EP1212751B1 - Method for suppressing spurious noise in a signal field - Google Patents

Description

The invention relates to a method for suppressing noise in a signal field containing a plurality of signal components, each of which has a value of Accept signal levels and can be applied over an ordinate range from which a distribution function is determined in the signal field, which is a function of the signal level for each of its possible signal level argument values indicates how large the portion is those signal components whose signal level is lower than the argument value that Signal level values of the signal field are modified so that the distribution function of the modified signal field is equal to a predetermined reference distribution, the Sequence of signal components regarding their energy level remains unchanged as well Signal components whose original signal levels are the same, modified the same Signal levels are assigned, and as a reference distribution function one from a distribution function, which has been determined for a set of reference patterns Function is used.

Signal fields to which the method according to the invention relates are, for example, in Pattern recognition systems used to describe the patterns to be recognized. The The process of recognizing a pattern can usually be roughly as follows Steps to be split: capturing the pattern, preprocessing and classification.

The first step, the pattern acquisition, is used to convert the original pattern, e.g. a spoken utterance by a user or a document written with text, in a format suitable for processing, e.g. in the form of an electronic signal that can be coded analog or digital, or a file of a predetermined format. Here also includes the conversion of a signal / file format, e.g. a raster image recording, in a format suitable for further processing. In the case of speech recognition, for example the utterance spoken by the user via an acoustic input, such as. a microphone, recorded, possibly pre-amplified and in an electrical Voice signal implemented in analog or digitized form.

The pattern captured in this way is fed to the preprocessing, which is a reduction in the processing data and a better differentiation of the samples to be determined reached. The result of the preprocessing is a signal field, in the example of speech recognition a spectrum of utterance that can be fed into the classification system. An essential step of preprocessing is often a signal analysis of the pattern signal, e.g. a signal analysis can be carried out for the electrical voice signal of the user utterance Form of a division into time frames (discretization) and a subsequent one Fourier transformation carried out within a time frame with decomposition into Frequency bands occur from which a time-frequency spectrum is obtained. So that is at the same time a - generally considerable - data reduction. Another, Under certain circumstances, an essential step of the preprocessing is the reduction of interference noise in the pattern signal or the signal field obtained therefrom.

The signal field comprises a large number of signal components, each of which takes on its own value, here referred to as signal level, of the same type. The signal components are naturally arranged within the signal field, this order being expressed with the help of one or more ordinate parameters. For example, a signal field realized as a time-frequency spectrum consists of many spectral components, each of which has its own energy level; the spectral components are sorted by time frame and frequency band. Each signal component can thus be assigned its own area element of the ordinate area in the ordinate area over which the signal field extends, so that the area elements as a whole cover the ordinate area of the signal field. Depending on the number of ordinate parameters, the ordinate range can be one, two or more dimensions; accordingly, the area elements are line, area or ( n- dimensional) volume elements.

The signal field obtained by the preprocessing becomes the classification system fed. This determines which recognition class - i.e. in the case of speech recognition a word of a given vocabulary or a word string - a match given is. The recognition result is then output, for example on an advertisement, or used for further processing, e.g. when entering a command language-oriented facility.

The execution of a pattern recognition is often made more difficult by noise, which the overlapping patterns to be recognized. For example, the performance of a speech recognition system greatly reduced or completely by acoustic background noise be thwarted.

In known methods for noise suppression, an estimation of the noise parameters underlying the signal is carried out in the preprocessing and a reference noise signal is subtracted on the basis of this estimate. Such methods of spectral subtraction for speech signals are described by SV Vaseghi and BP Milner in 'Noise Compensation Models for Hidden Markov Model Speech Recognition in Adverse Environments', IEEE Transactions on Speech and Audio Processing, Vol. 5, No. 1, January 1997, pp. 11-21. From the energy level E, a spectral component of the spectrum becomes the corresponding component of a reference noise signal Er according to the expression E '= s s (E, E r ) = (E b - α E r b ) 1 / b "Subtracted". The reference noise signal Er is simulated on the basis of predefined or estimated noise parameters. The subtraction of the energy levels can be carried out, for example, in relation to the linear energy levels or “convolutively” in the logarithmic range, ie in the formula mentioned the corresponding logarithms log E, etc. are used instead of the energy levels E, E _r , E '.

