A Speech Decoder (or simply "Decoder") is the software portion of the speech recognition engine.  In addition to a Decoder, Speech Recognition engines need an Acoustical Model and a Language Model or Grammar in order to recognize speech.