FR2897217A1 - DETERMINATION OF SIGNAL IDENTIFIER - Google Patents

Abstract

A system determines an identifier (ID) of a sound document (SG) signal that is robust to alterations such as compression and time shift of the signal. Repeatedly, a first segment of the digital signal is divided into second segments, average amplitudes of the second segments are compared to select a second segment, a characteristic segment (SCm) including the second selected segment is extracted, a sub-identifier (IDm) of the signal is determined as a function of the characteristic segment, and in the signal another first segment abutting the previously extracted characteristic segment is selected. The sub-identifiers determined according to the characteristic segments are aggregated into the identifier (ID) of the signal.

Description

Signal identifier determination

  The present invention relates to the field of signal processing applied to the identification of a signal. More particularly, it relates to determining an identifier of a signal corresponding to an audio document for indexing and recognizing the audio document.

  Download speeds and the growing number of files available for exchange, especially audio documents, on peer-to-peer (P2P) networks have contributed to rapid and large-scale dissemination. unlawful copies of documents without regard to the copyrights applied to those documents. Copyright protection techniques, such as Digital Rights Management (DRM) digital rights management, are used to counteract unlawful copies of multimedia documents. However, documents already exchanged and copied before the implementation of these techniques are forever lost for their beneficiaries. In addition, DRM techniques are inefficient at digital-analog-digital recordings of documents to overcome the protections placed on the media, the document can then be exchanged freely. To overcome these problems, alternative solutions recognize multimedia documents by determining a signal identifier corresponding to the document.

  According to a first known embodiment, the determination of an identifier of an audio signal

  is based on a transformation of the signal by pyramidal wavelet decomposition, separating the events of the signal which take place in different frequency bands. The identifier of the signal is then composed of the coefficients of the decomposition.

  According to other known embodiments, the determination of the identifier of a signal is based on a frame of the signal ("framing" in English) to separate the audio signal in intervals of a few hundreds of milliseconds overlapping and apply on each interval a Hamming window. The signal is thus considered as stationary in each interval. Signal characteristics are extracted from each interval to form an identifier of the signal. According to one of these known embodiments, the identifier of the signal is determined by applying a Fourier transform in each interval and the resulting Fourier transform is then applied to a bank of band-pass filters, for example between 300 Hz and 2 kHz, to extract a binary word whose bits are determined according to energies of the frequency bands of the following interval. The identifier is composed of successive binary words obtained by successive processing of the intervals. Another known embodiment is a discriminating distortion analysis DDA ("Distortion Discriminant Analysis" in English). The signal identifier is determined by successive applications of principal oriented analysis (OPCA) on each interval to obtain the resulting vectors. The signal identifier has components of the vectors.

  However, an identifier of an audio document determined from an original audio signal according to the previous known embodiments is different from the identifier determined once the audio signal has undergone alterations such as signal compression, deleting part of the signal or more particularly a time shift of the signal. For all these alterations, the identifier determined from an altered signal is different from the identifier determined from the original signal.

  The object of the invention is to overcome the aforementioned drawbacks by determining an audio document identifier that is robust to the alterations mentioned above, more particularly to the time shift of the signal so that the recognition of an original signal in the signal of altered origin is more likely.

  To achieve this objective, a method for determining an identifier of a segment division signal is characterized in that it comprises: recurrently, a division of a first segment of the signal into second segments, a comparison of middle amplitudes of the second segments to select a second segment, a determination of a sub-identifier of the signal based on the second selected segment, and in the signal a selection of another first segment butted to the second selected segment, and an aggregation sub-identifiers determined according to the second segments selected in the signal identifier.

  In order to limit the number of sub-identifiers making up the signal identifier, the second selected segment may be replaced by a characteristic segment including at its center the second selected segment. The segmentations of the signal according to the invention have the advantage of detecting segments having a characteristic average amplitude, such as a maximum or minimum average amplitude, included in the signal, and thus of synchronizing the altered signal with the original signal so as to determining an altered signal identifier very similar to the original signal identifier also determined by the method. Thus, if the original signal is shifted for example due to truncation of a part of the signal, this alteration is not a problem to determine an identifier of the signal since the determination is based on characteristic segments of the signal depending on the temporal pace of the signal and therefore significant moments of the signal a priori variables, and not framing ("framing" in English) frozen time signal. The determination of sub-identifiers of the signal according to characteristic segments amounts to synchronizing the altered signal with the original signal. The method of determining the signal identifier according to the invention is thus particularly robust to the compression of the original signal. According to another characteristic of the invention, the determination of a sub-identifier as a function of a second selected segment comprises: a frequency transformation of the second selected segment into a frequency transform,

