Abstract: The increased intimidation in pilfering copyrighted multimedia files demands the imprint of ownership identity called fingerprint in video/audio files. This would aid to indentify and accuse the imitators. Several hiding methods using image, audio, video or text as cover media have been found in literature with software based implementations like MATLAB. Some of these have also been reported in hardware platforms like FPGAs. As a trade-off between these implementations, embedded devices such as ARM core were used for image steganography on gray images to embed less payload in an attempt to abide the resource constrains on embedded devices. This study suggests an indicator based random LSB coding method that embeds fingerprint on a WAV audio file. The implementation was carried out on an embedded device LPC2148 with ARM7 core housed on MCB2140 evaluation board that supports audio play through on board amplifier and speaker. The original and embedded audio signal in their digitized form resides in FLASH and SRAM on-chip memories of LPC2148.
INTRODUCTION
High quality in image, audio and video files are attainable in today’s world with the modern word, digitization. The digital form makes it painless for an illicit person to make illegal copies of this digital content. This would distress the music, books, film and software development industries. These concerns over shielding the access to digital files and share out copies have given rise to noteworthy research that discover ways to warily conceal copyright information and serial numbers in digital files. These can help identify copyright violators and also to prosecute them. Data hiding techniques (Chan and Cheng, 2004; Amirtharajan and Rayappan, 2013) have been proposed and in exercise to carry secret digital information in out of sight over another digital media. These techniques at times may also use cryptography (Amirtharajan et al., 2013h, i; Janakiraman et al., 2014a, c; Salem et al., 2011; Zaidan et al., 2010) to scramble the data before the hiding takes place in spatial or transform domain (Praveenkumar et al., 2012a, b, 2013a, b, 2014a-j; Thenmozhi et al., 2012; Thanikaiselvan et al., 2012a, b, 2013a, b).
Depending on the reason behind the information to be carried, this hidden communication technique gets broadly classified in to steganography (Al-Azawi and Fadhil, 2010; Cheddad et al., 2010; Amirtharajan and Rayappan, 2012a, b, d; Amirtharajan et al., 2013e) and watermarking. A digital watermark is a sort of indication hidden in a noise-tolerant signal like audio (Gopalan, 2003), image (Thien and Lin, 2003; Rajagopalan et al., 2014a-d) or video data (Al-Frajat et al., 2010). It is usually used to imply actual ownership or copyright in addition to authenticity or integrity verification of the content in digital media. Digital watermarks may or may not be visible. That is, they can be made perceptible occasionally at times after the use of some algorithm and imperceptible at all other times. Ideally, the watermark should not disrupt the original signal. Distorting the carrier information in a perceptible manner through digital watermark is not worthy. Conventional watermarks can be found inside visible media (like images or video) whereas; for digital watermarking and cover signal can also be audio or texts.
A special case of watermark is called ‘digital fingerprint’ where a specific data or information embedded in a digital media file that can uniquely identify every digital file. Digital fingerprints are compact unique information carefully embedded into the original digital content which may be audio, image or video. This fingerprint represents the contents’ characteristics and has enough details to identify a content variant from various different sources and can effectively authenticate a file upon comparison with a database. Ideally, digital fingerprints should uniquely identify every audio or video files even when subject to modification like compression, re-sampling or even content degradation. While a digital watermark identifies ownership and traces the copyright infringements, digital fingerprints are intended to prosecute the defaulters. Fingerprint generation and identification forms an integral part of the content distributor’s media workflow with ability to identify, track, monitor and monetize their content. It empowers publishers with technology to prevent copyright infringement, provides means to extend the due financial benefits to rightful content owners and obviate themselves from legal liabilities arising due to unlicensed spread of copyrighted material.
