Digital Signal Processing Reference
In-Depth Information
were trained while for the MCMAC case, almost all of the memory locations
in a fairly large neighborhood were trained during the learning cycle.
To gain more inside about the implications of the remarkable results from
MCMAC experiments, we plot the Wiener filter profiles for both the
classical CMAC and the current MCMAC algorithms in Figures 8-7.a and
8-7.b, respectively. Thus, no further training is required for a fully trained
MCMAC memory unless the profile of the Weiner filters changes drastically.
The MCMAC memory once fully trained can be used without any further
training. Multiple of this MCMAC memory may be required from different
training noise environment to form a complete system.
Nevertheless we would like to caution the reader that the ultimate
judgment would come from the formal MOS -based evaluations performed
under controlled stimuli. We believe the success of the proposed method
could be attributed to the fact that only a subset of the CMAC memory was
trained. For the case of MCMAC, however, almost all of the memory
locations in a fairly large neighborhood have been trained during the learning
cycle.
5.
CONCLUSIONS
In this work we have presented a modified cerebellar model articulation
controller (MCMAC) and its deployment in the vehicular environment as an
integral part of the speech enhancement system. Even though, the
performance of the ASE-MCMAC algorithm is comparable to that of a
traditional LMS-based noise cancellation algorithm high SNR values greater
than -2dB, but for low SNR (high noise) regimes the proposed ASE-
MCMAC performs remarkably well and it has been shown to be very robust
in different type of noise in vehicles including the engine noise, road noise,
tire tractions, the wind and rain.
Provided sufficient computing power is available multiple MCMAC
memories can be employed to get additional gains as it is recently done in
microphone array-based implementations. In a recent report, this idea has
been explored with
encouraging
results
within the framework of
environmental sniffing [8].