  filtering the frequency transform in contiguous bandpass filters to produce elementary frequency amplitudes in each of the frequency bands, estimating an energy of each frequency band as a function of the amplitudes of the elementary frequencies in the frequency band; frequency, and a comparison of energies two by two so as to encode signs of energy differences and concatenate the signs coded into the sub-identifier. According to another characteristic of the invention, the comparison of the energies comprises a successive determination of the averages of the energies and a comparison of the averages of the energies two by two so as to code the signs of differences in energy means and concatenate the signs encoded in the sub-identifier. This subidentifier determination based on the selected second segments or the characteristic segments of the signal contributes to reducing the size of the signal identifier.

  The invention also relates to a system for determining an identifier of a signal by division into segments. The system is characterized in that it comprises: recursively, means for dividing a first segment of the signal into second segments, means for comparing average amplitudes of the second segments to select a second segment, means for determining a sub-identifier of the signal according to the second selected segment, and a means for selecting

  in the signal another first segment abutting the second selected segment, and means for aggregating sub-identifiers determined according to the second selected segments into the signal identifier.

  Finally, the invention relates to a computer program adapted to be implemented in a processor, such as a system for determining an identifier of a digital signal, said program comprising instructions which, when the program is loaded. and executed in said processor, perform the steps according to the method of the invention.

  Other features and advantages of the present invention will appear more clearly on reading the following description of several preferred embodiments of the invention, given by way of non-limiting examples, with reference to the corresponding appended drawings in which: FIG. 1 is a schematic block diagram of a signal identifier determination system according to the invention; FIG. 2 is an algorithm of a signal identifier determination method according to the invention; FIG. 3 is a detailed algorithm of a step of signal identifier segment identifier determination according to the invention; and FIG. 4 is a graph of a segmented audio document signal according to the invention.

  Although the invention is applicable to various signals, whatever their contents, the invention is described hereinafter for the identification of a

  digital audio signal according to a preferred embodiment. This realization is directed to determining the identifier of an altered digital signal near the identifier of an original digital signal. The identifier of the original digital signal and the identifier of the altered digital signal are successively determined according to the method of the invention. The alterations that a digital signal undergoes are, among others, the partial or total compression of the signal, the suppression of one or more parts of the signal causing a temporal shift of the signal. For example, a musical concert is recorded in the form of an original digital signal, the applause at the beginning of the concert is removed there which induces a time shift of the signal into an altered signal different from the original signal.

  Referring to Figure 1, an SDI digital signal identifier determining system according to the invention comprises a segmentation module MS for segmenting a digital signal SG entering the system to obtain SC 'to SCM characteristic segments of the system. signal SG, and a UDI identifier determining unit for determining an identifier ID of the digital signal SG according to the characteristic segments SC 'to SCM. The SDI system is represented in FIG. 1 in the form of functional blocks, most of which provide functions relating to the invention and can correspond to software and / or hardware modules. For example the SDI system is a computer and incorporates or can access locally or through a telecommunications network to a base

  data of the type used in artificial intelligence.

  The UDI unit determining the identifier ID of the signal SG comprises a frequency transformation module MTF, N parallel bandpass filters FI1 to FIN, an energy determination module MDE and an identifier construction module MCI. The MTF module frequency transforms each characteristic segment SCm of the signal SG produced by the segmentation module MS into a frequency transform TFm, with the index m between 1 and an integer M. For example, the module MTF implements a transformation of Discrete Fourier TFD which makes less sensitive the subsequent processing of the characteristic segments to alterations of the original signal. The N parallel bandpass filters FI1 to FTN filter the frequency transform TFm into N contiguous frequency bands BF1 to BFN. The module MDE determines a power spectral density Em, n of the segment SCm in each frequency band BFn, with the index n between 1 and N. The module MCI constructs the identifier ID of the signal SG by determining N-1 signs. Sm, 1 to Sm, N-1 as a function of the previously determined power spectral densities Em, 1 to Em, N to constitute an IDm subidentifier relative to the SCm segment. The identifier ID is aggregated into identifiers ID1 to IDM determined respectively from the segments SC 'to SCM produced by the segmentation module MS. The role of the different modules is described in more detail with reference to FIGS. 2 and 3.

  The method for determining the identifier ID of the signal SG comprises three main steps E1 to E3 shown in FIG. 2. The step E1 executed by the segmentation module MS segments the incoming digital signal SG in order to detect the signal SG of the segments. characteristics SC 'to SCM. Steps E2 and E3 are executed by the UDI unit determining an identifier ID of the SG signal.

