ROBUST ASR INSIDE A VEHICLE USING BLIND PROBABILISTIC BASED UNDER-DETERMINED CONVOLUTIVE MIXTURE SEPARATION TECHNIQUE - DSP for In-Vehicle and Mobile Systems - page 289

and 18-4, respectively. From this, it can be seen that our algorithm separated
all four speech signals fairly well. The speech recognizer that we have used
in this study is a segment based continuous speech recognizer with a
vocabulary size of 3000 words. Each segment and segment boundaries are
modelled using Gaussian mixture model (GMM). Duration models were also
used.
Figure 18-2. The block diagram of the proposed algorithm for BSS of convolutive mixture in
the underdetermined case.
Figure 18-3. Mixed speech signals from two channels.