An application can query a speech-recognition results object to get detailed information about a recognition. The information that is available depends on the engine, but it may include a digital-audio recording, features extracted from the audio stream, the most likely phrases recognized, and the name of the speaker.
An application can also use a results object to perform certain tasks, such as training the engine to a particular speaker's voice or reevaluating a recognition result.
The information in a results object can occupy a significant amount of memory, so an application may instruct the object to discard certain parts of its data. In addition, an engine vendor may permit the results object to "deteriorate" or discard data automatically as time passes. An application can preserve speech-recognition results objects by locking them, or it can store results objects by archiving them with the associated grammar.