Background

Audio Watermarking is one security technology for content protection. The watermark (WM) is a signature embedded within the data of the original audio signal and, in addi ^¬ tion to being inaudible to the human ear, should also be statistically undetectable, and should be resistant to any attempts of removing it. For example in a Digital cinema ap ^¬ plication, the requirements of a watermarking system can be specified on the basis of a set of properties: in addition to the general requirements of quality of the watermarked copies and robustness of the embedded watermarks, such cin ^¬ ema application includes additional security constraints which are to be complied with.

'Digital Cinema Initiatives' (DCI) is an entity founded by seven motion picture studios to establish specifications for an industry-wide standard for Digital cinema. A DCI Specifi ^¬ cation Version 1.2 was published on http://www.dcimovies.com . The workflow for movie production includes mastering, distribution and playback in the theatres, as depicted in Fig. 1.

In the Digital cinema scenario different components are specified, e.g.:

· Digital Cinema Distribution Master (DCDM) ;

• Compression specifies the DCI compliant JPEG 2000 code stream and the JPEG 2000 decoder;

• Packaging;

• Transport;

• Theatre System;

· Projection;

• Security.

The Digital cinema distribution master DCDM shall be based on a hierarchical image structure that supports 2K and 4K master. The Digital cinema system is built on a file-based design, i.e. the complete content is made up of data stored in files which are organised around the image frames. The included DCDM audio data has a bit depth of 24 bit/sample and a sample rate of 48 or 96kHz. The file format is PCM WAVE and the data are not compressed.

In the Digital cinema system, the security system as de ^¬ picted in Fig. 2 has different requirements which are imple ^¬ mented with appropriate security mechanisms. This includes cryptographic security from distribution to theatrical play- back or projection 21. Encryption protects against illegal access, unauthorised copying, editing and playback. The encrypted content is distributed as a single Digital Cinema Package (DCP) to every theatre. The DCP content components are selectively encrypted by the rights owner (e.g. studio or distributor) . The logging mechanisms log the access to protected content. Watermarking and Fingerprinting are implemented to enable forensic tracking.

Watermarking is implemented in embedder 22 within the media block MEDBL of the server located in the theatre.

The most important requirement regarding the watermarking of audio data is that the watermarking is not audible in ABX listening tests. The payload to be embedded includes an identification of the theatre by using 19 bit word length, corresponding to 2^-^=524.288 possible locations. In addition a time stamp is embedded which changes every 15 minutes and is repeated every year, requiring 16 bits for its binary representation :

4 time stamps/h * 24h/d * 366d/y = 35136 time stamps/h Thus, the total length 1 (w) of watermark w information bits is 1 (w) = 19+16 bits = 35 bits. The data-rate is set to

1 (w) bit/5 min = 7 bit/min

Since the embedding of the watermark is to be performed just before playback in the theatre it is to be performed in real-time or even faster. The minimal time segment required for detection should not exceed 30 minutes.

In the following, the original or carrier object is denoted by c ₀ · In the notation used, represents

the samples of the original signal in the time domain, wherein l _c denotes the number of samples of track C _Q . E # cj ^" # for a sampling rate f _s =44.1kHz one second corresponds to a number of 44100 samples. According to the amplitude resolu ^¬ tion of 8 or 16 bit, the range of sequence of numbers is

. An additional index ' j '

of the carrier elements c ₀ j denotes a subset of the audio signal. Very likely, all audio watermarking algorithms split the audio signal into different overlapping or non- overlapping blocks, usually same-size blocks. For this rea ^¬ son c _Q j[i] denotes the i-th sample in the j-th block, with length The individual blocks are used to embed one bit

of information of the whole watermark w.

The intention of the publication 0. Billet et al . , "Efficient Traitor Tracing from Collusion Secure Codes", 10

Aug.2008, Information Theoretic Security, Lecture Notes in Computer Science, Springer Berlin Heidelberg, pp.171-182, XP019102288, is to enable the tracing of traitors even if several traitors collude to produce a pirate decoder from the shared secrets (keys) which are illicitly extracted from the decryption box (decoder) . Therefore it is a tracing of traitors by cryptographic means not using a watermarking technique. In case a pirate decoder can decrypt the content, no further mechanisms are available for tracing the illicit user on the basis of the content. In turn, for tracing unau ^¬ thorised users the tracer needs access to the pirate de ^¬ coder. In case of watermarking, the tracer only needs access to the pirate content with the ID of the user (or the DCI watermark) embedded in the content in contrast to a set of keys integrated in the decoder.

