The Voice Dictation object maintains a mirror image of the text that is in the document. If a user loads in a large document, this might get unwieldy, especially if audio recordings are maintained for each of the words. Audio data is inherently large. Here are some ways to alleviate some of the memory problems:
The application can call IVDctAttributes::MemorySet() to specify how much memory should be used by storing the audio and other speech recognition correction information. Of course, if hundreds of pages are being mirrored, there's a minimum amount of memory that will be consumed.
Alternatively, an application might only keep the last page or paragraph that the user has dictated in the Voice Dictation buffer. This places a cap on the size of the data and may work well for the user. However, if the user scrolls back a few pages and wishes to hear what he/she said then they might be disappointed because the information won't be available.