The attack & release times could be set incorrectly for speech.
[ LevelSpeech2.ny does a good job of compressing speech ]