EP1368805B1 - Method and device for characterising a signal and method and device for producing an indexed signal - Google Patents

Abstract

In a method for characterizing a signal, which represents an audio content, a measure for a tonality of the signal is determined, whereupon a statement is made about the audio content of the signal based on the measure for the tonality of the signal. The measure for the tonality of the signal for the content analysis is robust against a signal distortion, such as by MP3 encoding, and has a high correlation to the content of the examined signal.

Description

The present invention relates to characterization of audio signals with regard to their content and in particular on a concept for classifying or indexing Audio pieces in terms of their content, for researchability to enable such multimedia data.

In the past few years, the availability of multimedia data, d. H. of audio data, has risen sharply. This Development has been due to a number of technical factors conditionally. These technical factors include, for example the wide availability of the Internet, the wide availability powerful calculator and wide availability powerful data compression techniques, d. H. Source coding, of audio data. An example of this is MPEG 1/2 Layer 3 called, which is also called MP3.

The huge amounts of audiovisual data, for example available on the Internet worldwide, require concepts which make this data possible according to content criteria assess, catalog or manage. It there is a need to provide targeted multimedia data to search for and find useful criteria.

This requires the use of so-called "content-based" Techniques based on so-called characteristics from the audiovisual data, which are also called "features" in technology, extract the important characteristic content properties of the signal of interest. Based on such features or combinations of such features can similarity relationships or similarities between the audio signals are derived. This is done generally by comparing or relating the extracted Characteristic values from different signals, which are also to be referred to here as "pieces".

U.S. Patent No. 5,918,223 discloses a method for the Content-based analysis, storage, recovery and Segmentation of audio information. An analysis of audio data generates a set of numerical values, also called Feature vector is referred to, and used for this can determine the similarity between individual audio pieces that typically in a multimedia database or on the World Wide Web are stored, classified and ranked.

The analysis also enables the description of user-defined Classes of audio pieces based on an analysis of a set of audio pieces that all members of a Are user-defined class. The system is able individual sections of sound within a longer piece of sound find what enables audio recording to automatically segmented into a series of shorter audio segments becomes.

As characteristics for the characterization or classification of Audio pieces regarding their content becomes the loudness of a Piece, the bass content of a piece, the pitch, the Brightness, the bandwidth and the so-called Mel Frequency Cepstral Coefficients (MFCCs) for periodic Intervals used in the audio piece. The values per Block or frame are saved and a first derivative subjected. Specific statistical values are then calculated, such as. the mean or standard deviation, from each of these features including the first Derivatives of these to describe a variation over time. This set of statistical quantities forms the Feature vector. The feature vector of the audio piece is in one Database stored in association with the original file, where a user can access the database for appropriate Retrieve audio tracks.

The database system is able to measure the distance in an n-dimensional Space between two n-dimensional vectors quantify. It is also possible to have classes of audio pieces to generate by specifying a set of audio pieces who belongs in a class. Example classes are twittering birds, Rock music, etc. The user is enabled to the audio track database using specific ones Search procedures. The result of a search is one List of sound files ordered by their distance from that specified n-dimensional vector are listed. The User can search the database for similarity characteristics, with regard to acoustic or psychoacoustic Characteristics, in terms of subjective characteristics or in terms of special noises, e.g. Bee buzz, search.

The specialist publication "Multimedia Content Analysis using both audio and visual clue", Yao Wang et al., IEEE Signal Processing Magazine, November 2000, Pages 12 to 36, discloses a similar concept to multimedia pieces to characterize. As characteristics for classification The content of a multimedia piece becomes time domain features or frequency domain features suggested. These include the volume, the pitch as the fundamental frequency of an audio signal form, spectral features such as B. the energy content of a Band based on the total energy content, limit frequencies in the spectral course etc. In addition to short-term characteristics that the mentioned sizes per block of samples of the audio signal long-term variables are also proposed, which affect refer to a longer period of the audio track.

