DE102009029615A1 - Method for processing audio data of e.g. guitar, involves removing spectral property from spectrum of audio data, and impressing another spectral property on audio data, where another spectrum is formed corresponding to latter property - Google Patents

Abstract

The method involves removing a spectral property from a spectrum of audio data. Another spectral property is impressed on the audio data, where another spectrum is formed corresponding to the latter spectral property. A minimal-phase filter is generated with cepstral-transformation of a linear-phase filter. The spectrums are determined by analyzing the audio data by interpolating maxima of a spectral representation of the audio data. A power intensity spectrum is generated from the spectral representation of the audio data before the interpolation. Independent claims are also included for the following: (1) an arrangement for processing audio data, comprising a chip (2) a computer program comprising a set of instructions to perform a method for processing audio data.

Description

The invention relates to a method and an arrangement for processing audio data as well as a corresponding computer program and a corresponding computer-readable storage medium, which can be used in particular in the field of audio software and sampling.

The sounds of natural instruments depict their volume and timbre depending on the intensity of the attack. The stronger a note is played, the louder and brighter it sounds. This essential feature of expression contributes significantly to the sound impression of an instrument.

In previous samplers, sounds are often only varied in their volume depending on their playing strength, resulting in a static and therefore very unnatural sound impression. In some cases, simple filter structures are also used which, for example, filter out overtones via low-pass filters for quiet sounds. However, this also improves the sound impression only insignificantly.

Other conventional methods use a variety of samples, each used for different Anspielstärken. In the transition between these samples, however, generally audible sound differences arise. To avoid this is often faded between the samples, but this has the disadvantage that two samples sound at the same time. This creates beats, which also creates an unnatural sound impression.

Natural instruments imprint a specific formant structure on all sounds that can be generated by them, which is determined by specific resonances. These resonances remain independent of the fundamental frequency of a played sound and are an essential characteristic of the instrument. When natural instruments are sampled, this characteristic is reflected in each individual audio sample. The influence of this characteristic is referred to below as a spectral fingerprint.

If a sound is directly transposed within a sampler, the formant structure is also transposed, which leads to unnatural sound results. To prevent this, a formant correction must be used. In the case of the invention, this ensures that the formantresonances are leveled by means of filters before the transposition and are introduced again after the transposition.

In the field of the invention, methods are also used in the prior art to manipulate the tone colors of the individual samples within a sampler and to convert them into one another. Some of these methods use a filter bank of EQ filters (EQ = equalizer). These are implemented in IIR technology (IIR = Infinite Impulse Response) and represent as far as possible a classical filter bank, as it is also known from mixing consoles or effect devices. The setting of the filters is made in this solution by the user.

Furthermore, applications are known for the mastering of complex music signals, wherein a finished mix an EQ curve of another signal is impressed. The filter used here is also an FFT-based FIR (Fast Fourier Transformation, FIR) filter, whereby only a single filter is used in this solution. The analysis in these applications is based on complex music signals and therefore does not analyze overtones and associated formant structures. An alternative solution for mastering complex music signals is to convert a spectrum that has been analyzed via an FFT into a 30-band filter bank.

The object of the invention is therefore to provide a method and an arrangement for processing audio data as well as a corresponding computer program and a corresponding computer-readable storage medium which eliminate the disadvantages of the known solutions and in particular produce a natural sound impression in synthetic instruments.

This object is achieved by the features in claims 1, 2, 20 and 25 to 27. Advantageous embodiments of the invention are contained in the subclaims.

One particular advantage of the invention consists, inter alia, in the fact that the range of variation of the timbre is mapped depending on the attacking strength, without accepting the abovementioned disadvantages. This is achieved in that in the method for processing audio data according to the invention at least a first spectrum of at least one first spectral property is removed from the spectrum of the audio data. Preferably, the at least one first spectrum of the first spectral characteristic is filtered out of the spectrum of the audio data. After the at least one first spectrum of the first spectral property has been removed from the spectrum of the audio data, in a first alternative of the invention at least one second spectrum of at least one second spectral property is impressed on the spectrum of the audio data. This imprinting is preferably carried out using a filter.

