Speech Synthesis by Rule From a Low Level Feature Based Input

Acoustic domain rule systems for generating highly intelligible continuous speech from a string of discrete linguistic elements using a terminal analog synthesisar have become an established research technique during the past decade and a half.Host early rule systems operated on the strategy of directly transforming a broad phonetic or phonemic transcription into a continuum of parametric data (1.2).More recently, rule systems (3,h) hava been developed which explicitly separate the rules dealing with high level and relatively abstract phonological processes from those dealing with low level phonetic detail.Such strategies are a recognition of the fact that these represent rather different aspects of the speech generation prooaaa.and allow appropriate modes of'rula expression to be exploited in each area.