1. INTRODUCTION

Recent progress in digital media technology and distribution systems, such as the Internet and cellular phones, has enabled consumers to easily access, copy, and modify digital audio. Several Digital Watermarking (DW) techniques for audio files have been proposed for hiding data for protecting their copyrights. There is generally a tradeoff between the quality of watermarked audio and the tolerance of watermarks to signal processing methods, such as compression.

In previous research [1], to essentially overcome this issue, we developed an authentication method for digital audio to protect the copyrights. In contrast to DW, no additional information is inserted into the original audio by the previously proposed method, and the digital audio is authenticated using features extracted using a Discrete Wavelet Transform (DWT) and characteristic coding of the previously proposed method [1]. However, in the previously proposed method [1], it is indispensable to determine the region to be authenticated in the audio data by using the fixed length and the fixed starting time from the beginning of the audio data. Therefore, the authentication tolerance to clipping of the audio data is essentially insufficient for practical use.

In the present study, to overcome this issue, we have improved the method by determining the region to be authenticated in the audio data by using the fundamental frequency characteristics.

2. OBSERVED PHENOMENON UNDERPINS THE AUTHENTICATION METHOD

The procedure and algorithm of our previously proposed method [1] is reviewed in this section, because it is very important for the present study.

It has been observed that when a DWT is applied to audio data, in the histogram of the wavelet coefficients of the Multi-Resolution Representation (MRR), the center of the distribution is very close to zero [2]. We exploited this phenomenon to develop an authentication method for audio data [1]. For further information on the DWT used, see Inoue and Yoshitomi [3] and Taniguchi and Yoshitomi [4].

3. AUTHENTICATION RATIO

We set the authentication parameters as described below [1,5].

In Figure 1, Th′(minus) was chosen so that it divides the nonpositive wavelet coefficients ( Sm′ in total frequency) into two equal groups, and similarly Th′(plus) was chosen so that it divides the positive wavelet coefficients ( Sp′ in total frequency) into two equal groups. Next, the values of the parameters T1′–T4′, which control the authentication precision, were chosen such that the following conditions were satisfied:

1. (1)

   T1′ < Th′(minus) < T2′ < 0 < T3′ < Th′(plus) < T4′.

2. (2)

   The value of ST1′ , the number of wavelet coefficients in (T1′, Th′(minus)), is equal to ST2′ , the number of wavelet coefficients in [Th′(minus), T2′), i.e., ST1′=ST2′ .

3. (3)

   The value of ST3′ , the number of wavelet coefficients in (T3′, Th′(plus)], is equal to ST4′ , the number of wavelet coefficients in (Th′(plus), T4′), i.e., ST3′=ST4′ .

4. (4)

   ST1′/Sm′=ST3′/Sp′ .

Figure 1

Three sets (F, G, and H) of MRR wavelet coefficients used for authentication [1].

In the present study, the values of both ST1′/Sm′ and ST3′/Sp′ are set to 0.3, which is the same setting used for creating the code for the original audio data [1]. When preparing the authentication codes, the wavelet coefficients V′ for each MRR sequence are divided as shown in Figure 1 into three sets, which are defined as follows:

• •

  F = {V′|V′ ∈ V′^AC, V′ < Th′(minus)}

• •

  G = {V′|V′ ∈ V′^AC, Th′(minus) ≤ V′ ≤ Th′(plus)}

• •

  H = {V′|V′ ∈ V′^AC, Th′(plus) < V′},

where V′^AC is the set of wavelet coefficients from the target audio data that is used to create the authentication code.

The wavelet coefficients Vi′ are then classified according to the following rules with the flags f_i used in creating the original code C:

• When f_i = 1 and Vi′∈G , bi′ is set to 0.

• When f_i = 1 and Vi′∈(F∪H) , bi′ is set to 1.

• When f_i = 0, bi′ is set to 0.5.

Note that the value 0.5 can be chosen arbitrarily, since the value of b_i that is the bit for creating the code for the original audio data [1] does not influence the method’s performance. Finally, this sequence of bi′ values is used to form the authentication code C′.

