A system and method for distributed speech recognition is provided. A prompt is provided to a caller during a call. One or more audio responses are received from the caller in response to the prompt. Distributed speech recognition is performed on the audio responses by providing a non-overlapping section of a main grammar to each of a plurality of secondary recognizers for each audio response. Speech recognition is performed on the audio responses by each of the secondary recognizers using the non-overlapping section of the main grammar associated with that secondary recognizer. A new grammar is generated based on results of the speech recognition from each of the secondary recognizers. Further speech recognition is performed on the audio responses against the new grammar and a further prompt is selected for providing to the caller based on results of the distributed speech recognition.