Laboratory of Biocybernetics

Processing and Recognition of Spoken Polish
 

• processing spoken Polish using neural networks - noise reduction with an MLP network, selection of sets of features describing the speech signal, processing pathological speech [6] (Wszo'ek, 1994 a) (Izworski, 1995 b),
• recognition of isolated words from a limited dictionary using nonlinear time-normalization (Dynamic Time Warping) and Hidden Markov Model [6] (Izworski, 1993),
• using MLP, LVQ and RBF neural networks in recognition of isolated words from a limited dictionary [6] (Izworski, 1995 a) (Izworski, 1995 b),
• recognition of Polish phonemes using time-delay neural networks [6][9] (Izworski, 1994 b) (Izworski, 1995 a) (see Fig.9),

  Fig. 8. TDNN neural network recognizing phonemes a) construction of the network b) rectangle of connections c) connections represented by the rectangle of connections