Speech synthesis in the time domain from text

A very efficient method to synthesize speech is the combination of digitally stored monophones and transients in the time domain, because of its good intellegibility and the easy implementation which allows a very inexpensive realisation. A disadvantage inherent in procedures in the time domain has so far been the fact, that they required a very large sized memory, the size of which increased in multiples if prosodic parameters were also taken into account. We developed a system which synthesizes an unlimited vocabulary with a memory of only 22 kBytes (8-Bit Bytes). Further investigations showed, that by means of the linear prediction method it is possible to control the fundamental frequency of the speech signal in a wide range without storing additional speech segments. In addition to this work, we developed a system to transform orthografic text into a phoneme string automatically. We optimized this algorithm for the 8000 most frequent words of the German language. The whole system which is implemented on a microprozessor, is placed on a single board, with a storage of total 32 kBytes (8-Bit Bytes).