Different categories are used to characterize audio pieces suggested, such as Animal sounds, bell sounds, Crowd sounds, laughter, machine noises, Musical instruments, male language, female language, Telephone noises or water noises.

The problem with the selection of the features used is that the computational effort to extract a feature is moderate to be able to achieve a rapid characterization that however at the same time the characteristic of the audio piece is characteristic should be such that two different pieces also have distinguishable features.

The robustness of the feature is also problematic. So is not based on robustness criteria for the concepts mentioned received. If an audio piece is immediately after its Generation characterized in the recording studio and with an index provided, which represents the feature vector of the piece and to a certain extent forms the essence of the piece, so is the probability relatively high to recognize this piece, if the same, undistorted version of this piece the same Process is subjected to, i.e. extracted the same characteristics be and the feature vector in the database with a Comparing a large number of feature vectors of different pieces becomes.

However, it becomes problematic when an audio piece is in front its characterization is distorted so that what is to be characterized Signal no longer identical to the original one Signal, but has the same content. A person who, for example knows a song, will recognize that song, if it is noisy, if it is louder or quieter or if it is played at a different pitch than originally added. Another distortion, for example achieved through lossy data compression have been, for example by means of a coding method according to an MPEG standard, e.g. MP3 or AAC.

Does a distortion or data compression mean that the Characteristic due to the distortion or data compression as well severely compromised, this would mean that the essence is lost while the content of the piece for a human is still recognizable.

U.S. Patent No. 5,510,572 discloses an apparatus for Analyze and harmonize a tune using results of a melody analysis. A melody in the form of a Sequence of notes played by a keyboard, is read in and broken down into melody segments, a melody segment, i.e. a phrase, e.g. B. four bars of the melody includes. A tonality analysis is done with each phrase, to determine the key of the melody in that phrase. To do this, the pitch of a note in the phrase is determined and then a pitch difference between the current one considered note and the previous note. Further becomes a pitch difference between the current note and the following note. Because of the pitch differences becomes a previous coupling coefficient and a subsequent coupling coefficient determined. The coupling coefficient for the current grade then results from the previous coupling coefficient and the following Coupling coefficient and the note length. This process will repeated for each note of the melody in the phrase to the Key of the melody or a candidate for the key of the Determine melody. The key of the phrase is used to a grade type classifier for interpretation to control the meaning of each note in a phrase. The key information, which were obtained by the tonality analysis is also used to create a transpose module to control the one in a reference key in a database stored chord progression in the by tonality analysis transposed certain key for a considered melody phrase.

Document US-B1-6185527 discloses a classification and an indexing of audio data based on a tonality determination.

The object of the present invention is an improved Concept for characterizing or indexing a To create signal that has audio content.

This task is accomplished through a characterization process of a signal according to claim 1, by a method for Generating an indexed signal according to claim 11, by a device for characterizing a signal Claim 14 or by a device for generating a indexed signal according to claim 15 solved.

The present invention is based on the finding that when selecting the characteristic for characterization or indexing of a signal especially for robustness Distortions of the signal must be taken into account. The usefulness of characteristics or combinations of characteristics depends on how strongly by irrelevant changes such as B. by a MP3 coding, can be changed.

According to the invention is used as a characteristic for characterizing or indexing of signals uses the tonality of the signal. It it has been found that the tonality of a signal, i. H. the property of a signal, a rather flat spectrum with pronounced lines or rather a spectrum with the same height Having lines that are more robust to distortion is more common Is like Distortion caused by a lossy coding method, such as. MP3. In essence, the essence of the signal is taken its spectral appearance, and related to the individual spectral lines or groups of Spectral lines. The tonality also provides great flexibility with regard to the computing effort to be carried out in order to to determine the tonality measure. The tonality measure can be taken from the Tonality of all spectral components of a piece derived or from the tonality of groups of spectral components, etc. In addition, tonalities of successive short-term spectra of the signal under investigation either individually or weighted or statistically evaluated be used.

