Digital Signal Processing Reference
In-Depth Information
Figure 8-6.
Performance of ASE-MCMAC and LMS under Different Noise Regimes.
At low noise (high SNR) cases, the improvement over the benchmark is
insignificant, i.e., any technique would work as expected.
However, at high noise (low SNR) regimes, the case for a vehicle moving
in heavy traffic, the LMS algorithm clearly fails. On the other hand, the
ASE-MCMAC algorithm has performed remarkably well. We would like to
note that this conclusion is valid as long as the SNR of the noise reference
input into the microphone
is less than that of the primary channel
microphone
The advantage of the ASE-MCMAC over the classical ASE approach is
that once the noise spectrum has been trained only a recall process is needed
for processing incoming speech. Even though fairly encouraging results were
obtained in our earlier work based on a CMAC, the performance of the ASE
based on this modified version (MCMAC-ASE) is remarkably superior to its
predecessor. Response from quite a number of experienced listeners has been
uniformly the same.
In addition during training, the corrupt speech is also processed as well.
Recalling the fact that only a subset of the overall memory of the CMAC