The subtraction approach, however, has the shortcoming that is used to describe the noise necessary parameters not with the required accuracy and completeness can be known. For example, for correct noise compensation it is not just that Knowledge of the noise amplitudes but also the phase relationships required what - if at all - only possible with great effort. Disorders that are not additive or represent a convolutive overlay, e.g. Mixed forms of additive and convolutive Disorders are even more difficult to deal with.

EP 0 062 519 A1 teaches the elimination of interference in radar signals, the distribution the disturbances are known, although arbitrary, in contrast to previously known ones Procedures that require a Rayleigh or Weibull distributed disorder. Knowing the Distribution or at least the associated probability density from which you can get them can derive is a necessary prerequisite for the application of the procedure this Document. Without knowledge of such a distribution, troubleshooting is possible this method is therefore not feasible.

EP 0 548 527 A2 teaches a method for generating a level scale transformation a digital radiographic image, e.g. X-ray image in which a cumulative Distribution function of the image is used to measure the level distribution of the image to modify it to be substantially linear in the area of interest. The The task underlying this method, namely a representation of the image in a form suitable for further investigation by viewing the image, differs significantly from that of the invention.

EP 0 720 358 A2 relates to the compression of video signal data. The level distribution of an image modified so that each input level range is larger Output level range is assigned, the more input levels fall in the former range, the entire output level range being limited. In this case too, the task is namely, more uniform signal compression, from that of the invention significantly different. Accordingly, the compression according to this document a goal distribution not aimed; rather, the compression rule only uses parameters derived from the input signal.

None of the documents mentioned use a training or Reference data obtained reference distribution function.

The document WHITE SA: "Restoration of non linearly distorted audio by histogram equalization" n JOURNAL OF THE AUDIO ENGINEERING SOOETY, Nov. 1982, USA, Vol 30, No. 11, pages 828-832, which is used to form the preamble of claim 1 discloses a noise suppression method in which the Amplitude distribution of an undisturbed signal is used as an aid. However assuming a complex modeling of the logarithmic histogram.

It is therefore an object of the invention to demonstrate a method for noise suppression, that the impairment of the signal field by the noise with regard to the following Evaluation, especially a classification, reliably reduced. Furthermore should the noise suppression without further knowledge of the properties of the noise and without a simulation of background noise can be carried out.

The object is achieved by a method of the type mentioned at the outset, in which according to the invention for the modification of the signal level values based on a division the range of values of the signal levels into a number of level ranges for each level range to a first level representing this level range using the Distribution function and the value of the reference distribution function at the first level second level is selected for which the value of the distribution function is the same Value as close as possible to the reference distribution function, and those signal components, whose signal level falls between the first and the second level, the value of the first level is assigned.

This solution enables noise suppression for both additive and convolutive Noise background as well as for mixed forms or more complicated disturbances. The effect of the disturbance on the signal parameters can be achieved by the method according to the invention of the signal field can be considerably reduced, even without further knowledge of Rauschparametem.

The requirement that the sequence of signal components in terms of their energy levels remains unchanged, means that for each (any) pair of signal components, for which is the original level of the first component less than that of the second, after the assignment of modified levels to the signal components, the modified level the first component is not greater than (or equal to or less than) the modified level of the second component.

It should be noted that there are no indications from the above writings indicate that a modification based on a reference distribution function without Considering the nature of the noise could be successful.

The parameter essential for the method according to the invention, the reference distribution function, can be e.g. be determined with the help of experiments. When a Training or comparison set of patterns is available, these or a selected one Part of these patterns are used to generate the reference distribution function. advantageously, can then be used as a reference distribution function from a distribution function that for function has been determined using a set of reference patterns become. The distribution function of the reference pattern set itself can function as a reference distribution function be used, or one of them, e.g. by simplifying the course of the curve, won function of the level.

• zu einem diesen Pegelbereich repräsentierenden, ersten Pegel unter Anwendung der Verteilungsfunktion und des Werts der Referenzverteilungsfunktion an dem ersten Pegel ein zweiter Pegel ausgewählt wird, für welchen der Wert der Verteilungsfunktion dem genannten Wert der Referenzverteilungsfunktion möglichst nahe kommt, und
• jenen Signalkomponenten, deren Signalpegel zwischen dem ersten und dem zweiten Pegel fällt, der Wert des ersten Pegels zugewiesen wird.