In other words, the tonality depends on the present Registration based on the audio content. Is the audio content or the signal under consideration with the audio content has noisy, so it has a different tonality than a less noisy signal. A noise-like signal typically has a lower one Tonality value as a less noisy, i.e. H. more tonal, Signal. The latter signal has a higher tonality value.

The tonality, i.e. H. the noise or tonality of a signal, is a quantity dependent on the content of the audio signal, which are largely unaffected by different types of distortion is. A concept based on a tonality measure Characterizing or indexing signals therefore provides a robust recognition, which shows that the tonality essence of a signal is not beyond recognition is changed if the signal is distorted.

Distortion is, for example, a transmission of the signal from a loudspeaker via an air transmission channel to a microphone.

The robustness property of the tonality feature is significant with regard to lossy compression methods.

It has been found that the tonality measure of a signal through lossy data compression such as according to one of the MPEG standards not or hardly being affected. It also provides a distinguishing feature based on the tonality of the signal a sufficiently good one Essence for the signal so that two different from each other Audio signals also have sufficiently different tonality measures deliver. The content of the audio signal is therefore strong correlates with the tonality measure.

The main advantage of the present invention is thus in that the tonality measure of the signal compared to disturbed, d. H. distorted, signals is robust. This robustness exists in particular against filtering, i. H. equalization, Dynamic compression, lossy data reduction, such as. MPEG-1/2 Layer 3, an analog transmission, etc. It also provides the tonality property of a signal has a high correlation to the content of the signal.

Fig. 1

ein Prinzipblockschaltbild einer erfindungsgemäßen Vorrichtung zum Charakterisieren eines Signals;

Fig. 2

ein Prinzipblockschaltbild einer erfindungsgemäßen Vorrichtung zum Indexieren eines Signals;

Fig. 3

ein Prinzipblockschaltbild einer Vorrichtung zum Berechnen des Tonalitätsmaßes aus der Tonalität pro Spektralkomponente;

Fig. 4

ein Prinzipblockschaltbild zum Bestimmen des Tonalitätsmaßes aus der Spectral Flatness Measure (SFM); und

Fig. 5

ein Prinzipblockschaltbild eines Mustererkennungssystems, in dem das Tonalitätsmaß als Merkmal (Feature) verwendet werden kann.

Preferred embodiments of the present invention are explained in detail below with reference to the accompanying drawings. Show it:

Fig. 1

a basic block diagram of an inventive device for characterizing a signal;

Fig. 2

a basic block diagram of an inventive device for indexing a signal;

Fig. 3

a basic block diagram of a device for calculating the tonality measure from the tonality per spectral component;

Fig. 4

a basic block diagram for determining the tonality measure from the Spectral Flatness Measure (SFM); and

Fig. 5

a block diagram of a pattern recognition system in which the tonality measure can be used as a feature.

Fig. 1 shows a basic block diagram of an inventive Device for characterizing a signal that a Represents audio content. The device includes an entrance 10, in which the signal to be characterized are entered can, the signal to be characterized compared to a original signal, for example a lossy one Has undergone audio coding. The one to be characterized Signal is in a device 12 for determining a measure for the tonality of the signal. The measure of that Tonality for the signal is via a connecting line 14 a device 16 for making a statement about the content of the signal supplied. The device 16 is designed to this statement based on the transmitted by the device 12 Measure of the tonality of the signal and delivers this statement about the content of the signal at an output 18 of the system.

2 shows an inventive device for generating an indexed signal that has audio content. The Signal, for example an audio piece as it is generated in the recording studio has been stored on a compact disc, is via an input 20 in the device shown in Fig. 2 fed. A device 22 that is basically the same how the device 12 of FIG. 12 can be constructed, determines a measure of the tonality of the signal to be indexed and delivers this measure via a connecting line 24 to a device 26 for recording the measurement as an index for the signal. At an output of the device 26, the at the same time the output 28 of the device shown in FIG. 2 to generate an indexed signal, the Signal fed in at input 20 together with a tonality index be issued. Alternatively, the one in FIG Device shown be designed so that at the output 28 a table entry is generated, the tonality index linked to an identification mark, the identification mark clearly assigned to the signal to be indexed is. Generally, the device shown in Fig. 2 provides one Index for the signal, where the index is assigned to the signal and indicates the audio content of the signal.