Invention

In contrast, the invention solves the problem that an adversary can potentially access the original content if embed ^¬ ding is performed at receiver site as it is the case in the digital cinema scenario. It can be combined with the method described above to increase the security level by providing additional security against collusion attacks due to the leaking of the security keys out of the decryption box (in the digital cinema case the DCI server) .

Due to the embedding of the watermark just before the pres ^¬ entation of the movie as shown in Fig. 2, that embedding is to be performed in real-time in the media block MEDBL of the theatre. This requires the audio watermarking algorithm to be implemented on a dedicated DSP hardware at receiver site (in the media server) . But this solution has several disad ^¬ vantages due to usage of DSP hardware and watermarking at receiver (theatre) site:

• Integration of DSP hardware at receiver site is expensive; · Updates of the watermarking algorithm (due to detected se ^¬ curity flaws) have to be performed in each single receiver (theatre) , which is time consuming and expensive;

• A potential security flaw is included in such system be ^¬ cause the original content is accessible after decryption and just before the embedding of the watermark in the me ^¬ dia stream; • The watermarking key is to be stored at, or transmitted to, the receiver site for the embedding, increasing the risk of cracking this security processing;

• The embedding of a correct time stamp cannot be ensured. A potential adversary could reverse-engineer the watermarking hardware and embed an incorrect time stamp by using previous decrypted versions of the frames having different watermark symbols embedded. Nevertheless the adversary can only construct watermarked copies with a time stamp within a range that can be controlled at sender site;

• Switching to a different (possible non real-time) water ^¬ marking processing may not be possible due to the con ^¬ straints of the integrated DSP hardware;

• Combining different watermarking procedures for watermark- ing the same media stream is impossible due to the hard ^¬ ware implementation of the processing.

These kinds of problems will occur in all applications where embedding is done at the receiver site and the watermarking algorithm is implemented in hardware.

A problem to be solved by the invention is to increase the overall security of such protection processing or system. This problem is solved by the method disclosed in claim 1. A system that utilises this method is disclosed in claims 2 and 3.

The invention shifts the burden of embedding the watermarks from the receiver site (theatre) to the sender site (post- production, studio) by using pre-watermarked and encrypted copies :

• A dedicated DSP implementation of the watermarking algorithm is no longer necessary. The watermarking can be carried out in software. This facilitates use of cheaper hardware;

• Updates of a new watermarking processing can be carried out easily because no hardware update is required;

• Access to the watermarking embedder and to the keys is not possible, thus preventing security problems regarding the watermarking encoding.

Several frame-wise pre-watermarked and encrypted copies of an audio PCM file are delivered. Frame-wise or block-wise watermarking and encryption is carried out at sender site (post-production or studio) , and the frame/block-wise en- crypted files are packed and delivered to the receiver site (theatre) . The different versions of the copies or tracks represent encrypted candidate watermarking with different symbols of the watermark. For example, the first copy of the audio or video signal is watermarked at sender site with value '0' watermark symbols for each block or frame, and the second copy is watermarked at sender site with value '1' wa ^¬ termark symbols for each block or frame. A further copy can contain sync watermark symbols.

In addition to the files delivered, a key sequence is trans- mitted to each receiver consisting of the keys for decrypting individual frames or blocks identified by the frame/ block number and the symbol embedded. At receiver site the decryption can be performed on the proper frames only, according to the transmitted key sequence. The pre-watermarked and decrypted frames are used to create a single watermarked copy that implicitly carries the correct watermark informa ^¬ tion bits, e.g. ID and time stamp in a DCI application. For creating that single watermarked copy, based on the bits or values of the desired watermark information bit word, the corresponding frames or blocks from the received or stored copies are taken and are assembled in a successive manner.

Due to pre-watermarking of the content at sender site, the embedding of the watermark has the following advantages: • Dedicated implementation of the watermarking processing in hardware is not necessary, reducing the costs of the hard- ware at the receiver site;

• Updating to a different watermarking processing is easy because no hardware update is necessary;

• Construction speed of the watermarked tracks depends on the I/O performance only and not on the speed of the un ^¬ derlying watermarking processing. Therefore a switching is possible to sophisticated watermarking processings which can not run in real-time;

• Combinations of different watermarking processings can be used, which alternately embed the watermark and thereby increase the robustness of the forensic tracking system. This can be controlled at the construction site by provid ^¬ ing timing information in case of different block sizes;

• The receiver (theatre) site receives pre-watermarked cop- ies having already a watermark (the ID of the cinema) em ^¬ bedded, which prevents unauthorised embedding of false wa ^¬ termarks at receiver site;

• The security of the watermarking system is ensured, since reverse-engineering of the watermarking algorithm is not possible. Furthermore there is no access to the keys of the watermarking system due to the pre-watermarking at sender site.

