Speech recognition relies on the language model in order to decode an utterance, and in general a better language model improves the performance of a speech recognizer. We have recently found that a time-based language model can improve on a standard trigram language model in terms of perplexity. This technical report presents the evaluation of this new language model in the context of speech recognition. First, a basic speech recognizer was built using the HTK tool. Then the recognizer was run using the standard language model and using the time-based one. On a testset of 39,147 words from the Switchboard corpus, there was a slight improvement, with the percentage of words correctly recognized going from 11.31% to 11.40%.