T 157/08 – Gobbledygook?

This decision shows what happens when an invention is not explained in sufficient detail. It will perish – and it will be of no avail to invoke a skilled person that comes close to an intellectual superhero.

The applicant appealed against the refusal of his application by the Examining Division (ED) on the ground that the claimed method could not be carried out (Nichtausführbarkeit).

Claim 1 before the Board read (in English translation):

Method for separating speech from noise (Geräusche), comprising the following steps:

the speech signal to be isolated from jamming (Störungen) is decomposed into its elementary tones (sounds) (Einzelklänge (Laute)) by means of an envelope method;

the speech signal is transformed from a representation “Intensity over time” into a representation “Amplitude over frequency” by means of integral transformations;

in the representation “Amplitude over frequency”, the frequency space, a sound has a plurality of frequencies;

intervals defining harmonic relationships between frequencies of speech are calculated by means of number-theoretic functions;

the proper separation of speech and noise is obtained in the following method steps:

the frequencies within a sound are compared with each other by pairs, by means of appropriate loop methods (Schleifenverfahren), wherein intervals are formed from the pairs of frequencies, wherein the intervals are equal to the ratio of the two frequency values;

the intervals determined in this way are compared with the calculated intervals of speech in order to check whether there are harmonic relationships between the corresponding frequencies;

only frequencies that are harmonically related with each other are transferred into the speech file.

In his statement of grounds of appeal, the applicant pointed out that the complexity of the subject-matter required a high-level skilled person. The skilled person to be taken into account could not be a normal graduate of a technical college (Fachhochschulabgänger) having experience in the field of speech recognition. Rather, the skilled person was a graduated physicist having many years of experience in the field of speech recognition but also a comprehensive knowledge of mathematics and in particular in group theory. Moreover, this skilled person had to have experience in musicology and in particular in the field of harmonics.

The applicant also submitted that an objection of lack of feasibility (Ausführbarkeit) could only be raised when there were corroborated doubts based on reproducible facts. The ED had not provided such facts.

*** Translation of the German original ***

Feasibility of carrying out the claimed invention (A 83 EPC 1973)

Qualification of the average skilled person

[3.1] The appropriate average skilled person does not have the qualifications which the appellant would want him to have […]. According to the established case law of the Boards of appeal, the skilled person is a practitioner having the usual knowledge in his technical field at a given time (i.e. the date of filing or the priority date) and having at his disposal normal means and abilities for routine work. He may also be expected to consider neighbouring fields and a more general field which he is familiar with in order to look for known solutions to similar problems.

Therefore, in the present case, the average skilled person cannot have a superior qualification than a physicist who is used to dealing with the separation of speech and noise, or perhaps a sound engineer. Such a skilled person has to have knowledge in the fields of signal processing, acoustics and audio technology. He is familiar with standard mathematical methods of spectral analysis, such as for example integral transformations. However, he cannot be attributed special knowledge in technical fields which are not significant for his routine work and which, as in the present case, are brought into view not because of the problem to be solved but because of the solution underlying the present invention. Therefore, the relevant skilled person cannot be attributed significant knowledge in the field of musicology in general and harmonics in particular. Nor does he have above-average knowledge in special fields of mathematics such as group theory and number-theoretic functions as well as their practical application in technology.

[3.2] The present application suffers from a lack of precision of the terminology used, non-uniform terminology, deficient or even inexistent definitions of mathematical expressions and over-all insufficient explanations of the mathematical/technical concepts that ensure a successful separation of speech and noise. The feasibility of steps D, F and G of claim 1 in view of A 83 is particularly problematic. […]

[3.2.1] A first problem concerning the feasibility of step D consists in the breadth of the claim wording regarding the number and nature of the number-theoretical functions to be used. It is not obvious, and the appellant has not made it credible, that comparative intervals suitable for speech separation can as a rule be calculated with any number-theoretical function that can be imagined, as encompassed by claim 1. As a matter of fact, the statements in […] the application as published indicate that the number-theoretical functions to be used are part of particular “groups” and that there are functions that are not suitable for the desired separation method. However, the identity of those “groups” is not explained. Nor are there criteria in the application documents that would allow to decide which number-theoretical functions would be suitable for the purpose of speech separation, and which would not. Paragraphs [0057] to [0074] do indeed exemplify some suitable number-theoretical functions but they do not provide any hint let alone rules for finding further functions that might be suitable.