High speed, expensive solutions with reconfigurable devices such as FPGAs (Janakiraman et al., 2013; Rajagopalan et al., 2012a, b; Ramalingam et al., 2014a, b; Rajagopalan et al., 2014b) and low cost compact solutions using ARM processors (Janakiraman et al., 2012a) are also reported in literature for information hiding with digital images. In addition, a few implantations on hardware reported the use of audio carrier for hiding information (Rajagopalan et al., 2014a, b). Steganography with audio as cover media (Gopalan and Shi, 2010) uses several methods at various complexity levels ranging from simple LSB coding (Asad et al., 2011; Cvejic and Seppanen, 2002, 2004), parity coding, phase coding, spread spectrum and echo data hiding (Bandyopadhyay et al., 2008). Audio fingerprinting (Cano et al., 2002) has many real world applications in the music and television industry.
A method called Pixel Indicator Technique (PIT) was proposed for RGB image steganography in order to strengthen the LSB algorithm (Amirtharajan et al., 2013c, d, j). Janakiraman et al. (2012b) suggested two variants for the use of PIT method on grey scale images (Janakiraman et al., 2012a). This study uses the simplest embedding technique on audio file namely the LSB coding combined with PIT considering a 16-bit audio sample as both indicator and channel for random embedding (Amirtharajan et al., 2012; Amirtharajan and Rayappan, 2012c; Amirtharajan et al., 2013a, b, f, g; Janakiraman et al., 2014b; Rajagopalan et al., 2014a, b) of fingerprint.
In this study, a finger print embedding algorithm implemented on ARM 7-LPC2148 device with audio file in WAV (Waveform Audio) file format is discussed. The device can be operated at 3.3V and runs up to the maximum frequency of 60 MHz. Its on-chip memory of 512 KB FLASH and 32 KB of SRAM puts a limitation on the size of audio file to be used in-turn, limiting the amount of data to embed. The fingerprint algorithm manages a balance between ideal amount of data embedded to enable comparison while keeping fingerprints lightweight for manageable access, indexing, search and storage. It can also be used to prosecute the copyright infringements legally. A typical digital fingerprinting process involves content owners registering their content for fingerprinting and creating reference digital representation of their content in a database which is used for future comparisons. Spatial domain is more suitable for hiding techniques in embedded processors like ARM (Daniela Stanescu et al., 2009).
METHODOLOGY
Audio Fingerprint Indicator (AFI) method described here two bits uses from MSBs (b15 to b2) in the selected 16-bit audio samples to indicate the presence of a fingerprint bit in the LSB position (b0) of that audio sample. Initially, the algorithm selects the indicator bits (bn,bm) based on which the fingerprint bit is to be embedded in audio samples. Based on the size of cover audio file (Amax) and size of fingerprint (Fpmax), the minimum embedding interval (EImin) between the audio samples is selected to distribute the fingerprint bits all over the cover. In every audio sample Ai separated by the interval EImin, the fingerprint embedding is done using LSB coding based on the value of indicator bits bn, bm. The AFI choices for embedding are given in Table 1.
The implementation of embedding algorithm detailed in Fig. 1 is done on an embedded device LPC2148 comprising ARM7 core. The digitized audio input file is in WAV format, which is a 16-bit Pulse Code Modulated (PCM) version of original analog audio signal for the duration of about 0.9 sec, sampled and quantized at 128 bits per second. This is first converted to 16-bit hexadecimal format (HEX386 or H86) using a DOS application utility program called Bin to Hex converter that runs on 32-bit Windows systems producing 14428 samples of each 16-bit in length. The hex file created using the above said utility program is loaded directly in to the 512 KB of on-chip FLASH memory.
The algorithm receives a 32-bit fingerprint value in the beginning from the user through asynchronous serial communication with a baud rate of 115.2K bits using the on-chip USART of LPC2148. During the course of embedding process, the embedded and non-embedded audio samples are stored in on-chip SRAM of 32KB provided by LPC2148. The algorithm interacts with the user during demonstration through the USART of LPC2148. Once, the device sends the embedding completion message, the user can select their options to play the embedded file or non-embedded audio files from SRAM and FLASH memories, respectively. Alternatively to verify the embedding, the user can extract the fingerprint. These options can be given as input to the algorithm through USART.
