SPRAAK does not have its own tools for computing statistical language models but interfaces well with models generated with well established open source toolkits such as the SRI or CMU Language Model toolkits.

The language models produced by the mentioned toolkits come in the '.arpabo'-format (N-grams with backoff). The SPRAAK continuous speech decoder uses a compact binary format for efficiency reasons. The program spr_lm_arpabo.c converts back and forth between the ASCII arpabo and the SPRAAK binary format.