Pitch assignment rules for speech synthesis by word concatenation

This paper describes an algorithm that has been developed to generate appropriate prosodic patterns for synthesized sentences produced by concatenating LPC-encoded words and phrases of English. The method involves the assignment of one of three accent levels to each syllable of all encoded vocabulary items. The highest accent marking for a given word is dependent on its grammatical category. These accent assignments are then used to determine the relative pitch for each word in a synthesized sentence. The pitch assignment algorithm is currently being used in applications requiring automated speech output.