When a plurality of signals by the one shown in Fig. 2 Device is processed, gradually creates a database from indices for audio pieces, for example for the pattern recognition system outlined in FIG. 5 can be used can. In addition to the indices, the database optionally contains the Audio pieces themselves Tonality properties can be easily searched to identify a piece by the device shown in FIG. 1 and classify them, in terms of tonality or in terms of similarities to others Pieces or distances between two pieces. Generally however, the device shown in Fig. 2 provides one possibility to create pieces with an associated meta description, d. H. the tonality index. Therefore it is possible Records e.g. to index according to given tonality indices and search so that according to the present invention an efficient search and find of Multimedia pieces is possible.

Various can be used to calculate the tonality measure of a piece Procedures are applied. As shown in Fig. 3 is a time signal to be characterized by means of a device 30 are implemented in the spectral range, to a block from a block of temporal samples of generating spectral coefficients. As explained later can, for each spectral coefficient or for each spectral component a separate tonality value can be determined in order for example using a yes / no determination, whether a spectral component is tonal or not. Under Using the tonality values for the spectral components and the energy or power of the spectral components, the Tonality values can be determined by the device 32 then by means of a device 34 the tonality measure for the Signal calculated in a variety of different ways become.

Due to the fact that, for example, in FIG concept described receive a quantitative tonality measure , it is also possible to see distances or similarities between specify two tonality-indexed pieces, where Pieces can be classified as similar if their tonality measures only about a difference less than one differentiate predetermined threshold while pieces other than can be classified differently if their tonality indices differ by a difference that is greater than is a dissimilarity threshold. In addition to the difference between Two tonality measures can be used to determine the tonality distance other sizes are used between two pieces, such as B. the difference between two absolute values, the square a difference, the quotient between two tonality measures less one, the correlation between two tonality measures, the distance metric between two tonality measures, the n-dimensional Are vectors, etc.

It should be noted that the signal to be characterized does not necessarily have to be a time signal, but that it is the same can also be an MP3-encoded signal, for example, which consists of a sequence of Huffman code words consisting of quantized spectral values have been generated.

The quantized spectral values were from the original Spectral values generated by quantization, the quantization was chosen such that the quantization introduced quantization noise below the psychoacoustic Masking threshold is. In such a case, as shown, for example, with reference to FIG. 4, directly the encoded MP3 data stream can be used, for example the spectral values using an MP3 decoder calculate (device 40 in Fig. 4). It is not necessary before the determination of the tonality an implementation in the time domain and then again implement a conversion into the spectral range, but it can be inside the MP3 decoder calculated spectral values can be taken directly to the Tonality per spectral component or as shown in FIG. 4 is the SFM (SFM = Spectral Flatness Measure = measure for the spectral flatness) to be calculated by the device 42. If so to determine the tonality spectral components are used and if the signal to be characterized is an MP3 data stream, device 40 is like one Decoder built, but without the inverse filter bank.

The measure for spectral flatness (SFM) is calculated using the following equation.

In this equation, X (n) stands for the square of one Spectral component with the index n, while N for the total number is the spectral coefficient of a spectrum. Out The equation shows that the SFM is equal to the quotient from the geometric mean of the spectral components to arithmetic mean of the spectral components. As known the geometric mean is always smaller or at most equal to the arithmetic mean so that the SFM has a range of values between 0 and 1. A value indicates close to 0 to a tonal signal and a value close to 1 to a closer noise-like signal with a flat spectral curve. It it should be noted that the arithmetic mean and the Geometric averages are only the same if all X (n) are identical are what is completely atonal, d. H. intoxicated or impulsive Signal corresponds. In the extreme case, however, is only a spectral component very large in amount, while other spectral components X (n) are very small in amount are, the SFM will have a value close to 0, indicating a indicates very tonal signal.

