Direct sample interpolation (DSI) speech synthesis: An interpolation technique for digital speech data compression and speech synthesis

Direct transcription of the waveform is a potentially simple method to generate speech. However, it has a poor reputation due to the enormous data store required, although many digital encoding techniques have been suggested which reduce the data store significantly. Digital encoding is not invoked but a method called direct sample interpolation (DSI) is described which will compute bridging sections quite simply between, in the first instance, very short vowel sections called here phoneme fragments (PF's). This reduces the data store for vowels. DSI will significantly compress the store for other sounds, notably the fricatives. The method relies on the property in speech that the spectrum changes relatively slowly with time. Perhaps, most importantly, DSI can produce bridging sections of predetermined durations and pitch contouring and can therefore provide facilities for pitch inflexion and stress. Speech generation can be effected using DSI and the hardware is simpler than required for file "straight" LPC method. Speech synthesizers using the method have been built and some details of the data store and the hardware are given.