Nevertheless, the invention described so far has still two disadvantages:

• There is a potential security vulnerability at theatre site because the decrypted pre-watermarked copies can be used to construct a watermark carrying not the correct wa ^¬ termark (for example correct ID of a cinema but a differ- ent time-stamp) ;

• Due to using pre-watermarked copies the required bandwidth is nearly doubled.

Therefore the encryption can be carried out block-wise as described below.

In principle, the inventive method is suited for protecting by watermarking against non-authorised use, e.g. non- authorised recording or copying, an original audio or video data signal that is to be presented, said method including the steps:

- at sender site, generating from said original audio or video data at least two differently pre-watermarked versions for successive blocks or frames of said original audio or video data, wherein said versions are derived by applying a repeated watermark symbol value to a version and different watermark symbol values to the different versions, and en ^¬ crypting said versions using corresponding encryption keys, wherein in addition each block or frame of said differently pre-watermarked versions is encrypted individually using a corresponding key sequence;

- transferring said pre-watermarked and encrypted versions and said encryption keys and said key sequence to one or more receivers or receiver entities, e.g. as one or more data files;

at receiver site, decrypting said pre-watermarked and en- crypted versions of audio or video data using said encryp ^¬ tion keys and said blocks or frames individually using said key sequence and assembling, according to the values of a desired watermark information word, corresponding frames or blocks from said decrypted and pre-watermarked versions in a successive manner, so as to provide a watermarked version of said original audio or video data that carries said water ^¬ mark information word;

presenting said decrypted and watermarked version of said original audio or video data.

In principle, the inventive sender system is suited for pro ^¬ tection by watermarking against non-authorised use, e.g. non-authorised recording or copying, of an original audio or video data signal that can be presented at a receiver site, said sender system including:

means being adapted for generating from said original au- dio or video data at least two differently pre-watermarked versions for successive blocks or frames of said original audio or video data, wherein said versions are derived by applying a repeated watermark symbol value to a version and different watermark symbol values to the different versions, and by encrypting said versions using corresponding encryption keys, wherein in addition each block or frame of said differently pre-watermarked versions is encrypted individu ^¬ ally using a corresponding key sequence;

- means being adapted for transferring said pre-watermarked and encrypted versions and said encryption keys and said key sequence to one or more receivers or receiver entities, e.g. as one or more data files,

for decrypting at said receiver site said pre-watermarked and encrypted versions of audio or video data using said en ^¬ cryption keys and said blocks or frames individually using said key sequence and assembling, according to the values of a desired watermark information word, corresponding frames or blocks from said decrypted and pre-watermarked versions in a successive manner, so as to provide for presentation a watermarked version of said original audio or video data that carries said watermark information word.

In principle, the inventive receiver system is suited for protection by watermarking against non-authorised use, e.g. non-authorised recording or copying, of an original audio or video data signal that is to be presented, said receiver system receiving or storing at least two differently pre- water-marked versions for successive blocks or frames of said original audio or video data, wherein said versions were derived by applying a repeated watermark symbol value to a version and different watermark symbol values to the different versions, and by encrypting said versions using corresponding encryption keys, wherein in addition each block or frame of said differently pre-watermarked versions is encrypted individually using a corresponding key se- quence, said receiver system also receiving or storing said encryption keys and said key sequence, and said receiver system including:

means being adapted for decrypting said pre-watermarked and encrypted versions of audio or video data using said en ^¬ cryption keys and said blocks or frames individually using said key sequence and assembling, according to the values of a desired watermark information word, corresponding frames or blocks from said decrypted and pre-watermarked versions in a successive manner, so as to provide a watermarked ver ^¬ sion of said original audio or video data that carries said watermark information word;

means being adapted for presenting said decrypted and wa ^¬ termarked version of said original audio or video data.

Advantageous additional embodiments of the invention are disclosed in the respective dependent claims.

