結合浮水印與音頻指紋之音訊認證研究

數位音訊的傳遞過程易遭受非法竄改，竄改的音訊在認證上是困難的，包含音訊發送者的身分識別與音訊遭受竄改的偵測等都是問題。因此，我們計畫透過浮水印與音頻指紋結合來解決上述問題。其方法有四個步驟，首先，利用頻率掩蔽效應，將浮水印嵌入於離散小波轉換(Discrete Wavelet Transform)的頻率域中。第二，將含浮水印的音訊，經過梅爾倒頻譜(Mel-Frequency Cepstral Coefficient, MFCC)及線對譜(Line Spectral Pairs, LSP)的特徵擷取建構音頻指紋。第三，從音訊載體中萃取出浮水印。第四，驗證比對階段。在這個階段中，我們利用四個量化值來進行驗證及比對，它們分別為正規化關聯值(Normalization Coefficient, NC)、音訊峰值信噪比(Peak Signal to Noise Ratio, PSNR)、音頻指紋訊號雜訊比(PSNR)、及含浮水印音頻指紋(Audio Watermark Fingerprint, AWF)與不含浮水印音頻指紋(No Audio Watermark Fingerprint, NAWF)的差值。實驗結果顯示，利用本論文所提出的方法，浮水印嵌入音訊中人耳無法察覺，且能抵抗常見的音訊攻擊，如低通濾波、裁剪、加噪聲、MP3失真壓縮、加大振幅等等。攻擊後的音訊浮水印萃取過程不需原始音訊，且 NC值平均可達到90%以上的強韌性，使浮水印可成為音訊來源身分識別的標誌；此外，透過音頻指紋AWF減去NAWF的差值比對，我們可以偵測出音訊遭竄改的位置。

關鍵字

音訊認證；浮水印；音頻指紋；梅爾倒頻譜；線對譜

並列摘要

The digital audio is easily tampered with during the process of transmission. However, authenticating the identity of the sender or even detecting tamper is difficult. Therefore, in this thesis the watermarking and audio fingerprinting techniques are combined to solve these issues. The scheme is in four parts. The first part involves the audio masking effect where the watermark is embedded in the audio’s frequency domain after Discrete Wavelet Transform. The second part takes the audio signal containing the watermark and parses it through for feature acquisition by Mel-Frequency Cepstral Coefficient (MFCC) and Line Spectral Pairs (LSP) in order to construct the audio fingerprint. The third part involves the watermark extraction from the audio carrier. The fourth and final part is the verification and comparison stage. The fours variables for verification and comparison are the respective Normalization Coefficient (NC), Peak Signal to Noise Ratio (PSNR), audio fingerprint PSNR, and the difference between Audio Watermark Fingerprint (AWF) and No Audio Watermark Fingerprint (NAWF). Experimental results show that the embedded watermark is not detectable by the human ear. Furthermore, the scheme is robust against some common audio attacks, like, low pass filtering, cutting, noise adding, MP3 loss compression and amplitude amplification. After any of these attacks, the audio signal watermarking retrieval does not require the original audio signal. Also, the NC value showed a robustness of more than 90%. The combined watermark and fingerprint allows the detection of tampering. Furthermore, the difference value between audio fingerprint AWF and NAWF can reveal the location of the tampered audio signal.

並列關鍵字

Audio authentication ； Watermarking ； Audio fingerprint ； Mel-Frequency Cepstral Coefficient ； Line Spectral Pairs

參考文獻

[1]I. Biehl, and B. Meyer, “Cryptographic Methods for Collusion-Secure Fingerprinting of Digital Data,” Computers and Electrical Engineering, Vol. 28, No. 1, pp. 59-75, 2002

[2]C. Baras, N. Moreau, and P. Dymarski, “Controlling the Inaudibility and Maximizing the Robustness in an Audio Annotation Watermarking System,” Transactions on Audio, Speech, and Language Processing, Vol. 14, No. 5, PP. 1772-1782, 2006

[3]S. Baluja, and M. Covell, “Waveprint: Efficient Wavelet-Based Audio Fingerprinting,” Pattern Recognition, Vol. 41, No. 11, pp. 3467-3480, 2008.

[4]C.C. Chang, H.C. Tsai, and Y.P Hsieh, “An Efficient and Fair Buyer-Seller Fingerprinting Scheme for Large Scale Networks,” Computers & Security, Vol. 29, No. 2, pp. 269-277, 2010.

[5]J. Chen, T.S. Chen, and C.Y. Cheng, “New Image Tampering Detection and Recovery System of JPEG2000 Region of Interest Area,” Imaging Science Journal, Vol.53, No. 1, pp. 12-19, 2005.

國際替代計量

結合浮水印與音頻指紋之音訊認證研究

未授權

主題瀏覽