4.4.1 Database and condition
Experiments were carried out to evaluate inaudibility and robustness of the proposed method with 102 RWC music tracks [110] that have a sampling frequency of 44.1 kHz and 16-bit quantization. The frame size is set to 500 ms. The FFT size is equal to the frame size and the rectangle window was used. The watermarks were randomly generated. The parameters,F,ǫ,L, andM, were determined by experimental analysis and set to 1.6 kHz, 10−4, 5, and 0.005 respectively. TheQIM step sizes are chosen as integer divisions ofπ to reduce wrapping errors. We investigated the proposed method with three sets of fiveQIM step sizes as shown in Tab.4.2.
Inaudibility was tested byPEAQ [33] which rates sound quality by theODGfrom−4 (very annoying) to 0 (imperceptible). Detection accuracy was measured byBDR, the ratio between the numbers of correct bits and total bits. Robustness was investigated with the following processing: MP3 128 kbps, MP4 96 kbps, adding white Gaussian noise 36 dB, requantization 8 bits, resampling 22 kHz and 16 kHz, and bandpass filtering with passband [0.1, 6] kHz and stopband attenuation −12 dB/octave.
−4
−3
−2
−1 0
(a) PEAQ
PEAQ (ODG)
99.9 100
(b) No attack
BDR (%)
Proposed Static (π/2) Static (π/6) Static (π/10)
6 12 18 30 50 100 150 200 300 400
60 70 80 90 100
(c) MP3 64kbps
BDR (%)
Bit rate (bps)
Figure 4.6: Sound quality (PEAQ) and BDR with respect to bit rate in comparison with the static scheme: (a)PEAQ, (b) BDR without attack, and (c)BDR against MP3 attack
4.4.2 Effectiveness of the proposed method
We carried out experiments to test the PEAQ and BDR of the proposed method (DPC) in comparison with an audio data hiding method based on a static phase coding (SPC) scheme, in which the QIM step size is fixed for all the frequency components. The first QIMstep set (Set 1) was used for the proposed method. We investigated theSPC scheme with the QIM step size ofπ/2,π/6, and π/10.
Figure4.6shows the results of thePEAQand theBDR. For theSPCscheme, thePEAQ increases from−2.17ODGto−0.46ODGwhile theBDR decrease in both the cases of no attack and MP3 attack as the QIM step size decreases from π/2 to π/10. Especially, the BDRdrastically decreases when the bit rate increases. These results suggest that the SPC scheme meets a trade-off among inaudibility, robustness, and capacity.
In contrast, the proposed method has the PEAQ approximate to −1ODG and a high
0 0.5 1 1.5 2 2.5 3
Org-Org DPC-Org CDb-Org DSS-Org
Mean score
Test pair
5 9 10 12 14 21 22 23 26 27
29 59 63 88 90 92 97 99 100 101
Track no.
Figure 4.7: Results of subjective listening test.
BDRfor both the cases of no attack andMP3attack, even at high bit rate (up to 300bps).
The results reveal that the proposed method has a better trade-off among inaudibility, robustness, and capacity than the SPC scheme. Accordingly, the proposed DPC scheme is effective for audio data hiding in obtaining a reasonable trade-off among the properties.
4.4.3 Subjective listening test
The subjective listening test was carried out with the settings as the same as those in Sect.3.3.3. For each track, there are four pairs in a random order, in which the first one in the pair is the original track (Org) and the other one is the original track or an watermarked track created by the proposed method (DPC), theCDb method [112], or the DSS method [17]. Each subject evaluated the similarity of 80 test pairs in total, which is 20 tracks multiplied by 4 pairs (Org-Org, DPC-Org, CDb-Org, and DSS-Org).
Figure 4.7 shows the mean scores of four test pairs for each track. The mean scores of the pairs DPC-Org are approximate to those of the pairs Org-Org, revealing that the distortion caused by the watermark embedding of the proposed method is hardly percepti-ble. In contrast, the distortion caused by theCDbmethod and theDSS method is slightly perceptible and perceptible for most of the tracks.
4.4.4 Robustness of the proposed method
We carried out experiments to evaluate the robustness of the proposed method. Figure4.8 shows the results ofPEAQand BDRof the proposed method with three sets ofQIM step size. The bit rate was varied from 6 to 400 bps. All thePEAQs are greater than−1ODG (not annoying) and the sound quality of watermarked signals remains unchanged as the bit rate increases. The sound quality becomes better when the QIM step size decreases from the values in Set 1 to those in Set 3.
In the cases of no attack and resampling, the BDRsare greater than 99.9% for all the bit rates and do not change much among three sets ofQIMstep size. In the cases of MP3, MP4, adding white noise, and requantization, the BDRsare greater than 99% for the bit rates less than or equal to 200 bps with Set 1 and slightly decrease with Set 2 and Set 3.
The bandpass filtering seems to be the strongest attack which makes the BDRs around 90% for the bit rates less than or equal to 200 bps with Set 1 and decrease much more with Set 2 and Set 3.
These results suggest that the proposed method is effective with regard to inaudibility and robustness. The proposed method provides good sound quality in the watermarked signals and high robustness against most types of processing.
4.4.5 Effectiveness of ECC
We evaluated the effectiveness of incorporation of ECC into the watermarking system in the case ofMP3 64 kbps. We chose to investigate MP3attack because it is popularly used in practice and is the strongest attack except for bandpass filtering. If ECC can correct errors after MP3 compression, it can also correct errors from the other attacks. The five codes with the length of 1023 and different values of k0 have been used.
Figure 4.9 shows the bit-error rate (BER) after BCH decoding, with respect to em-bedding bit rate with three sets of QIM step size. The results show that the system can extract watermarks without any detection error at a bit rate of 102, 51, and 28 bps with Set 1, Set 2, and Set 3, respectively. Compared with the case that ECC is not used to encode watermarks, the incorporation of ECC is remarkably effective in correcting all the errors at relatively high bit rates.