Drawings

Exemplary embodiments of the invention are described with reference to the accompanying drawings, which show in:

Fig. 1 Distribution workflow for a Digital cinema application from packaging to pro ection;

Fig. 2 Application of watermarking technology in the Digital cinema;

Fig. 3 Preparation of pre-watermarked and encrypted copies; Fig. 4 Constructing a watermarked audio track;

Fig. 5 Preparation of pre-watermarked and block-wise en ^¬ crypted copies;

Fig. 6 Constructing a watermarked track from block-wise decrypted copies;

Fig. 7 Pre-watermarked and encrypted copies with variable part ;

Fig. 8 Pre-watermarked and encrypted copies with difference signals .

Exemplary embodiments

Pre-watermarking and encryption

According to the invention, several block-wise pre-watermarked tracks for one audio PCM file are delivered. All pre- watermarked tracks are encrypted and stored in a file. At receiver site the pre-watermarked tracks are decrypted and, depending on the watermark information word (e.g. ID of the theatre and the time-stamp) , the watermarked audio stream is constructed for the presentation.

The inventive processing is portable to other watermarking systems if it conforms to the following criteria:

• The underlying watermarking system operates in a block- based manner. In fact, most watermarking systems divide the original signal into disjoint regions, either in space or in time. In audio watermarking systems, the audio track is split into a temporal sequence of consecutive audio blocks .

• Embedding of the watermark in one block does not depend on the content of adjacent blocks even if an overlap occurs between the different blocks. At least, such dependency can be ignored due to an insignificant decrease of the quality of the watermarked track and the robustness of the watermark .

Besides the above-mentioned characteristics, in the special case of Adaptive Spread Phase Modulation (ASPM) processing the following can be assumed:

• The N information bits of a watermark w are represented by a sequence of separate symbols aj_, i=l,...,N, and each symbol is embedded into one block consisting of Ν _β sub- blocks;

• In the watermarking system, each sub-block contains the same number of samples determined by usage of a psycho- acoustic model. For example, the size 1 of a sub-block is 1 (c ₀ ^SB ) = 1024. Before the audio tracks are delivered, the original file data c _or g is watermarked times with special watermarks

resulting in watermarked copies. Fig. 3

shows versus time t two versions eg, C]_ of the same data file, and a version C5 with synchronisation symbols embed- ded, each version of which is watermarked (and encrypted) differently. Each one of the watermarked track copies is represented by a temporal sequence of identical watermark information bit patterns out of

. For example, in file version eg the bit pattern '000...00' is used to generate watermark symbol wg for the time periods to t _n , and in file version C]_ the bit pattern '111...11' is used to generate watermark symbol W]_ . Such data preparation is referred to as pre- watermarking . The embedding of these watermarks is performed by using a unique key K for generating the bit pattern representing the different bits or sync words ' 0 ' , ' 1 ' , ' S _] _ ' , ' S2 ' , ' S3 ' . In the special case of ASPM processing being used, three different sequential synchronisation sequences ( ' S _] _ ' , ' S2 ' , ' S3 ' ) have to be inserted at the beginning of the watermarks wg / W]_ as shown in Fig. 3 in version C5 of the data file, in order to ensure success of the detection against other signal processing operations. Since the length of the watermark information word is _Z(w)=35 bits in the ex ^¬ ample application, the synchronisation bits have a fixed block position and need not be stored as a separate pre- watermarked copy. This reduces the number of necessary pre- watermarked copies to a number lower than if all water

marked blocks are counted, as shown in Fig. 3.

In order to increase the security, the watermark keys can be different for the individual audio files. However, this re- suits in an identification problem during watermark detection. But this problem can be solved by incorporating fingerprinting techniques.

The watermarked copies of the track are encrypted e.g. as a whole by with the encryption function

using key K _E (cf. Fig. 3) . The resulting encrypted

files are packed and delivered or transferred to the re ^¬ ceiver (theatre) . In the Digital cinema application the pre- watermarked encrypted files are stored in the media server in the cinema.

Before presenting an audio track at receiver site the pre- watermarked copies are decrypted with

the decryption function using the same key K _E . The em-

bedding of the 1 (w) watermark information bits in an audio track is performed by constructing block or frame-wise the watermarked copy from the pre-watermarked copies. The whole number of blocks is np = + 1 (w) , with _Z(w)=35 watermarked blocks and -"5=3 synchronisation blocks.

