These programs use pre-recorded allophones, unlike as with the hardware based [[SP0256]] speech synthesizer chip, the allophones aren't computed an the fly. However, the overall result is the same: One can combine allophones to produce any kind of words and sentences, the software solution requires more memory and more CPU load though.
Speech! uses about 8K of memory for player and sample data. The samples are stored as packed 4-bit values to send to the AY. To play the speech delay loops are often used to control the frequency at which the AY is updated, and therefore the frequency of the sample. However, when Speech is playing the CPU time is effectively used completely until it is done, unless the player could be modified?
== Games believed to use Software Speech Synthesizer ==