Dies erlaubt eine möglichst weitgehende Anpassung des Signals an die Referenzverteilungsfunktion. Im einfachsten Falle der Aufteilung des Signalpegel-Wertebereichs in Pegelbereiche wird für jeden auftretenden Signalpegel ein eigener Bereich zugeordnet, sodass jeder Pegelbereich mit dem zugehörenden Signalpegel identifiziert werden kann.The signal level values are advantageously modified by starting from a division of the value range into a number of level ranges for each level range

• for a first level representing this level range, using the distribution function and the value of the reference distribution function at the first level, a second level is selected for which the value of the distribution function comes as close as possible to the mentioned value of the reference distribution function, and
• those signal components whose signal level falls between the first and the second level are assigned the value of the first level.

This allows the signal to be adapted as far as possible to the reference distribution function. In the simplest case of dividing the signal level value range into level ranges, a separate range is assigned for each signal level that occurs, so that each level range can be identified with the associated signal level.

Furthermore, a particularly expedient implementation of the invention for a time and / or frequency-dependent spectrum of an acoustic signal realized signal field executed.

Fig. 1

ein Spektrogramm einer Äußerung unter geräuschfreien Bedingungen;

Fig. 2

die Energieverteilungsfunktion zu dem Spektrogramm der Fig. 1;

Fig. 3 und 4

ein Spektrogramm und die zugehörende Energieverteilungsfunktion einer Äußerung mit Geräuschhintergrund;

Fig. 5 und 6

ein Spektrogramm und die zugehörende Energieverteilungsfunktion, welche sich durch spektrale Subtraktion aus dem Spektrogramm der Fig. 3 ergeben;

Fig. 7

eine Referenzverteilungsfunktion zur Anwendung der Erfindung;

Fig. 8 und 9

ein Spektrogramm und die zugehörende Energieverteilungsfunktion, welche sich aus dem Spektrogramm der Fig. 3 mittels der erfindungsgemäßen Rauschreduktion anhand der Referenzverteilungsfunktion der Fig. 7 ergeben.

The invention is explained below using an exemplary embodiment which relates to the speech recognition of a spoken word in a motor vehicle. The attached figures are used, which show:

Fig. 1

a spectrogram of an utterance in silent conditions;

Fig. 2

the energy distribution function to the spectrogram of Fig. 1;

3 and 4

a spectrogram and the associated energy distribution function of an utterance with a noise background;

5 and 6

a spectrogram and the associated energy distribution function, which result from spectral subtraction from the spectrogram of FIG. 3;

Fig. 7

a reference distribution function for applying the invention;

8 and 9

a spectrogram and the associated energy distribution function, which result from the spectrogram of FIG. 3 by means of the noise reduction according to the invention using the reference distribution function of FIG. 7.

Speech signals which are generated against a background of noise, e.g. that inside one Motor vehicle vehicles in operation, are spoken by noise, from various sources, e.g. the vehicle engine, other vehicles, wind etc., and often a mixture of sound components of high energy with unpredictable statistics regarding their timing and frequency. The performance of speech recognition systems therefore quickly decreases when the background noise increases, for example because the vehicle speed is higher becomes. The exemplary embodiment of the invention shown below relates to detection the English words 'zero', 'one', 'two', etc. to 'nine' for the digits 0 to 9 using of a speech recognition system in a car of the small car type.

1 shows a spectrogram S1 of a spectrum for an utterance of the English word 'seven', spoken by a male speaker in the car under noiseless Conditions.

In the spectra treated in the exemplary embodiment, the time axis records a time period of 0.992 s, which is divided into 31 frames T of the same duration (so-called 'frames'). The frequency range extends from f = 200 Hz to 3.4 kHz and is in 9 bands F with roughly logarithmically graded bandwidth and spacing. The spectral energy is logarithmic in all figures as energy level E, with the unit dB and related to a basic level common to all figures.

Spectra of this type were used in the applicant's speech recognition attempts for utterances used about the vocabulary mentioned. In the speech recognition system used takes place after preprocessing the utterance to be recognized by means of a Noise suppression as explained in more detail below is a classification in which a layered neural network, which has been trained with a training vocabulary was used as a pattern recognition system. For the training vocabulary the Vocabulary of a number of speakers - advantageously both male and female female persons - in an environment that corresponds to the speaking environment of the car, spoken, for each word several times under noise-free conditions of the background noise (rest of the car).