The SFM is in "Digital Coding of Waveforms", Englewood Cliffs, NJ, Prentice-Hall, N. Jayant, P. Noll, 1984 and was originally used as a measure of the maximum to be achieved Coding gain defined from a redundancy reduction.

The SFM can then be determined by a device 44 of the tonality measure the tonality measure can be determined.

Another way to determine the tonality of the spectral values, performed by a device 32 of FIG. 3 can be determined by determining peaks in the Power density spectrum of the audio signal as found in MPEG-1 audio ISO / IEC 11172-3, Annex D1 "Psychoacoustic Model 1" is. Here the level of a spectral component determined. Thereupon the levels of two become the one spectral component surrounding spectral components determined. A Classification of the spectral component as tonal then takes place instead when the level of the spectral component is a predetermined Factor is greater than a level of a surrounding Spectral component. The predetermined threshold is in the state of technology adopted as 7dB, being for the present invention however, any other predetermined thresholds are used can be. This allows for each spectral component whether it is tonal or not. The measure of tonality can then by means 34 of FIG. 3 under Use of the tonality values for the individual components and the energy of the spectral components can be specified.

Another way to determine the tonality of a Spectral component consists in evaluating the temporal Predictability, d. H. Predictability, the spectral component. Here again MPEG-1 Audio ISO / IEC 11172-3, Annex D2 "Psychoacoustic Model 2". General will a current block of samples of the to be characterized Signal converted into a spectral representation to a to get current block of spectral components. hereupon become the spectral components of the current block of spectral components using information from samples of the signal to be characterized that corresponds to the current Go ahead block, so using historical information, predicted. This will result in a prediction error from which a tonality measure is then derived can.

Another possibility for determining the tonality is in U.S. Patent No. 5,918,203. Another positive one real-value representation of the spectrum of the to be characterized Signal used. This representation can show the amounts include the amount squares etc. of the spectral components. In one embodiment, the amounts or squares of amounts of the spectral components initially logarithmic compressed and then using a filter with differentiating Characteristic filtered to differentiate a block of to get filtered spectral components.

In another embodiment, the amounts of Spectral components first with a filter with differentiating Characteristic filtered to get a counter and then with a filter with an integrating characteristic filtered to get a denominator. The quotient of one differentially filtered amount of a spectral component and the integrally filtered amount of the same spectral component then gives the tonality value for this spectral component.

Both of these approaches make slow changes suppressed between neighboring amounts of spectral components, during abrupt changes between neighboring amounts of spectral components in the spectrum. Slow changes between neighboring amounts of Spectral components indicate atonal signal components, while abrupt changes indicate tonal signal components. The logarithmically compressed and differentiating filtered spectral components or the quotients then in turn be used to measure a tonality to calculate the considered spectrum.

Although in the previous text it was said that a Tonality value is calculated per spectral component, it will preferred in view of a lower computing effort, for example always the amount squares of two neighboring ones Add spectral components and then for each result the addition of a tonality value by one of the above Calculate procedure. Any kind of additive grouping of amount squares or amounts of spectral components can be used to set tonality values for more than one Calculate spectral component.

Another way to determine the tonality of a Spectral component is the level of a spectral component with an average of levels of spectral components to compare in a frequency band. The width of the frequency band, in which the one spectral component lies, the Level with the mean z. B. the amounts or amount squares the spectral components can be compared depending on the requirement to get voted. For example, there is one possibility in that the band is chosen narrow. alternative the band could also be chosen broadly, or also according to psychoacoustic Aspects. As a result, the influence can be brief Performance drops in the spectrum can be reduced.

Although the tonality of an audio signal was determined based on its spectral components, this can also in the time domain, i.e. using the samples of the Audio signal happen. This could be an LPC analysis of the signal be performed to gain a prediction for the Estimate signal. The prediction gain is inversely proportional to the SFM and is also a measure of tonality of the audio signal.

