When a audio stream is coming from an open microphone or telephone handset etc., the moment the user speaks is a barge-in event. The audio data before the barge-in is usually discarded (except for a small amount right before the barge-in). Commonly the user application is playing a prompt to the user and streaming a recording of the user to the Speech Engine. Until the user speaks, this data is discarded. When barge-in occurs, the user application can stop playing the prompt, but continue to stream recorded audio to the Speech Engine until end-of-speech is recognized.
Complete Help Topic List | Speech Engine Product Information