Figure 4 shows an example of embedding a watermark for bit sequence '10...11' by constructing in a constructor 40 the required track c _w . The upper part of Fig. 4 corresponds to Fig. 3. According to a current bit value 'i' in the water ^¬ mark bit sequence w = '10...11', pre-watermarked section ^c w (t _x ) is taken and assembled in a successive manner by the constructor 40 from pre-watermarked tracks eg or C]_ .

The content of the watermark information word 'w' can be determined individually by the present receiver.

Advantageously, the construction speed of the watermarked track depends on the I/O performance only and not on the kind of watermarking processing being used. The watermarking performed in the data preparation step is the same for all receivers (cinemas) and therefore has to be performed only once for an audio track. Block-wise encryption

In the embodiment described before, it could be possible to construct watermarked tracks carrying the wrong watermark (i.e. cinema ID) . The following embodiment modifies the pre- watermarking and construction processing.

The watermarking performed to create the pre-watermarked copies is identical to the previous embodiment. However, the watermarked copies are encrypted block-wise.

Each block is en-

crypted by using a different key K for the block starting at position tj in the watermarked track and for symbol type i, as depicted in Fig. 5. The block-wise encrypted files are packed and delivered, e.g. to the theatre where they are stored in the media server. In addition to the media files, a key sequence 1 ^S transmitted to each receiver,

consisting of the keys required for decrypting only those blocks which are carrying the bits for constructing the wa- termark.

For Digital cinema application, the watermark w = { w _ID , w _time } consists of the ID of the cinema and the time stamp. The key sequence can be partitioned into a fixed part consisting of the synchronisation blocks and the ID of the cinema , and

a variable part for the time stamp

consists of 35 bits for the DCI watermark and 3 synchronisa- tion bits.

Different time stamps can be implemented by varying the last sub-sequences K _time . This offers the possibility to limit the time stamps to a certain period by transmitting only the corresponding key sequences.

By using the key sequence the receiver (theatre) will be able to properly decrypt only the blocks having the watermarks w _ID and w _tiine embedded, by getting the block at po ^¬ sition j from encrypted block

as depicted in Fig. 6 (the upper part corresponds to Fig . 5 ) .

In addition to the previous embodiment, this embodiment has the advantage that the correct watermark (and thereby the ID of the cinema) is implicitly embedded because only the proper blocks can be decrypted by the assigned key sequence. But there are remaining disadvantages:

• In the Digital cinema application the embedding of the correct time stamp cannot be ensured by this processing. An potential adversary can embed an incorrect time stamp by using already decrypted versions of the blocks having different symbols embedded. Nevertheless, watermarked cop ^¬ ies can be constructed with a time stamp only within a range according to the transmitted key sub-sequences K _time . An incorrect time stamp can be detected due to the knowl- edge of the transmitted key sequences.

• The bandwidth is nearly doubled since it is still neces ^¬ sary to send two pre-watermarked and block-wise encrypted copies . Reducing the bandwidth

A reduction of bandwidth can be achieved by taking into account the static and the dynamic part of the watermark. This can be performed by sending the pre-watermarked and block- wise encrypted part of the stream directly with the fixed watermark portion carrying synchronisation bits and the ID of the watermark. Only for the variable part the two ver ^¬ sions of the pre-watermarked and encrypted blocks are sent to the receiver (theatre) . The time stamp range can be con ^¬ trolled at sender site.

The watermark embedding is the same as presented in the pre- vious embodiment, whereby selecting the blocks for decryp ^¬ tion has to be done only for the variable part w _tiine of the watermark. This is depicted in Fig. 7. A further reduction of the necessary bandwidth can be achieved by sending for the variable portion w _tiine of the wa ^¬ termark a track marked with one watermark symbol and the difference of the watermark signals for different watermark symbols. As depicted in Fig. 8, for each block at position tj a watermarked version for the first symbol '0' is

embedded, as well as the difference to the watermark

signals of all other symbols:

Because the watermark difference signals have a lower dy ^¬ namic range than the original watermark signal they can be compressed effectively.

For embedding the watermark by construction, in addition to the selection of the blocks from the variable watermark portion w _tiine as presented in the previous embodiment a block- wise addition of the difference signals according to equa ^¬ tion (3) is performed if a symbol other than that of the pre-watermarked copy is to be embedded.

As mentioned earlier, the invention is applicable to Digital cinema applications. It is further applicable to all appli ^¬ cations where a watermark embedding step has to be performed at receiver site in an insecure device.