In a preferred embodiment of the present invention not only one value is given per short-term spectrum, but the tonality measure is a multidimensional vector of tonality values. For example, the short-term spectrum in four adjacent and preferably not overlapping areas or frequency bands are divided, with a tonality value for example for each frequency band by the device 34 of FIG. 3 or by the device 44 of Fig. 4 is determined. This is for a short-term spectrum of the signal to be characterized is a 4-dimensional one Preserve tonality vector. For better characterization it would also be preferred, for example four successive short-term spectra as described above edit, so that overall a tonality measure results, which is a 16-dimensional vector or generally an n x m-dimensional Is vector, where n is the number of tonality components per frame or block of samples, while m for the number of blocks or short-term spectra under consideration stands. The tonality measure would then, as stated, a 16-dimensional vector. To the course of time take better account of the signal to be characterized, it is further preferred to have several such 16-dimensional ones Calculate vectors and then statistically too process, for example, variance, mean or central moments higher order from all n x m dimensional Tonality vectors of a piece with a certain Calculate length to index this piece.

Generally speaking, the tonality can thus consist of parts of the whole Spectrum can be calculated. So it is possible to Tonality / noiseiness of a sub-spectrum or several Determine sub-spectra and thus a finer characterization to achieve the spectrum and thus the audio signal.

Furthermore, short-term statistics from tonality values, such as e.g. Mean, variance and central moments of higher order, can be calculated as a measure of tonality. These are calculated using statistical Techniques based on a temporal sequence of tonality values or tonality vectors are determined and thus deliver an essence over a longer section of a piece.

In addition, there can also be differences of consecutive times Tonality vectors or linearly filtered tonality values are used, for example as a linear filter IIR filters or FIR filters can be used.

It also becomes when calculating the SFM (block 42 in FIG. 4) preferred in order to save computing time, for example two to add or add adjacent squares of frequencies average and the SFM calculation on this coarsened positive and real-value spectral display. This also leads to greater robustness compared to narrowband Frequency drops and less computing effort.

5, which is a schematic Overview of a pattern recognition system shows at which the present invention can be used advantageously can. A distinction is made in principle with one shown in FIG. 5 Pattern recognition system between two operating modes, namely training mode 50 and classification mode 52.

In the training mode, data is "trained", i.e. H. the System added and then recorded in a database 54.

In classification mode an attempt is made to characterize one Signal with the entries in the database 54 to compare and order. The invention shown in Fig. 1 Device can be in classification mode 52 be used when there are tonality indices of other pieces, with which the tonality index of the current piece can be compared to a statement about the piece too to meet. The device shown in Fig. 2, however, is advantageous used in training mode 50 of Fig. 5 to the Database to be filled gradually.

The pattern recognition system comprises a device 56 for signal preprocessing, a downstream device 58 for Feature extraction, a device 60 for feature processing, a device 62 for cluster generation, and means 64 for performing a classification to for example, as a result of classification mode 52 such a statement about the content of the signal to be characterized to meet that signal with the signal xy that is in a Previous training mode has been trained identically is.

Below is the functionality of each Blocks of Fig. 5 received.

Block 56, together with block 58, forms a feature extractor, while block 60 represents a feature processor. Block 56 sets an input signal to a uniform one Target format, such as B. the number of channels, the sampling rate, the resolution (in bits per sample) etc. This is insofar as it makes sense and is necessary because there are no requirements about the source from which the input signal originates should.

The feature 58 for feature extraction is used to do the usual large amount of information at the exit of the facility 56 to a small amount of information. The too investigating signals usually have a high data rate, so a high number of samples per time period. The restriction on a small amount of information must take place that the essence of the original signal, that is, the peculiarity the same, is not lost. In facility 58 are given characteristic properties, as general for example loudness, fundamental frequency, etc. and / or, according to the present invention, tonality features or the SFM, extracted from the signal. The tonality characteristics thus obtained are said to be the essence of the signal under investigation include.

