Occurrences of at least one of a plurality of telephone number-related speech patterns are identified in an audio stream using speech analysis techniques. At least one number pattern corresponding to the occurrences are thereafter identified and stored for later presentation to a user of a mobile device. The occurrences and corresponding number pattern(s) may be filtered to reduce the likelihood of false detections. Furthermore, the detected number pattern(s) may be augmented to ensure completeness. Users may edit and persistently store the resulting number pattern(s). In one embodiment, a portion of the audio stream may be stored and subsequently subjected to the speech analysis process. The stored portion of the audio stream may be reproduced at substantially the same time that at least one number pattern detected therein is provided to the user. In this manner, telephone numbers in audio streams may be captured safely and conveniently.