Speech depersonalization based on the long-term spectrum of voice