In block 60, the previously calculated feature vectors can are processed. The processing is simple Standardization of the vectors. Possible processing of characteristics are linear transformations, such as the Karhunen-Loeve transformation (KLT) or linear discriminant analysis (LDA), which are known in the art. More in particular nonlinear transformations are also available Feature processing applicable.

The class generator is used to process the feature vectors to combine into classes. These classes correspond a compact representation of the associated signal. The Classifier 64 is finally used to generate a feature vector a predefined class or a predefined Assign signal.

The table below provides an overview of detection rates under various conditions. Type of distortion detection rate
(Loudness as a characteristic) detection rate
(SFM as a characteristic) MP3 coding, 96kbps, 30s cutout 83.9% 100% MP3 coding, 96kbps, 15s cutout 76.1% 74.1%

The table presents detection rates using a database (54) of FIG. 5 with a total of 305 pieces of music, of which the first 180 seconds each as reference data were trained. The detection rate gives the percentage Number of correctly recognized pieces depending on the signal influence on. The second column represents the detection rate if loudness is used as a characteristic. In particular the loudness was calculated in four spectral bands, then logarithmizing the loudness values, and then a difference of logarithmic loudness values for corresponding spectral bands in succession carried out. The result obtained was used as a feature vector used for loudness.

In the last column, the SFM was used as the feature vector for four bands used.

It can be seen that the use of tonality according to the invention as a classification feature for a 100% recognition rate of MP-3 encoded pieces when a snippet of 30 seconds is considered while the detection rates are both in the inventive feature as well as in the Decrease loudness as a characteristic if shorter sections (e.g. 15 s) of the signal to be examined is used for detection become.

As has already been stated, the one shown in FIG Device used to do the shown in FIG Train detection system. In general, however, the in Fig. 2 device shown can be used for any Multimedia records meta descriptions, d. H. Generating indexes so that it is possible to view records regarding their Search for tonality values or records from a database to output that have a certain tonality vector or are similar to a certain tonality vector.

Claims (15)

1. Method for characterizing a signal, which represents an audio content, comprising:

   determining (12) a measure for a tonality of the signal, wherein the tonality depends on the audio content, and wherein the tonality for a noisy signal differs from the tonality for a tone-like signal, wherein the step (12) of determining a measure for the tonality comprises:

   calculating (40) a block of positive and real-valued spectral components for the signal to be characterized;

   forming (42) a quotient with the geometric mean value of a plurality of spectral components of the block of spectral components as numerator and the arithmetic mean value of the plurality of spectral components in the denominator, wherein the quotient serves as measure for the tonality, wherein a quotient with a value near 0 indicates a tonal signal, and wherein a quotient near 1 indicates an atonal signal with flat spectral curve; and

   making (16) a statement about the audio content of the signal based on the measure for the tonality of the signal.
2. Method according to claim 1, wherein the step (16) of making a statement comprises:

   comparing (64) the measure for the tonality of the signal with a plurality of known tonality measures for a plurality of known signals, which represent different audio contents;

   determining that the audio content of the signal to be characterized corresponds to the content of a known signal, when the tonality measure of the signal to be characterized has a lower than predetermined deviation from the tonality measure, which is associated with the known signal.
3. Method according to claim 2, further comprising:

   outputting a title, an author or other metainformation for the signal to be characterized, when a correspondence is determined.
4. Method according to claim 1, wherein the measure for the tonality is a quantitative quantity, wherein the method further comprises:

   calculating a tonality distance between the determined measure for the tonality of the signal and a known tonality measure for a known signal; and

