A Compression-domain Audio Fingerprint Technique Based on The Mathematical Expectation of MDCT Spectrum