Feature Extracing in OCR

4.2 Experimental Procedure

Since the project involves two types of recognition, it required implementing two different types of systems - an isolated word recogniser and a connected word recogniser. Both were speaker independent systems with small vocabularies.

4.2.1 Data

A random selection of 48 speakers was used for training, of which, 24 were male and 24 were female. The isolated digit models were trained using only isolated digit data. A total of 1056 isolated digit utterances were used for training. The connected digit recogniser was trained using the connected digit data, which included the isolated digit utterances. A total of 2640 utterances were used for training the connected digit recogniser.

Although the amount of data used for training both systems is comparatively small to the amount contained on the CD, it is sufficient for these baseline systems.

4.2.2 Documentation

Selected scripts and files generated for use in the experiments have ...