Example Phonem Recognition

A possible application of SECoS is given in Watts/Kasabov (2002)[343](cf. figure 8.11). Given are data with spoken phonems. The raw speech signal is transformed according to the mel-scale. From one sample they extract 26 features^8.3. Then they combine three consecutive feature vectors into one input vector with $3 \times 26 = 78$ values (cf. [343]:p.235).

Figure 8.11: An application of SECoS for phonem recognition Watts/Kasabov (2000)