Male Speech
OEDK adaptive filter output sound files.
Signal LMS NLMS BLMS
Impulse
Female Speech
Male Speech

    Signal lengths

  • Impulse: .10 seconds
  • Male Speech: 5.0 seconds
  • Female Speech: 11 seconds

Rms values of the error signal

Error signal RMS for each algorithm implementation in the echoic classroom.
Echoic classroom
Signal LMS NLMS BLMS
Impulse .0240 .0182 .0490
Female Speech .0705 .0385 .0133
Male Speech .0235 .0133 .0564
Error signal RMS for each algorithm implementation in the OEDK classroom.
Oedk classroom
Signal LMS NLMS BLMS
Impulse .0253 .0166 .0480
Female Speech .0560 .0373 .1141
Male Speech .0223 .0117 .0590
Average error signal RMS for each algorithm implementation.
LMS NLMS BLMS
Average RMS Value .0369 .0752 .0226

Since the purpose of the adaptive filter is to minimize the mean square error, the root mean square of the signal can be used to evaluate the effectiveness of each algorithm on the signals. Table 1 and Table 2 show the RMS values of the error signal when the corresponding algorithm has been used to train an adaptive filter with their corresponding desired signals (impulse to echoed impulse, male to echoed male, etc.)

Oedk male speech echo cancellation with lms algorithm

Oedk male speech echo cancellation with nlms algorithm

Oedk male speech echo cancellation with blms algorithm

Echo cancellation of male speech in the OEDK classroom using three different adaptive filter algorithms.

Visually, it is apparent that the NLMS error signal's amplitude is smaller. This implies that the NLMS's output is closest to the desired signal.

Get Jobilize Job Search Mobile App in your pocket Now!

Get it on Google Play Download on the App Store Now




Source:  OpenStax, Characterization and application of echo cancellation methods. OpenStax CNX. Dec 17, 2012 Download for free at http://cnx.org/content/col11468/1.3
Google Play and the Google Play logo are trademarks of Google Inc.

Notification Switch

Would you like to follow the 'Characterization and application of echo cancellation methods' conversation and receive update notifications?

Ask