The authentication ratio AR(%) is defined as follows [Equation (1)]:

where N is the number of wavelet coefficients chosen to create the authentication code for the original audio data [1]. As can be seen in Equation (1), neither b_i nor bi′ influences the value of Authentication Ratio (AR) when f_i = 0, which occurs when the corresponding V_i that is the wavelet coefficient of the original audio is not selected for coding in the original audio data [1].

To use the proposed method, we need to store the flags f_i and the original code C for each copyrighted file that we want to protect. When calculating (1) to authenticate audio data, we do not use the original audio data; instead, we use the flags f_i and the code C for that file [1].

4. FUNDAMENTAL FREQUENCY CHARACTERISTICS OF AUDIO DATA

Figure 2 shows the fundamental frequency of the first entry of the rock music genre category in the music database RWC for research purposes [6]. As shown, several local maximums of the fundamental frequency exist within the stream of the music. We have proposed a method for determining the region to be authenticated in the audio data by using local maximums of the fundamental frequency of the audio data, as described in the next section.

Figure 2

Fundamental frequency of audio data.

5. PROPOSED METHOD FOR DETERMINING THE REGION TO BE AUTHENTICATED IN THE AUDIO DATA

The audio data clipping procedure for the authentication is as follows:

• Step 1: The fundamental frequency f(i)(1, 2, 3,..., n) at the start time index i is measured every 0.01 s from the beginning to the end of T s of the original audio data, i.e., n = 100T. Then, the absolute value of difference ADF(i) = |f(i + 1) – f(i)| is calculated for all i(i = 1, 2, 3,..., n – 1).

• Step 2: The sum S(i)=∑j=0999ADF(i+j) is calculated for all i(i = 1, 2, 3,..., n – 999).

• Step 3: The start time indexes i are determined by whether the value of S(i) is among the top 10 of all S(k) under the restriction |i – j| ≥ 1000 for all j such that S(i) < S(j).

• Step 4: For the start time indexes i selected in Step 3, 10 s of audio data are clipped for the authentication.

6. EXPERIMENTS

7. CONCLUSION

In general, there is a tradeoff between the quality of watermarked audio and the tolerance of watermarks to signal processing methods, such as compression. To overcome this inevitable tradeoff, we previously developed an authentication method [1] for digital audio using a DWT. In the present study, we have improved the method by determining the region to be authenticated in the audio file by using the fundamental frequency characteristics. The experimental results show that the method has a high authentication tolerance to clipping small parts from the audio data.

CONFLICTS OF INTEREST

The authors declare they have no conflicts of interest.

Authors Introduction

Dr. Yasunari Yoshitomi

He received his B.E, M.E. and PhD degrees from Kyoto University in 1980, 1982 and 1991, respectively. He works as a Professor at the Graduate School of Life and Environmental Sciences of Kyoto Prefectural University. His specialties are applied mathematics and physics, informatics environment, intelligent informatics. IEEE, HIS, ORSJ, IPSJ, IEICE, SSJ, JMTA and IIEEJ member.

Mr. Shohei Tani

He received his B.S. degree from Kyoto Prefectural University in 2013. He works at Fukuchiyama City Hall.

Mr. Masaki Arasuna

He received his B.S. degree from Kyoto Prefectural University in 2017. He works at Nissay Information Technology Co., Ltd.

Mr. Ryota Kan

He received his B.S. degree from Kyoto Prefectural University in 2017. He works at Shimazu Business Systems Co., Ltd.

Dr. Taro Asada

He received his B.S., M.S. and PhD degrees from Kyoto Prefectural University in 2002, 2004 and 2010, respectively. He works as an Associate Professor at the Graduate School of Life and Environmental Sciences of Kyoto Prefectural University. His current research interests are human interface and image processing. HIS, IIEEJ member.

Dr. Masayoshi Tabuse

He received his M.S. and PhD degrees from Kobe University in 1985 and 1988 respectively. From June 1992 to March 2003, he had worked in Miyazaki University. Since April 2003, he has been in Kyoto Prefectural University. He works as a Professor at the Graduate School of Life and Environmental Sciences of Kyoto Prefectural University. His current research interests are machine learning, computer vision and natural language processing. IPSJ, IEICE and RSJ member.