Moreover, it is not clear from the application what is meant by the expression “number-theoretical function” and how one could use them to calculate “intervals defining harmonic relationships between frequencies of speech.” The indications of the application regarding this topic are vague or incomprehensible. For instance, in the part of the description […] entitled “target intervals” it is stated that the “interval values” or “target intervals” are “very particular elements of the body of prime numbers” and the “result of complicated number-theoretical functions” and that they are calculated “from the number-theoretical functions of the body of prime numbers”. The list of preliminary calculation steps […] contains the statement “A number of number-theoretical functions calculates intervals of speech and generates the space of equivalent classes of tones of speech”; the explanation of the expression “equivalent classes” does not go beyond the indications “Speech intervals are calculated by means of number-theoretical functions and a space of equivalent classes is generated” […] and “Number-theoretical functions of prime numbers and powers of two constitute fields for equivalent classes” […]. [Another] paragraph mentions a “reference tone from the equivalent class of number-theoretical functions of the body of prime numbers”. There is no concrete embodiment allowing to concretely understand the execution of the required mathematical steps for generating the “fields” and the “space” of “equivalent classes” and of their elements, i.e. “tones of speech” and/or “reference tones”. Therefore, it is not clear how to calculated the needed comparative intervals and the “target intervals”, respectively, are to be calculated even when using the concrete exemplary functions provided in […] the application as published. It is indeed shown how the degree of consonance of a number n is obtained for Euler’s Gradus function, but it is left open how comparative intervals are determined with this function. It also remains unclear how such intervals are obtained from the Gamma function, i.e. the function

Moreover, it is to be noted that the indications in the [application] are mathematical terms but not functions stricto sensu.

The appellant’s reference to annex B1 to the statement of grounds of appeal rather contributes to confusion than to clarifying the matter because it is not clear how the intended determination of “harmonic” frequency intervals relates to functions which, as defined in annex B1, connect an element of the set of positive, natural numbers with an element of the set of complex numbers.

[3.2.2] As far as steps F and G of claim 1 are concerned, the corresponding explanatory parts of the description allow to understand the underlying idea of how the signal is analysed and the speech part separated from noises but not how this is concretely carried out from a process point of view.

For instance, it is not apparent how “appropriate loop methods” according to step F are to be carried out, in view of the fact that [the description] points out that the order of frequencies is never to be inverted when intervals in an envelope are to be determined but that there is no more precise indication of what the correct order would be. Instead of that the “summary” of […] the application as published refers to “astute combinatory operations of comparison” by which “systematically all tones of a sound, i.e. all frequencies of a wave packet, are divided by the adjacent frequencies until all elements (frequencies of a wave packet) have been compared with each other”, and it adds: “The intervals obtained from the wave packets of the speech sample by combinatorics and comparison are then inserted into mathematical formulae which take into account, inter alia, the particularities of the physics of hearing”. The application documents remain silent on which formulae and which particularities are meant.

Regarding the comparison (according to step G) of the intervals determined in steps D and F, the summary of […] the application mentions the “confrontation of a specific formula structure” and [notes] that “thereby group-theoretical structures expressing symmetry laws can be taken into account (in Ansatz gebracht)”, wherein “the separation [can] obviously take place when the tones of a speech sample exactly correspond to the calculated congruence groups of the tones in the form of intervals”. There is no explanation of the “specific formula structure” or of the “symmetry laws”. [Two] paragraphs refer to “symmetry relationships” and “symmetry operations”, respectively:

“All frequencies of the frequency space of the transformed speech signal have to be examined as to whether they have symmetry relationships with other frequencies of the same envelope. There are invariant properties of speech encrypted in the symmetry relationships of the elements.”

“The separation of speech and non-speech is carried out by means of a series of several symmetry operations”

However, none of these expressions is explained. The expression “congruence groups” reappears once more […] together with the expressions “equivalence class” and “equivalence relations”, without there being any explanation. The corresponding instructions

“The equivalence classes are related to the speech signal in the form of equivalence relations”.

and

“Inversions of elements are specified by a double- or multi-digit relation on the set of all sounds of a pitch space (Tonsystem)”

are equally incomprehensible as […]

“A two-digit relation of the intervals is defined such that a reference tone exactly corresponds to the element of the signal to be examined within several coordinates of the space of equivalence classes.”

and

“The elements of the existing set of equivalence relations (Faktormenge) of frequency intervals from equivalence classes are now defined as harmonies.”

[3.3] The arguments invoked by the appellant do not deal with the above mentioned deficiencies of the application and are not persuasive for this reason alone.

[3.4] For the reasons given above, a method for separating speech from noise, as claimed in present claim 1, cannot be carried out by the skilled person on the basis of the pieces of information contained in the application documents, as required by A 83 for the grant of European patents.

Therefore, the request of the appellant for the grant of a patent cannot be allowed. […]

The appeal is dismissed.

Should you wish to download the whole decision (in German), just click here.

The file wrapper can be found here.