Non-deterministic audio synthesis?

[erbsenzaehla]

I am using the slt_arctic demo on merlin without any changes.
I ran the demo without problems, but when it comes to synthesizing, everytime I run the script, it produces a different output. Sometimes it's the words I want to hear, but often it is only noise. Shouldn't this be a deterministic program with the same output every time (with always the same input)?