T 1118/10 (Recognition of audio broadcast segments/ARBITRON) 30-11-2012

Reasons for the decision

1. First auxiliary request - Article 76(1) EPC

1.1 In accordance with Article 76(1) EPC, "[a European divisional application] may be filed only in respect of subject-matter which does not extend beyond the content of the earlier application as filed". In the present case the earlier application is the parent application as referred to at point II above. The established criterion, analogous to Article 123(2) EPC, is that the subject-matter of a divisional application must be directly and unambiguously derivable from the parent application. This is the criterion adopted by the board in the present case.

1.2 Claim 1 includes the feature "wherein the comparisons comprise producing difference values representing differences between the first and second frequency band values and comparing the difference values to zero".

This feature was said by Appellant II to be based on the description of the parent application on pages 43 and 44, the relevant part of which reads:

"The signature generation module 312 utilizes a NOW-THEN processing technique to produce sixteen-bit audio signatures such that each signature bit is obtained based on a current value (or NOW value) of a corresponding frequency band and a previously obtained value (or THEN value) of the same frequency band produced from a frame preceding the current frame by a predetermined frame offset.

The signature generation module 312 produces a value DVAL for each frequency band in accordance with the following relation:

DVAL=(NOW-THEN)/(NOW+THEN)

The value of each of the 16 bits in the audio signature for the current frame and the bit values of corresponding mask word are determined in accordance with the value DVAL. That is, a signature bit is set to 0 if DVAL for the corresponding band is greater than 0, otherwise it is set to a value of 1."

1.3 Claim 1 includes the step of forming a difference "NOW-THEN" and comparing it to zero, but omits any division by the value "NOW+THEN". Claim 1 is therefore at variance with the description to the extent that a feature has been omitted from an originally disclosed feature combination, leading to a so-called "intermediate generalisation".

1.4 Appellant II argued that the so-called "essentiality test" should be applied as set out in T 331/87. This decision sets out various tests to determine whether a claim may be broadened by omission of an inessential feature. In the present case, it was argued that the skilled person would recognise that division by the value NOW+THEN, which Appellant II referred to as "normalisation", was not essential to the signature formation and clearly only had relevance to the forming of the mask word, for the reason that the ensuing comparison with zero would produce the same result with or without normalisation.

1.5 However, with regard to intermediate generalisations, the approach followed by the boards is that such amendments are only justified in the absence of any clearly recognisable functional or structural relationship between the features presented in combination (cf. eg T 25/03, not published). In the present case, in the board's view, when elements are presented in the same formula they are prima facie related. The skilled person who is only presented with the instruction "The signature generation module .. produces a value DVAL for each frequency band ... DVAL = (NOW-THEN)/(NOW+THEN)" would not conclude, even implicitly, that division by "NOW+THEN" was an optional step. In particular he would not conclude from the information given in the description that an embodiment was envisaged in which both the mask word and the division step could be dispensed with; this would be a new embodiment developed by the skilled person.

1.6 The board concludes that claim 1 of the first auxiliary request includes an intermediate generalisation not directly and unambiguously disclosed in the parent application as filed, contrary to Article 76(1) EPC.

2. Second auxiliary request

2.1 Article 76(1) EPC

Claim 1 is based on claim 17 of the parent application combined with details taken from the description on page 43, line 15 to page 44, line 16. Claim 1 therefore complies with Article 76(1) EPC. Appellant I did not argue otherwise.

2.2 Articles 123(2) and (3) EPC

Claim 1 is based on original claim 1 together with paragraphs [0111] and [0112] of the published application. Further, the scope of claim 1 is more limited than that of claim 1 as granted. It therefore complies with Articles 123(2) and (3) EPC, which was not contested by Appellant I.

2.3 Article 84 EPC

Neither the board nor Appellant I saw any reason to object to claim 1 on the ground of lack of clarity (Article 84 EPC).

2.4 Novelty and inventive step (Article 52(1) EPC)

2.4.1 Background

The present invention relates to the recognition of audio signal "segments" in a broadcast audio signal. More particularly, it concerns the formation of a signature, ie a digital signal (in general, highly compressed) derived from and representative of the audio signal segment.

2.4.2 Closest prior art

There was agreement at the oral proceedings that document E1 represents the closest prior art in respect of the subject-matter of claim 1 of the second auxiliary request.

E1 discloses a method for forming a signature primarily from a broadcast video signal, although it is also mentioned that the broadcast signal can be a radio broadcast for which audio information is used for the recognition process (cf. col. 3, lines 41-43).

E1 only describes in any detail how to form a signature from a video signal. The video signature is formed from luminance values of particular areas of the video field or frame. An average value is calculated for each area. This value is "normalized to a bit value of 0 or 1", by, in one embodiment, "comparing the value to the average luminance of the same area in some previous field or frame" (col. 5, lines 18-25 and 29-30).

E1 further discloses the generation of a "mask word". As disclosed in col. 6, line 2 ff.,

"The mask word represents the reliability of the frame signature. For each bit in the frame signature, if the absolute value of the luminance difference used to calculate that bit value is less than a threshold or "guard band" value, the luminance value is assumed to have been susceptible to error (because of noise in transmission), and so the mask bit for that bit is set to 0, indicating suspect data. If the absolute value of the luminance difference is greater than or equal to the guard band value, the luminance value is assumed to be much greater than the noise level and the corresponding mask bit is set to 1, indicating reliable data."

