This paper describes the design and generation of the acoustic database for a Spanish text-to speech synthesizer, developed jointly by AT&T BELL LABORATORIES and TELEFONICA I+D, based on the concatenation of LPC coded units. These units can be referred to as "paraphones" since they can be subphones, diphones, triphones or even polyphones of a greater order. For our work, we first established a list of the allophones required; then constructed a dictionary of paraphones; devised a list of sentences containing all this paraphones; recorded all these sentences and finally segmented from them the units for our database.