FIG. 2 shows the energy distribution function P1 (E) for the spectrum S1 shown in FIG. 1 , An energy distribution function P (E) assigned to a spectrum S gives as a function of Energy level E on how many of the spectral components S (T, F) of the spectrum in question S have an energy level which is lower than the specified energy level E, this number as a value between 0 and 1 based on the total number of spectral Components is expressed. For example, the energy distribution function has P1 at 48 dB the value 0.6, because 60% of the energy levels of spectrum S1 are below 48 dB. A large (small) slope in the energy distribution function P (E) corresponds to an energy level, whose value in a large (small) number of components of the associated Spectrum S occurs. An energy distribution function can also be used for a large number of Spectra are determined and then gives the proportion of the components of all spectra with energy level below the specified level E divided by the total number of components of all of these spectra.

3 shows the spectrogram S2 for an utterance of the word by the same speaker at a car speed of 113 km / h (70 mph). As from the comparison of the Spectrograms S1 and S2 (Fig. 1 and 3) visible, only the speech components remain high energy little affected, while the rest of the noise are masked. The background energy level increases from about 25 dB to about 65 dB, the peaks of the utterance are at 85 dB, the speech components below 70 dB go in the noise background under. The associated power distribution function P2 (E) is in FIG. 4 shown.

The energy distribution functions P1 and P2 (Fig. 2 and 4, respectively) show that the spectral distribution of the noise-free signal S1 is significantly different from that of the noisy one Signal S2 is in which the background energy is approximately 40 dB higher than in the case of the noise-free signal.

E = S(T,F) und

E_r = S_r(T,F)

"subtrahiert" wird. Die Rauschreduktion nach der spektralen Subtraktion wurde im Rahmen der weiter unten beschriebenen Versuche der Anmelderin für das Spektrum S2 durchgeführt. In Fig. 5 und 6 sind das Spektrum S3 = s_s( S2, S_r ), das sich bei der Anwendung der spektralen Subtraktion auf das Spektrogramm S2 ergibt, und die zugehörende Energieverteilungsfunktion P3 dargestellt; dabei wurden jene Parameter b und α verwendet, bei denen die Ergebnisse von durchgeführten Spracherkennungstests für verschiedene Parameter b und α am besten waren, sowie ein aus der Aufnahme der Äußerung S2 gewonnenes Referenzrauschen S_r. Wie aus Fig. 5 und 6 ersichtlich ist, ist das Hintergrundrauschen um ca. 10 dB niedriger als im unbehandelten Signal S2, jedoch ist ein beträchtlicher Anteil der Sprachanteile niedriger Energie immer noch vom restlichen Rauschen verdeckt. Daher verbessert sich die Erfolgsquote bei der Spracherkennung nur geringfügig.A noise reduction of the noisy signal can be achieved by means of the spectral subtraction according to SV Vaseghi and BP Milner mentioned at the beginning. According to what has been said above, the spectrum S is transformed using a reference noise signal S _r in that in each spectral component S (T, F) the corresponding component S _r (T, F) of the reference noise according to the expression S '(T, F) = E0 = s s (E, E r ) = (E b - α E r b ) 1 / b . in which

E = S (T, F) and

E _r = S _r (T, F)

is "subtracted". The noise reduction after spectral subtraction was carried out for the spectrum S2 in the course of the applicant's experiments described below. 5 and 6 show the spectrum S3 = s _s (S2, S _r ), which results when the spectral subtraction is applied to the spectrogram S2, and the associated energy distribution function P3; those parameters b and α were used for which the results of the speech recognition tests performed were best for various parameters b and α, and a reference noise S _r obtained from the recording of the expression S2. As can be seen from FIGS. 5 and 6, the background noise is approximately 10 dB lower than in the untreated signal S2, but a considerable proportion of the speech components of low energy are still covered by the remaining noise. Therefore, the success rate in speech recognition only improves slightly.

Since the signal used as the reference noise signal S _r only statistically matches the noise which is present as the background of the noisy signal S2, the spectral subtraction achieves a reduction in the noise level only on individual components of the resulting spectrum S3. Depending on the relative phase position of the reference noise and the actual background, only a part of the components of the spectrum is canceled out, the noise component of the component in question, in other components the level remains approximately the same, in some cases there is even an amplification (albeit whose effect is mitigated due to the logarithmic representation of the energy level). This can be seen in FIG. 5 in particular from the low-level components starting from time frame 20.

According to the invention, the noise suppression takes place for the present speech signal S2 using a given "template function", namely one as a reference serving power distribution function. This is advantageously done in such a way that the level of the spectral components of the speech signal spectrum S2 adapted to the template function become. The energy distribution function of the resulting spectrum is then correct essentially match the template function.