  The step E2 is recurrent with the step E1 and determines an IDm subidentifier of each characteristic segment SCm contained in the signal SG. Step E3 aggregates the IDI IDM sub-identifiers of the characteristic SC 'to SCM segments of the SG signal into the identifier ID.

  Step E1 comprises steps E10 to E13 for the detection of characteristic segments of the signal SG. In step E10 and with reference to FIG. 4, the segmentation module MS selects in the digital signal SG a first digital segment, called the observation segment SOI, whose start corresponds to the beginning of the signal SG and whose duration is predefined and fixed, for example equal to 100 ms. The module MS divides the observation segment SO1 into second very short digital segments, called segments of average amplitude SA 'to SA1, having a fixed duration for example of 10 ms. The average amplitude segments SA 'to SA' overlap two by two with a time offset of, for example, 1 ms. For each segment SAi, with the index i between 1 and I, the segmentation module determines the average amplitude AMi of the segment SAi in step E11 and stores it. Once the average amplitudes AMI to AMI respectively of all segments SA 'to SA' are

  determined, in step E12, the segmentation module MS searches, in step E13, among all the segments of average amplitude SA1 to SA1, a segment SAmax whose average amplitude AMmax is the most characteristic, by comparing the average amplitudes of the second segments SA1 to SA1. Preferably, the segmentation module MS seeks a segment SAmax whose average amplitude AMmax is the maximum, to which reference will be made later in the description. According to other examples, the segmentation module MS seeks a segment whose average amplitude is minimal or whose average amplitude is the second maximum amplitude. The overlap of the segments SA 'to SA' two by two has the advantage of a better accuracy of the characteristic average amplitude AMmax, the latter is considered at the center of the SAmax segment. In a variant, the average amplitude segments do not overlap.

  The segmentation module MS defines a third digital segment, called the characteristic segment SC1, which includes in its center the segment SAmax of maximum average amplitude and which therefore has a duration greater than or possibly equal to that of the segment SAmax, for example a few tens milliseconds. The segmentation module MS extracts the characteristic segment SC 'including the segment SAmax in order to transmit it to the identification unit UDI.

  In step E2, the UDI identifier determining unit determines a sub-identifier ID1 according to the previously defined characteristic segment SC 'and stores it. As long as the SG signal is not fully processed by the MS segmentation module at step

  E30, the steps E10 to E13 and the step E2 are repeated by selecting in step E10 another observation segment SOm, with the index m lying between 2 and M, SOM being the last observation segment of the signal SG. As shown in FIG. 4, the beginning of the SOm segment begins at the end of the preceding characteristic segment SCm_1 in order to define another characteristic segment SCm by executing the steps E11 to E13 described above.

  The selection of the observation segments which can overlap in pairs gives the advantage of a more precise selection of characteristic average amplitudes which are temporally close in the signal S. For example, if the signal S contains two close characteristic average amplitudes, of which the first is more relevant than the second, for example the first is maximum or minimum, and which are included in the same observation segment SOm_1, only the first characteristic average amplitude is selected. Following the abutment of an observation segment following SOm to the characteristic segment including the first characteristic average amplitude selected, the second characteristic average amplitude is also selected. In one variant, the observation segments SOm and SOm_1 do not overlap, and the segment SOm ends up with the preceding segment SOm_I.

  An IDm sub-identifier relating to the characteristic segment SCm is determined in step E2.

  With reference to FIG. 3, the main step E2 determines the identifier IDm of each characteristic segment SCm defined and extracted by the module

  segmentation MS at step E1, with the index m between 1 and M. Step E2 comprises steps E20 to E27 executed by the UDI identifier determining unit.

