TODO normalize volume of segments use different timbres for each voice In the transcriptions, the dropoff in volume from the loudest to the next loudest note is too great