The at least one first and / or second spectral property may, for example, be the timbre of a particular instrument. In this way, the sound of another sound can be impressed on the audio data without, as in the conventional solutions, for. B. beats caused by the blending of two samples.

Another alternative of the invention provides that after removing the at least one first spectrum of the at least one first spectral property, the spectrum of the audio data is changed, for. B. by the frequency or pitch of the audio data is changed. After such a change in the spectrum of the audio data, at least part of the at least one first spectrum of the at least one first spectral property and / or at least one second spectrum of at least one second spectral property is impressed on the altered spectrum. This ensures that an audio sample can be changed in terms of pitch, volume or other sound parameters, in each case the natural timbre is achieved by the impressed spectrum of the spectral property, so that a natural sound impression is created. With the same method, for example, an audio sample of a piano, the tone (sound characteristic) of a guitar can be impressed.

The first and second spectral properties may, for example, be the spectral property at the beginning or end of a desired or realizable variation width of the audio data. For example, the first spectral characteristic can represent the timbre of very soft or very deep sounds and the second spectral feature the timbre of very loud or very high sounds. In imposing the first and second spectral characteristics, the spectra are preferably weighted depending on where the variation width of the audio data is the target sample. For example, if the target sample is close to the quiet end of the range of variation, then the at least one spectrum representing the tones of soft sounds is weighted higher, and the at least one spectrum representing the tones of loud sounds is weighted less. The weighting can be done for example by specifying a corresponding Anspielstärke.

As mentioned, in a preferred embodiment the spectra of the (at least one) first spectral property are removed by filtering from the spectrum of the audio data and the spectra of the (at least one) second spectral characteristic are impressed on the spectrum of the audio data. In an advantageous embodiment of the invention, to remove the at least one first spectrum, the inverted at least one first spectrum is used as the filter spectrum. In another advantageous embodiment of the invention, it is provided that the spectrum of the spectral property to be imprinted is used directly as the filter spectrum for imprinting, optionally weighted spectrum. A preferred embodiment of the invention provides that the filter spectra are transformed into the time domain, for example by an IFFT. The transformed into the time domain spectra form the impulse response. It proves to be advantageous if the impulse response represents a linear-phase filter. It may also prove advantageous if a minimum-phase filter is used to remove and / or impress the at least one first or second spectrum of the first or second spectral property. Such a minimal-phase filter can be obtained from the linear-phase filter by cepstral transformation.

A preferred embodiment of the invention provides that the at least one first spectrum of the at least one first spectral property and / or the at least one second spectrum of the at least one second spectral property is obtained by analyzing audio data. The audio data to be analyzed can be, for example, the already mentioned samples at the beginning or end of the variation width of the audio data. In this case you would z. B. by the analysis, the tone of very soft and / or very deep (beginning of the variation) or very loud and / or very high (end of the range of variation) received. The analysis preferably includes the determination of harmonics.

In a preferred embodiment it is provided that at least part of the analysis takes place in the spectral range. The analysis in the spectral range preferably takes place using a Fourier transformation, in particular a short-time Fourier transformation (STFT). In this case, parts of the audio data (sample) can be subjected to this analysis, but preferably the STFT is applied to the entire sample.

A further preferred embodiment of the invention provides that the determination of the harmonics is based on (local) maxima (peaks) in the spectrum of the audio data. It proves to be advantageous if an interpolation curve is laid through at least part of the (local) maxima. Preferably, the (local) maxima also remain as (local) maxima of the interpolation curve. Since the harmonics are imaged by the maxima, this form of interpolation curve detects the formant character of the analyzed sample. To obtain a timbre For example, a guitar can be analyzed samples of guitar sounds.