  In step E20, the frequency conversion module MTF of the determination unit UDI applies a discrete Fourier transform TFD to the characteristic segment SCm of the signal SG to transform it into a frequency transform TFm in order to study the frequency behavior of the segment SCm characteristic. In step E21, the frequency transform TFm of the segment SCm is applied to the bank FT of N bandpass filters FT1 to FTN for filtering the frequency transform TFm of the segment SCm into N frequency transforms TFm, 1 to TFm, N respectively in the N contiguous frequency bands BF1 to BFN in order to analyze in detail the frequency behavior of the SCm segment. Each bandpass filter FTn has a low start cutoff frequency Fdn equal to the high end cutoff frequency Ffn_1 of the previous bandpass filter FTn_1 and a high end cutoff frequency Ffn equal to the low start cutoff frequency. Fdn + 1 of the bandpass filter according to FTn + 1, with the index n lying between 1 and N. For example, the widths of the contiguous frequency bands BF1 to BFN increase in proportion to a factor equal to 21/9 -1 or Ffn = Fdn x 21/9 The set of frequency bands BF1 to BFN preferably extends over an audible frequency band, for example between 400 Hz and 5 kHz. Each frequency transformation TFm, n provided by each bandpass filter FTn is described by the spectral amplitudes ITFm, nlfn of elementary frequencies fn equidistributed in the frequency band BFn of the filter FTn and between the cutoff frequencies Fdn and Ffn. In step E22, the module MDE determines acoustic characteristics of the segment SCm as the power spectral density of the segment SCm, hereinafter referred to as "energy" Em, n, for the frequency band BFn of each filter FTn. Each energy Em, n is estimated as a function of the sum of the amplitudes ITFm, nFdn to ITFm, nFfn of the elementary frequencies fn included in the frequency band BFn of each filter FTn. The module MDE determines the energy Em, n by the following relation: fn = Ffn 2 11TFm, n fn = Fdn fn Ffn - Fdn At the end of the step E22, N energies Em, 1 to Em, N are estimated for a characteristic segment SCm. In steps E23 to E26, the identifier building module MCI "compresses" the information represented by the N energies Em, 1 to Em, N so that the IDm sub-identifier of the segment SCm to be constructed characterizes the segment SCm. The information represented by the N energies is "compressed" by coding the N-1 sign differences between successive averages of energies Em, 1 to Em, N previously estimated. In step E23, the MCI module successively determines the averages of the energies Em, 1 and Em, 2,... Em, 1 to Em, n, _ Em, 1 to Em, N and progressively compares these averages of energies two by two for the index n increasing from 1 to N-1. For the sign Sm, n associated with the frequency band BFn of the filter FTn with n between 1 and N-1, if the average of the energies Em, 1 to Em, n respectively relating to Em, n

  frequency bands BF1 to BFn, is greater than the average of the energies Em, 1 to Em, n + 1 relative respectively to the frequency bands BF1 to BF11 + 1, whereas the sign Sm, n is positive and represented by a bit at 1 state "1" in step E24; otherwise it is negative and represented by a bit in state "0" in step E25. As a variant, the coding of the N-1 sign differences is established by successive comparisons of the energies Em, 1 to Em, N, without determining the successive averages of the energies Em, 1 to Em, N. Once all the signs of energy differences Sm, 1 to Sm, N-1 have been coded in step E26, the identifier construction module MCI concatenates the coded signs Sm, 1 to Sm, N-1. to obtain the IDm subidentifier of the segment SCm, in step E27. Then, the IDm sub-identifier is memorized.

  Returning to FIG. 2, step E3 concerns the construction of the unique identifier ID representative of the signal SG and comprises steps E30 and E31. In step E30, the MCI identifier construction module in the UDI unit verifies that the SG signal has been entirely processed by abutting iteration of an observation segment SOm + 1 to a characteristic segment SCm previously extracted. of the signal SG, as shown in FIG. 4, and that all the ID1 to IDM sub-identifiers of all the characteristic segments SC 'to SCM contained in the signal SG are determined, the steps E1 and E2 being repeated M times. Then in step E31, the module MCI concatenates the sub-identifiers ID1 to IDM in order to obtain the unique identifier ID of the digital signal SG of size M x (N-1) bits, M being the

  variable number of characteristic segments detected in the digital signal SG in step El.

  According to a first variant, the module MCI completes the identifier ID of the signal SG by determining the chronological positions of the second segments of maximum amplitude SAmax and thus of the characteristic segments SCm in the signal SG. According to a second variant, the module MCI adds additional information to the identifier ID in the signal SG by determining the time differences between the second segments of maximum amplitude SAmax and therefore between the characteristic segments SCm.

  An exemplary application of the invention relates to the identification of an altered digital signal corresponding to an audio document according to a list of original digital signal identifiers corresponding to original, unaltered audio documents, said identifiers being pre-determined according to the method of the invention and indexed in a database in connection with the SDI digital signal identifier determination system. The identifier of the altered signal is first determined according to the method of the invention. Then, each sub-identifier relating to a maximum amplitude segment of the altered signal is compared to each sub-identifier of the identifiers of the list, according to a predetermined approximation. For this purpose, the MCI module evaluates, for example, similarities between the identifier of the altered signal and the identifiers of the original digital signals, expressed respectively by scores.