To facilitate the testing of proposed AFI algorithm in real time, in this study, MCB2140 evaluation board from KEIL is used. The board supports access to the FLASH memory of LPC2148 via., either by In System Programming (ISP) or by JTAG based debugger units like KEIL ULINK2. To play the audio files, the audio samples are sent to the 10-bit on-chip DAC of LPC2148 the output of which is amplified by a low power amplifier that drives the on board speaker module. During the play of audio signals through the speaker, the speaker volume can be adjusted using a potentiometer that controls the analog input applied to the 10-bit successive approximation based on-chip ADC of LPC2148. The block diagram of the entire hardware setup is revealed in Fig. 2.
| Table 1: | Embedding operation based on AFI |
The absolute process done by the embedded software is described in the pseudo code given below.
| Fig. 1: | Flowchart for Audio Fingerprint Indicator (AFI) embedding process |
RESULTS AND DISCUSSION
The embedded software for this Audio Fingerprint Indicator (AFI) algorithm was developed using KEIL mdk, an Integrated Development Environment (IDE) for ARM devices. Here, we embedded a 32-bit fingerprint comprising of 4 characters PIT! [0x50, 0x49, 0x54, 0x21]. The embedding process was mainly carried out using the undemanding LSB coding method which is amalgamated here with an indicator based embedding technique that is usually seen with image steganography.
The proposed AFI method reinforces the conventional audio LSB coding method by the introduction of two algorithmic parameters namely the minimum Embedding Interval (EImin) and the embedding indicator bits (bn, bm). Though the minimum interval between the samples getting embedded with fingerprint is ensured by EImin, the average of actual distribution interval (EIavg) varies based on the indicator (AFI) bit values. Still choosing the value of EImin large enough helps spreading the fingerprint over large duration of audio signal measured as coverage of distribution in terms of percentage.
| Fig. 2: | Block diagram for complete hardware setup |
The quality of fingerprint embedded audio signal against the real audio signal is measured by the factor Signal to Noise Ratio (SNR) given by:
where, So and Se are the original and embedded sample values of audio file with length Amax. The value of SNR, coverage of fingerprint distribution and average embedding interval against the selection values for algorithmic parameters EImin and bn, bm are tabularized in Table 2.
The inference from Table 2 shows that for the AFI values (3, 2), (7, 6) and (9, 8) when Eimin is taken as 20 coincidentally brings out equal distribution percentage and EIavg while their SNR values are slightly differing. The graph in Fig. 3 shows the randomness between the audio samples selected for fingerprint embedding based on AFI values.
The obtained results showed no perceptible difference in the audio quality between embedded and original files when played at maximum volume selected through the volume control POT. The snapshots of partial audio signals with and without embedding taken during their play from logic analyzer of KEIL mdk 4.7 are shown in Fig. 4 and 5 correspondingly.
| Fig. 3: | Randomness in distribution of fingerprint based on selection of AFI bits |
| Fig. 4: | Original audio signal |
| Fig. 5: | Fingerprint embedded audio signal |
| Table 2: | Analysis of SNR and randomness based on AFI algorithm parameters |
| *EER (Embedding Error) occurs when required numbers of samples are not available in audio cover with the value of selected AFI bits that leads to incomplete embedding of fingerprint | |
The real time values of execution time on MCB2140 evaluation board was measured using the hardware debugger unit ULINK2. The embedding of 32-bit fingerprint takes about 65.89 ms whereas the extraction (decoding) takes 65.07 ms when LPC2148 is operated at 60 MHz.
CONCLUSION
The extended features of the ARM cores make it suitable for processing audio signals. Though the original input signals can be housed in larger FLASH area, the available SRAM for the storage of processed information limits the input size (cover). By embedding 32-bit finger over 14428 samples, this study achieved tolerable SNR in the range of 79.7dB-82.1dB. During the audibility test carried out by playing the original and embedded audio files by adjusting the volume control POT to different levels, it was found that the audio quality of embedded file degrades proportionally for the reduction in volume whilst, the original audio file quality remains unaffected. This can be considered as a severe glitch that makes the job of spotting the difference between original and embedded audio files effortless. On the other side, AFI method serves as pseudo random generator in the selection of audio samples for fingerprint embedding and thus strengthens the simple LSB coding.