When a user listens to the audio, also allow them to propose fixes for specific segments, which can then be fed back into the ASR training process.