Ideally, the energy distribution function would be the sum of those as a reference function Spectra are used which are used in training the speech recognition system for the the relevant word (here 'seven') can be used; because the word to be recognized is the speech recognition system is naturally not known in advance, this is not possible. It Instead, an energy distribution function is selected as a template function, which in Relative to the entirety of the words of the vocabulary to be recognized is appropriate. For example, that energy distribution function can be used as template function P0 which have been derived from the spectra of the entire training vocabulary.

The noise suppression according to the invention by adapting the levels to a template function is carried out in such a way that spectral components whose level E = S (T, F) is originally the same have a common level E0 = S '(T, F) even after the adaptation, ie for the adaptation condition applies to all spectral components S '(T 1 , F 1 ) = S '(T 2 , F 2 ) if S (T 1 , F 1 ) = S (T 2 , F 2 ).

Furthermore, the sequence of the components should not be changed with regard to their energy levels, ie S '(T 1 , F 1 ) ≤ S '(T 2 , F 2 ) if S (T 1 , F 1 ) <S (T 2 , F 2 ); this monotonous condition preserves the structures of the spectrum, at least qualitatively, when the spectrum S is suppressed into a modified spectrum S '.

The noise suppression can be fully described as a consequence of the adaptation condition (1) by an adaptation function R (E) which assigns a modified level E0 = R (E) to each original level E, to which those spectral components are reduced (or increased) that were originally had the level E. The fitting function is monotonic due to the monotony condition (2), ie R (E ₁ ) ≤ R (E ₂ ) if E ₁ <E ₂ . According to the invention, this adaptation of the spectrum takes place in such a way that P0 (E0) = P (E) applies to the assigned energy distribution function. The adaptation function R (E) is therefore clearly determined by comparing the energy distribution function P2 of the present signal with the reference function P0. Since the energy distribution functions P, P0 are also monotonically increasing functions, the adaptation function can be formally determined from this by reversing the reference function P0.

Table 1 shows an exemplary program pseudo code through which the invention Adaptation of a spectrum takes place. The spectrum S to be adjusted is here in the Field variables S stored, which over the intervals Tmin .. Tmax and Fmin .. Fmax des Time-frequency space is defined. The energy levels of the spectrum can be discrete values assume in the range of values between the energy levels Emin and Emax. In the field variable A reference energy distribution function is specified as a reference function P0. The energy distribution functions are as fields over the given interval Emin. , Emax Are defined.

PS, deren zugeordneter Energiepegel über diesem Pegelwert liegt, werden inkrementiert. Hierbei bezeichnet inc die Inkrementierfunktion.First (from the brand PS / S) the associated power distribution function is determined and stored in the field variable PS. For this purpose, the level value is determined for each component S [T, F] of the spectrum, and all components of the energy distribution function

PS whose assigned energy level is above this level value are incremented. Here inc denotes the increment function.

Then (from the brand RED / S) is in a for loop for each of the discrete values E0, provided that at this level the energy distribution function PS [E0] is smaller than the template function P0 [E0], the following steps are carried out: First the level value E0 assigned energy level E0 + dE determined. This happens because the distance dE this level is incremented from the value 0 (while loop) until the Value of the energy distribution function at the assigned level PS [E0 + dE] the value of the Template function comes closest to the given level value P0 [E0]. For this, the Function abs used to determine the absolute amount. The one after the while loop Decrementing step dec (dE) is used to correct the value for which the condition mentioned actually applies. Now the level value E0 represents the modified one Level to the energy level E0 + dE. Then it is checked whether the level difference dE is positive (greater than 0); in this case all components S [T, F] of the spectrum, whose energy level falls in the interval between E0 and E0 + dE, to the energy level E0 posed. After the last run through the outer for loop, the field S contains the invention noise suppressed spectrum S '.

7 shows the template function P0 (E0) used in the exemplary embodiment, namely the Energy distribution function for the above training vocabulary, d.s. the English Numerals 'zero' to 'nine'. For the noisy utterance S2, the result according to the invention Noise suppression with the aid of the mentioned template function P0 as the spectrogram S4 8 shown spectrum; the associated energy distribution function P4 is in FIG. 9 played.

To reduce the effort involved in carrying out the method according to the invention A level range of the original spectrum can be treated together in this way be that the associated spectral components have a uniform modified level is assigned. This modified level is related to a representative level value the relevant level range, e.g. the mean of the level range or the median the level across the components falling within the level range as described above determined, for example by means of the adaptation function.