  The highest score reflects the likelihood that the altered signal results from an alteration of one of the original digital signals associated with the highest score and therefore is all or part of the original digital signal associated with the highest score. Student. According to the first variant, the search for the highest score is also based on the similarity between the chronology of the sub-identifiers of the altered signal and the sub-identifiers of the original digital signals, that is to say on the similarity chronological positions of the maximum amplitude segments of the altered signal and the original digital signals. According to the second variant, the search for the highest score is also based on the similarity between the time differences relative to the maximum amplitude segments of the altered signal and the original digital signals.

  The invention described herein relates to a method and system for determining a signal identifier. According to a preferred implementation, the steps of the method of the invention are determined by the instructions of a computer program incorporated in the SDI system. The program comprises program instructions which, when said program is loaded and executed in the SDI system whose operation is then controlled by the execution of the program, perform the steps of the method according to the invention. Accordingly, the invention also applies to a computer program, including a computer program on or in an information carrier, adapted to implement the invention. This program can use any language of

  programming, and be in the form of source code, object code, or intermediate code between source code and object code such as in a partially compiled form, or in any other form desirable for implementing the method according to the invention. The information carrier may be any entity or device capable of storing the program. For example, the medium may comprise storage means or recording medium on which is stored the computer program according to the invention, such as a ROM, for example a CD ROM or a microelectronic circuit ROM, or a USB key, or a magnetic recording medium, for example a diskette (floppy disc) or a hard disk. On the other hand, the information medium may be a transmissible medium such as an electrical or optical signal, which may be conveyed via an electrical or optical cable, by radio or by other means. The program according to the invention can in particular be downloaded to an Internet type network. Alternatively, the information carrier may be an integrated circuit in which the program is incorporated, the circuit being adapted to execute or to be used in the execution of the method according to the invention.

Claims (10)

  1 - Method for determining an identifier (ID) of a signal (SG) by division into segments, characterized in that it comprises: recurrently, a division (E10) of a first segment (SOm) of the signal in second segments (SA1-SA1), a comparison (E13) of mean amplitudes (AMI-AMI) of the second segments to select a second segment (SAmax), a determination (E2) of a subidentifier (IDm) of the signal as a function of the second selected segment, and in the signal a selection (E30) of another first segment abutting the second selected segment, and an aggregation (E31) of the determined sub-identifiers as a function of the second selected segments, in accordance with identifier (ID) of the signal.   2 - Process according to claim 1, wherein the second selected segment (SAmax) is replaced by a characteristic segment (SCm) including in its center the second selected segment.   3 - Process according to claim 1 or 2, wherein the second segments (SA1-SA1) of a first segment (SOm) overlap two by two.   4 - Process according to any one of claims 1 to 3, wherein the determination of a sub-identifier (IDm) according to a second selected segment (SAmax) comprises: a frequency transformation (E20) of the second segment selected in a frequency transform (TFm), a filtering (E21) of the frequency transform in filters (FI1, FIN) with contiguous bandwidths (BF1, BFN) in order to produce elementary frequency amplitudes (fn) in each (BFn) ) frequency bands, an estimate (E22) of an energy (Em, n) of each frequency band (BFn) as a function of the amplitudes of the elementary frequencies in the frequency band, and a comparison (E23) of the energies two in pairs so as to encode (E24, E25) signs (Sm, i, Sm1N-1) of energy differences and concatenate (E27) the signs coded as the sub-identifier (IDm).   5 - Process according to claim 4, wherein the comparison of the energies comprises a successive determination of the averages of the energies and a comparison of the averages energies two by two so as to code (E24, E25) the signs (Sm, i, Sm1N -1) of differences in energy averages and concatenate (E27) the signs coded in the sub-identifier (IDm).   6 - Process according to any one of claims 1 to 5, comprising a determination of chronological positions of the second selected segments (SAmax) in the signal.   7 - Process according to any one of claims 1 to 6, comprising a determination of temporal gaps between the second selected segments (SAmax) in the signal.   8 - System for determining an identifier (ID) of a signal (SG) by division into segments, characterized in that it comprises: recursively, means (MS) for dividing a first segment (SOm) second segments ( SAI-SAI), a comparison of the average amplitudes second segments in order to segment (SAmax), a means of the signal in means (MS) for (AMI-AMI) to select a second (UDI) to determine a sub-identifier (IDm ) of the second selected segment, signal according to the and means (MS) for selecting in the signal another first segment abutting the second selected segment, and means (UDI) for aggregating sub-identifiers determined according to the second selected segments , in the identifier (ID) of the signal.   9 - computer program comprising instructions for implementing a method according to any one of claims 1 to 7 when said instructions are executed by a processor.   10 - Recording medium on which is stored a computer program according to claim 9.