The spectra of timbre thus obtained, d. H. The first or second spectrums of the first and second spectral characteristics can now be used to impose, for example, a piano sample the timbre of a guitar, for example, by removing the timbre of the piano from the piano sample by removing the first spectrum, which in this case the spectral property 'tone color of the piano', and then imprinting the second sample, which represents the spectral property 'tone color of the guitar'.

For each sample, you can define as many Spectral Figerprints as you want, each describing the spectral character of a defined time segment.

For obtaining the normalized source signal, the source signal can be sent, depending on its playback position, by a weighted combination of the normalization filters of the different temporal sections defined as belonging to each, the outputs of which are combined on a standardization bus. An advantageous embodiment provides for the continuous cross-fading between two adjacent sections. From the normalization bus, the thus standardized source signal can be sent to any combination of different denormalizing filters. The denormalization filters can originate from other target signals, but also from the source signal itself. Combinations of the denormalization filters can, for example, map the temporal spectral course of the target signal. In this case, as with standardization, a temporal blending between adjacent sections of advantage.

• 1. Direkt am Anfang des Signals bildet sich ein prägnanter Rauschcharakter beim Einschwingen der stationären Grundschwingung aus.
• 2. Direkt nach dem Einsetzen der lauten, stationären Grundschwingung ist die Klangfarbe recht hell und obertonreich.
• 3. In der Mitte des Signals zeigt sich die Klangfarbe aufgrund der durch das Decrescendo nachlassenden Lautstärke weniger obertonreich.
• 4. Kurz vor dem Ende des Klangs ist die Klangfarbe aufgrund der durch das Decrescendo nachlassenden Lautstärke recht dumpf und hat kaum noch Obertöne.

As an example application here can serve a saxophone sound, with which a Decrescendo is played. Here the following four fingerprints could be defined:

• 1. Directly at the beginning of the signal forms a peculiar noise character when settling the stationary fundamental.
• 2. Immediately after inserting the loud, stationary fundamental, the timbre is quite bright and rich in overtones.
• 3. In the middle of the signal, the timbre is less overtone-rich due to the declining volume of the decrescendo.
• 4. Shortly before the end of the sound, the timbre is quite dull due to the declining sound of the decrescendo and barely has overtones.

In order to obtain an optimally adapted, normalized source signal, the source signal can be weighted in a time-dependent manner into the standardization filter belonging to the various fingerprints.

For the denormalization, the signal could now be routed from the normalization bus in real time, depending on the desired effect, weighted back into the denorming filter. Thus, for example, in the middle of the signal again continuously in the sound impression of the transient phase could be blended.

Instead of subdividing the signal into a few sections (as in the previous example, the envelope sections), the signal can also be detected completely over its entire duration. For this purpose, separate spectra are generated for all adjoining periods. Also, this "spectral scanning" can be done with overlapping time ranges, which further increases temporal resolution.

In a further preferred embodiment of the invention, it is provided that the samples resulting from the Fourier transformation, in particular the samples resulting from the STFT, are converted into power density spectra. It proves to be advantageous if the power density spectra are averaged and converted back into an averaged spectrum. Furthermore, it proves to be advantageous if the spectral representation of the audio data before the evaluation of the peaks for the detection of overtones is converted into a constant-Q spectrum. Preferably, the conversion is from the averaged spectrum, i. H. the result of the return transfer from the averaged power density spectra. Further preferably, the conversion is energy conserving.

An arrangement according to the invention comprises at least one chip and / or processor and is arranged such that a method for processing audio data can be executed, wherein at least a first spectrum of at least one first spectral property is removed from the spectrum of the audio data and the spectrum of the audio data at least a second spectrum of at least one second spectral property is impressed.

A preferred embodiment of the invention provides that the arrangement comprises at least one filter for removing the at least one first spectrum and / or at least one filter for impressing the at least one first spectrum and / or the at least one second spectrum.

A further preferred embodiment of the invention provides that the outputs of the at least one filter for removing the at least a first spectrum and the inputs of the at least one filter for impressing the at least one first spectrum and / or the at least one second spectrum are connected via a bus.