During the first speech recognition attempts carried out by the applicant with the above described speech recognition system, the inventive method was tested and at the same time compared with the method of spectral subtraction. The ones to be recognized Utterances were spoken under different background noise conditions, namely driving at 80 km / h (50 mph) and at 113 km / h (70 mph). It was here counted the events in which the speech recognition system misrecognized the utterance has, whereby only substitution errors were taken into account. In a control series in which the Signals without noise reduction were fed to the pattern recognition, 30% of the Expressions misrecognized. When using spectral subtraction as a noise reduction method The proportion of incorrect detections decreased to 23.3%. With the invention Procedures, the percentage of errors decreased to 13.3%, i.e. a reduction the error rate by almost half compared to the known method.

The method according to the invention is particularly suitable for suppressing superimposed ones Disorders that the monotonic relation of the spectral components of the utterance do not or interfere only slightly. Such disturbances include e.g. white noise, one linear or nonlinear amplification or attenuation of the entire spectrum as well various phenomena of the Lombard effect, which is known to change the Voice and pronunciation depending on the mental state of the speaker, e.g. Stress, describes.

8 is around time frame 16 in the upper frequency bands an artifact recognizable which is not contained in the actual utterance (FIG. 1) and was not eliminated by the method according to the invention. Such artifacts can be found in in most cases e.g. with the help of a median filter following the noise suppression be eliminated.

The method of noise suppression according to the invention changes what is to be processed Signal even in the absence of noise, since the template function P0 generally from the energy distribution function of the undisturbed utterance is different. This can may create a source of noise-free detection errors. To do this To avoid, for example, training the speech recognition system with the help of spectra that are already performed with the method according to the invention the template function used has been adjusted. The training vocabulary can contain these spectra instead of or together with the original spectra.

Another approach is to use the method according to the invention only if if the presence of noise is detected, e.g. in the period shortly before the statement; otherwise, the speech signal of speech recognition without noise suppression fed. This approach does not require an estimate of the noise beyond that Detection of noise would go out.

In a simplified variant of the method according to the invention, the adaptation of the Spectrum can be significantly simplified by the fact that only a fixed number of Parameters of the template function are used, and the adjustment with regard to these parameters are done. For example, the mean and the spread of the distribution of the Template function can be used. The mean and Scattering the distribution of the energy distribution function is determined, and from the comparison this parameter with those of the template function becomes a linear transformation for the Energy level of the spectrum determined. By applying this linear transformation there is a modified spectrum in which the disturbing effect of the background noise is significantly reduced. Unless the application of a linear transformation is not is sufficient, e.g. a higher order transformation can be used which results from the Comparison of a corresponding number of parameters of the energy distribution function and the template function, e.g. higher moments of the distributions.

The method according to the invention is not only suitable for reducing interference for acoustic signals, e.g. Speech signals; rather, it can also be used for other types of patterns be used, which is characterized by a one-dimensional or multi-dimensional field applied feature size can be described. Accordingly, possible areas of application are e.g. character recognition in written, text or the like, reconstruction and / or evaluation of pictures etc.

Claims (2)

1. A method for suppressing spurious noise in a signal field (S2) containing a plurality of signal components which each adopt a value of a signal level and may be assigned to an ordinate range, where from the signal field (S2) a distribution function (P2(E)) is determined, which, as a function of the signal level, indicates for each of its possible argument values of signal level (E), the size of the fraction of those signal components whose signal level is lower than the argument value
   the signal level values of the signal field (S2) are modified in such a manner that the distribution function (P4(E)) of the modified signal field (S4) equals a predetermined reference distribution (P0(E)), wherein the sequence of the signal components remain unchanged with regard to their energy level and signal components, whose original signal levels are identical, are assigned the same modified signal levels, and a function obtained from a distribution function that was determined for a set of reference models is used as reference distribution function (P0),
   characterized in that
   to modify the signal level values, starting from a division of the value range of the signal levels into a number of level ranges, for each level range

   to a first level (E0) representing the respective level range, using the distribution function (P2) and the value of the reference distribution function at the first level (P0(E0)), a second level is selected for which the value of the distribution function (P2(E)) is as close as possible to the said value of the reference distribution function (P0(E0)), and

   those signal components, whose signal levels fall between the first and the second level, are assigned the value of the first level (E0).
2. The method of claim 1, characterized in that it is carried out for a signal field realized as a time and/or frequency dependent spectrum of an acoustic signal.

Images