2.4.3 It was not in dispute that the subject-matter of claim 1 of this request was new, either with respect to E1 or to any other document on file (Article 54 EPC).

2.4.4 Starting out from the method disclosed in E1 for creating a signature and a mask word for a video signal, with a hint towards forming an audio signature, Appellant I argued that the problem to be solved by the skilled person was to create a signature for a broadcast audio signal. The board agrees.

2.4.5 In order to solve this problem, Appellant I argued that the skilled person would arrive at the subject-matter of claim 1 in the following manner:

(i) E1 teaches that an audio signature should be derived in the same way as for a video signal. In this respect, Appellant I referred to col. 6, lines 17-21, which reads:

"An audio "'frame' signature", if such a signature is used, can be constructed in the same format as a video frame signature so that it can be processed in the same way".

(ii) E1 teaches that instead of luminance values, the audio signature should be based on the frequency spectrum (cf. claim 11). It is well-known to the skilled person, see eg document E2, to represent the frequency spectrum in terms of a spectral analysis to determine frequency band values. Analogous to the method for producing a video signature based on the difference between a current and a previous frame, it would be obvious to form an audio signature based on the difference between frequency band values representing a current segment and corresponding values of a previous segment.

(iii) The division by the sum value required by claim 1, ie normalisation, is a routine measure for the person skilled in the art.

In consequence, Appellant I argued that the subject-matter of claim 1 did not involve an inventive step.

2.4.6 The board however finds the above arguments unconvincing for the following reasons:

Re (i): The expression "in the same format" appears to the board to refer to the format of the signature and/or mask word rather than the method of their production, ie this passage plausibly means that these signatures should each have the same structure in order to be subsequently processed (by being compared to a database of stored signatures) in the same way.

Re (ii): Claims 11 and 12 of E1 read as follows:

"11. The method of claim 3 wherein said signatures and said parametized signals are derived at least in part from the frequency spectrum of said audio portion.

12. The method of claim 11 wherein said signature and said parametized signals each comprise a plurality of digital words, each digital word of said signatures and said paramatized [sic] signals being derived by comparing selected frequency bands of said audio portion to at least one reference band, each of said selected bands providing a bit of said digital word."

These claims, which are the only part of the disclosure containing any detail concerning the generation of the audio signature, thus teach comparing frequency bands to a reference band or bands, with no suggestion that the reference band should be associated with a previous segment, or might be the corresponding frequency band value of a previous segment, as required by claim 1. In fact, E1 teaches (for the video signature) in col. 5, lines 33-35 that the goal in selecting a comparison to make is to maximise entropy, ie to minimise correlation. However, audio frequency signals are highly correlated in time. Therefore E1 arguably teaches away from the claimed solution of comparing to the corresponding frequency band value of a previous segment.

Re (iii): Even if it might occur to the skilled person to carry out some form of normalisation of the difference value on the basis of common knowledge (despite the lack of any suggestion in E1), there are plausibly other ways to calculate the normalising parameter. For example, it would be possible to divide by the sum taken over all the bands of the frequency spectrum.

2.4.7 Remaining documents

The board considers that none of the remaining documents cited in the procedure could be combined with document E1 to arrive at the subject-matter of claim 1.

During the written proceedings, Appellant I referred to the combination of E1 with any of the documents E2 to E9 (albeit not in connection with claim 1 of the second auxiliary request). Appellant I also cited further documents E11 to E16.

In the first place, the board considers that only documents in the same field as E1, ie recognition of audio broadcast signals, would be likely to be considered by the skilled person. These are: E2, E3, E4, E5, E8 and E12.

However, although E2, E4, E5, E8 and E12 make use of a spectral analysis, they do not suggest forming differences between corresponding frequency band values collected at different times. E3 does not use any spectral analysis at all. Hence, none of these documents in combination with E1 could render obvious the subject-matter of claim 1.

All other documents are considered to concern fields sufficiently remote from E1 that the skilled person would not consider them in combination: E6 resides in the field of speech recognition. E7, E9, E13 and E14 relate to speech coders, E10 relates to speech synthesising, E11 to identifying signals on a telephone line, and E15 relates to modulation.

E16 was filed shortly before the oral proceedings. It is a patent document with publication date of 1953, and discloses an arrangement with analogue circuitry. Prima facie, the board considers that the skilled person would not combine any teaching from this document with E1. The board therefore disregarded this late-filed document (Article 114(2) EPC).

2.4.8 The board accordingly concludes that claim 1 of the second auxiliary request meets the requirements for novelty and inventive step (Articles 52(1), 54 and 56 EPC).

3. Independent Claim 4 and dependent claims (Second auxiliary request)

3.1 The above remarks in connection with claim 1 apply, mutatis mutandis, to independent system claim 4. No objection was raised with respect to dependent claims 2 and 3.

4. Third and fourth auxiliary requests

The board concludes that the claims of the second auxiliary request are allowable. Hence, there is no need to consider Appellant II's third and fourth auxiliary requests.

5. Conclusion

In view of the above, the board concludes that the patent can be maintained as amended on the basis of the claims of the second auxiliary request (Article 101(3)(a) EPC).