It also proves to be advantageous if the arrangement is set up such that the filtering takes place at least in part by means of rapid folding in the overlap-add and / or overlap-save method.

A computer program for processing audio data allows a data processing device, after being loaded into the memory of the data processing device, to perform a method for processing audio data, wherein at least a first spectrum of at least one first spectral property is removed from the spectrum of the audio data and the spectrum of the Audio data at least a second spectrum of at least one second spectral property is impressed.

In a further preferred embodiment of the invention, it is provided that the computer program according to the invention is modular in construction, with individual modules being installed on different data processing devices.

Advantageous embodiments additionally provide computer programs by which further method steps or method sequences specified in the description can be executed.

Such computer programs can be made available for download (for a fee or free of charge, freely accessible or password-protected) in a data or communication network, for example. The computer programs thus provided can then be made usable by a method in which a computer program according to claim 25 is downloaded from an electronic data network, such as for example from the Internet, to a data processing device connected to the data network.

In order to carry out the method according to the invention for processing audio data, it is provided to use a computer-readable storage medium on which a program is stored which, after being loaded into the memory of the data processing device, allows a data processing device to carry out a method for processing audio data. wherein at least a first spectrum of at least one first spectral property is removed from the spectrum of the audio data and the spectrum of the audio data is impressed on at least one second spectrum of at least one second spectral property.

Thus, with the invention, sounds can be dynamically manipulated in their timbre by imprinting the timbres on other sounds. This allows in particular an expression control by blending the timbres of sounds with different volume, as well as a Formantkorrektur by blending the timbres of sounds with different pitch.

The invention represents a very accurate Formantkorrektur, which stands out from other conventional methods, since it treats overtones much more accurate.

In summary, the invention can be described as follows:
Each sample is assigned one or more spectra of a spectral property, also referred to below as spectral fingerprints, which describe the timbre as spectral states of the sample. The spectral fingerprints are determined in an analysis step.

The tone color manipulation is then done in a two-step process using a normalization filter and a denormalization filter.

The source sample (eg a sample of a first instrument) is deprived of the timbre described by a spectral fingerprint via a normalization filter, by leveling all the resonances. This results in a signal whose harmonics have a level of about 0 dB. The sound impression of this normalized sample is reminiscent of white noise, which was filtered with a comb filter.

A denormalization filter can then be used to impress on this normalized sample a timbre of a target sample described by another spectral fingerprint (eg a sample with a different timbre, which is caused, for example, by another intonation or a sample from a second instrument). Here, the harmonics are set to the level characteristic of the sound impression.

Spectral fingerprints:

• – Short-Time Fourier Transform (STFT) des gesamten Samples;
• – Überführung der einzelnen Spektral-Frames in Leistungsdichtespektren (LDS);
• – Mittelung der LDS unter Vernachlässigung von Frames mit zu geringer Gesamtleistung;
• – Rücküberführung der gemittelten LDS in ein mittleres Spektrum;
• – energieerhaltende Überführung des linearen Spektrums in ein Constant-Q-Spektrum;
• – Obertondetektion als Maxima des Spektrums in festgelegten Bereichen rund um die Vielfachen des Grundtons;
• – Interpolation der einzelnen Peaks, wobei die Stützstellen exakt wiedergegeben werden.

The extraction of the spectral fingerprints takes place in an exemplary embodiment via the following analysis process:

• Short-time Fourier Transform (STFT) of the entire sample;
• - transfer of the individual spectral frames into power density spectra (LDS);
• - averaging the LDS neglecting frames with too low overall performance;
• - return of averaged LDS to a medium range;
• - energy conserving conversion of the linear spectrum into a constant Q spectrum;
• - overtone detection as maxima of the spectrum in defined ranges around the multiples of the fundamental;
• - Interpolation of the individual peaks, whereby the interpolation points are reproduced exactly.

From a technical point of view, the spectral fingerprints obtained by this process thus represent classical amplitude spectra which can be further processed by known methods of FFT or filtering.

