In designing a speech recognition expert system, there are a number of
aspects to be considered-segmentation method, recognition unit, struc
ture of the rule, and user aids. To avoid the difficulties in segmenta
tion, an irregular unit based on spectral transition measure is propos
ed. Frames are used as the structure of the speech recognition rules.
The structure provides the user with an easy way of making rules, and
enables the building of an automatic rule generator. Fuzzy linguistic
variables are used in representing rules. The rule-generating cycle, w
hich includes a rule generator, a rule tester with an error reporter,
and a state describer with other modules, could save much time in maki
ng rules and could result in high performance. The experimental result
s and comparison with the SPREX system, which is the basis of this sys
tem, are described.