Word recognition
Automatic speech recognition

As already mentioned, the absence of phoneme and word boundary indicators in the speech signal is a major difficulty in speech recognition. As a result, the recognition of artificially pronounced words in isolation represents a significant simplification of the problem. Word recognition is then a typical pattern recognition problem. Any pattern recognition system comprises the following three parts:

• a sensor (in our case, a microphone) for recording the physical phenomenon in question;

• a shape parameterization stage (e.g. a spectral analyzer);

• a decision stage responsible for classifying an unknown shape into one of the possible categories.

These three stages can be found in an isolated word...