(De-) Scaling filter

The structure of the filter used for normalizing and denormalizing is identical except for a deviation in the processing of the spectral fingerprints:

• – als Filterspektrum für das Denormierungsfilter das Spektrum des Spectral Fingerprints H(jw) direkt verwendet wird, und
• – als Filterspektrum für das Normierungsfilter hingegen das invertierte Spektrum des spektralen Fingerprints H(jw)^–1 verwendet wird.

The deviation is that

• - the spectrum of the spectral fingerprint H (jw) is used directly as filter spectrum for the denormalization filter, and
• On the other hand, the inverted spectrum of the spectral fingerprint H (jw) ^{-1 is} used as the filter spectrum for the normalization filter.

• – das (Filter-)Spektrum wird über eine IFFT in den Zeitbereich zurücktransformiert und bildet die Ausgangsimpulsantwort;
• – die Ausgangsimpulsantwort wird vorzugsweise um die halbe Blocklänge verschoben, um die Energie der Impulsantwort in der Mitte zu konzentrieren.
• – Soll die Impulsantwort gekürzt werden, geschieht dies vorzugsweise symmetrisch von den Blockgrenzen her, wenn nötig auch mit einer anschließenden, zusätzlichen Fensterung zur Glättung der Blockgrenzen.
• – Die so entstandene Impulsantwort stellt ein linearphasiges Filter dar und kann über die reellwertige Cepstraltransformation in ein minimalphasiges Filter überführt werden.

The same processing steps include the following:

• The (filter) spectrum is transformed back into the time domain via an IFFT and forms the output impulse response;
• The output impulse response is preferably shifted by half the block length to concentrate the energy of the impulse response in the middle.
• If the impulse response is to be shortened, this is preferably done symmetrically from the block boundaries, if necessary also with a subsequent, additional windowing for smoothing the block boundaries.
• - The resulting impulse response represents a linear-phase filter and can be converted via the real-valued cepstral transformation into a minimal-phase filter.

filtering

By subdividing the filtering into the two steps of normalization and denormalization, the described tone color transformation can be carried out with reasonable effort. This allows for real-time processing, which is impossible with the use of a single filter or made possible only by the preceding calculation of all possible individual transformation filters. However, this leads to a computational effort and storage requirements square or higher order.

In a preferred embodiment of the invention, the filters are connected in such a way that the outputs of all normalization filters go to a normalization bus. This normalization bus is connected to all inputs of the denormalization filters. The actual filtering takes place in all filters by means of fast convolution, for example in the overlap-add or overlap-save method.

Due to the design with denormalization and normalization filters, an ideally noisy signal is present on the standardization bus at all times. This can be converted in real time proportionally over any other Denormierungsfilter in a different timbre. This real-time capability can only be realized by the setup with a standardization bus. If this bus were not available, any possible tone color transformation would have to be known in advance and implemented in a separate filter. However, this would only be possible to a very limited extent due to the increasing effort.

The quality of the standardization depends on the constancy of the respective timbre. For example, a (de) crescendo or a strong vibrato does not have a uniform timbre or only an approximate uniform timbre. However, the invention allows to be arbitrarily accurate by the above-mentioned use of multiple fingerprints in the normalization.

Within a sampler, the inventive method can be used to standardize the transitions between different samples in terms of different pitches and different Anspielstärken. Classically, the samples are arranged within a sampler in a sample map, with different pitches arranged horizontally and different pitch strengths vertically. The moment you press the key, the pitch and the attacking power are now clearly defined so that a particular sample can be played from the map.

If the sampled sample in its timbre is now transferred to another sample, a crossfade of the time signals is often used. However, this leads to beats. To adjust the timbre, often too However, they are clearly inferior to the normalization and denormation filters according to the invention.

The inventive method can be placed on the sample map an additional fingerprint map. This consists of fingerprints, each sample being assigned one or more fingerprints. If only one fingerprint is assigned, this describes the timbre of the entire sample. If several fingerprints are assigned, they each describe the timbre of a signal segment or time. Each fingerprint is in this case transferred in advance to a normalization and a denormalization filter.

