Rather than ignore the time stamps, use the time stamps in an imported transcript to help the AI match up the text to the audio file. If the person doesn't start talking again for thirty seconds, the text won't try to 'match up' to the silence.