Speech/Music Classification Using Features from Spectral Peaks