   indicating a similarity measure for the signal to be characterized, wherein the similarity measure depends on the tonality distance and represents the similarity of the content of the known signal to the content of the signal to be characterized.
5. Method according to one of the previous claims,
   wherein the signal to be characterized is derived by encoding from an original signal,
   wherein the encoding comprises a block-wise conversion of the original signal into the frequency domain and a quantizing of spectral values of the original signal controlled by the psychoacoustic model.
6. Method according to one of claims 1 to 4,
   wherein the signal to be characterized is provided by outputting an original signal via a speaker and by recording via a microphone.
7. Method according to claim 1, wherein at least two spectral components adjacent in frequency are grouped, thereupon not the individual spectral components but the grouped spectral components will be further processed.
8. Method according to one of the previous claims,
   wherein in the step (12) of determining a short-time spectrum of the signal to be characterized is divided into n bands, wherein a tonality value is determined for every band,
   wherein further for m successive short-time spectra of the signal to be characterized n tonality values are determined each, and
   wherein a tonality vector is formed with a dimension, which is equal to m x n, wherein m and n are greater or equal to 1.
9. Method according to claim 8, wherein the measure for the tonality is the tonality vector or a statistic quantity from a plurality of timely successive tonality vectors of the signal to be characterized, wherein the statistic quantity is a mean value, a variance or a central moment higher order or a combination of the above-mentioned statistic quantities.
10. Method according to claim 8, wherein the measure for the tonality is derived from a difference of a plurality of tonality vectors or a linear filtering of a plurality of tonality vectors.
11. Method for generating an indexed signal, which comprises an audio content, comprising:

    determining (22) a measure for a tonality of the signal, wherein the tonality depends on the audio content, and wherein the tonality for a noisy signal differs from the tonality for a tone-like signal, wherein the step (12) of determining a measure for the tonality comprises:

    calculating (40) a block of positive and real-valued spectral components for the signal to be characterized;

    forming (42) a quotient with the geometric mean value of a plurality of spectral components of the block of spectral components as numerator and the arithmetic mean value of the plurality of spectral components in the denominator, wherein the quotient serves as a measure for the tonality, wherein a quotient with a value near 0 indicates a tonal signal, and wherein a quotient near 1 indicates an atonal signal with flat spectral curve; and

    recording (26) the measure for the tonality as index in association to the signal, wherein the index refers to the audio content of the signal.
12. Method according to claim 11, wherein the step of determining (22) a measure for the tonality comprises:

    calculating tonality values for different spectral components or groups of spectral components of the signal; and

    processing the tonality quantities (60) to obtain the measure for the tonality; and

    associating (62) the signal with a signal class depending on the measure for the tonality.
13. Method according to claim 11, which is performed for a plurality of signals, to obtain a data bank (54) of references to the plurality of signals together with associated indices which refer to tonality properties of the signals.
14. Apparatus for characterizing a signal, which represents an audio content, comprising:

    means for determining (12) a measure for a tonality of the signal, wherein the tonality depends on the audio content, and wherein the tonality for a noisy signal differs from the tonality for a tone-like signal, wherein the means for determining is configured to:

    calculate a block of positive and real-valued spectral components for the signal to be characterized (40); and

    form a quotient with the geometric mean value of a plurality of spectral components of the block of spectral components as numerator and the arithmetic mean value of the plurality of spectral components in the denominator (42), wherein the quotient serves as a measure for the tonality, wherein a quotient with a value near 0 indicates a tonal signal, and wherein a quotient near 1 indicates an atonal signal with flat spectral curve; and

    means for making (16) a statement about the audio content of the signal based on the measure for the tonality of the signal.
15. Apparatus for generating an indexed signal, which comprises an audio content, comprising:

    means for determining (22) a measure for a tonality of the signal, wherein the tonality depends on the audio content, and wherein the tonality for a noisy signal differs from the tonality for a tone-like signal, wherein the means for determining is configured to:

    calculate a block of positive and real-valued spectral components for the signal to be characterized (40); and

    form a quotient with the geometric mean value of a plurality of spectral components of the block of spectral components as numerator and the arithmetic mean value of the plurality of spectral components in the denominator (42), wherein the quotient serves as a measure for the tonality, wherein a quotient with a value near 0 indicates a tonal signal, and wherein a quotient near 1 indicates an atonal signal with flat spectral curve; and

    means for recording (26) the measure for the tonality as index in association to the signal, wherein the index refers to the audio content of the signal.

Images