You may be able to improve the accuracy of speech recognition by experimenting with a couple of GetInput XML features.
Speech recognition hints
Provide a set of hint words and phrases to improve speech recognition accuracy. This feature can greatly improve transcription accuracy of proper nouns, homophones (such as one and won), and domain-specific words rarely used in the general parlance. You can build a repository of words and phrases expected from speakers in the hints attribute of the GetInput XML element.
Speech models
Experiment with the prebuilt speech recognition models. We recommend using the command_and_search model for command-driven IVR applications, and the phone_call model for more informal speech.