A text-to-speech development system

A research system is described which was created at TI to aid the development of economical text-to-speech systems. Notable features of these text-to-speech systems are: 1.) Rules are compiled into an object data structure which is then interpreted; 2.) Source rule format is a quasi-linguistic generalization of that used by the NRL; 3.) Rules are organized into an arbitrary number of passes; 4.) In object-code form the rules are context-sensitive byte string transducers. The research system includes on-line data bases and utility programs for testing. The utility of the system is demonstrated by our development of a rule set having 1/2 the phonemic error rate of the NRL rules without a significant increase in execution time or memory used.