If a note is played now, the pitch and the sticking strength are fixed at the moment of pressing the key as in the conventional sampler. However, the sample underlying these parameters will not play directly. Rather, the source signal (time-dependent) is proportionally fed into the normalization filter corresponding to its position within the fingerprint map, the outputs of which converge on the normalization bus. The signal on the normalization bus is then routed proportionately to the denormalization filters that are closest to any point within the fingerprint map. This point can be modulated in real time, which directly blends the timbre and constitutes a great strength of the method according to the invention.

The invention will be explained in more detail below with reference to the figures of the drawings of various embodiments.

Show it:

1 : Illustration of an exemplary analysis of an audio sample for determining a spectral fingerprint,

2a FIG. 2 illustrates an exemplary processing of an audio sample to which a source fingerprint is withdrawn in order to obtain a normalized source signal, and FIG

2 B : Illustration of an exemplary processing of a normalized source signal to which a target fingerprint is impressed in order to obtain a transformed source signal.

Based on 1 intended to obtain a spectral fingerprint 100 be described by an exemplary analysis method.

In a first step is on the input signal 104 an STFT applied (step 106 ), creating a mid-range 108 is produced. In step 110 become the individual spectral frames of the middle spectrum 108 in power density spectra 112 transferred. In a subsequent step 114 the single spectra are averaged and converted back to a middle spectrum, and in step 116 smoothed and into a constant Q spectrum 118 transferred.

In a next step 120 the detection of the overtones takes place as peaks (maxima) of the constant-Q spectrum 118 , Through these maxima will step in 122 placed an interpolation curve, the interpolation curve runs exactly through the maxima as support points. This interpolation curve represents the spectral fingerprint 100 represents.

At the input signal 104 it can be an audio signal of any sound or instrument. In particular, it may be very quiet, very loud, very deep, very high sounds, or generally sounds that are essentially at the limits of a range of variation of the audio signal or sound.

Based on 2a and 2 B Now is the change of a spectral property, for example, by the change of the Anspielstärke, a source signal 200 be explained in more detail. At the in 2a illustrated exemplary source signal 200 it is a powerfully played sound with many overtones. This source signal 200 becomes a spectral representation 202 transferred. Subsequently, this source signal 200 a normalization filter is applied, with the spectrum being the filter spectrum 206 of the inverted spectral fingerprint of the source signal 200 is used (the fingerprint 204 himself is in 2a together with the spectral representation 202 of the source signal 200 shown).

The filtering results in a normalized source signal 208 with associated spectrum 210 , Like from the can be seen, the overtones of the spectrum 210 of the normalized source signal 208 an approximately uniform level of 0 dB. The spectrum 210 of the normalized source signal 208 can now be transformed without the effect that the spectral properties are undesirably affected by the transformation since they were previously filtered out. For example, the volume of the source signal 200 be reduced. The normalized source signal 208 is now filtered using a denormalization filter, using the spectral fingerprint as the filter spectrum 212 of the target signal is used, its spectrum 214 in 2 B is reproduced. The target signal may, as in this example, be a weakly played sound. As a result of this filtering, the transformed source signal is obtained 216 with the associated spectrum 218 , The transformed source signal 216 represents a signal to which the timbre of a softly played sound (target signal) has been impressed (in the figure by the comparatively few overtones of the transformed source signal) 216 illustrated). Then, regardless of the described method for changing the timbre, the volume of the transformed source signal 216 be reduced.

In another exemplary embodiment, it is provided that the normalized source signal 208 optionally modified by further processing steps (such as volume or pitch change) to apply at least two filters, using the spectra of different fingerprints as filter spectra. A preferred embodiment provides that these spectra are weighted. For example, if the transformation is the change in timbre according to a change in volume, and the transformed source signal 216 should lie in the middle of the range of variation, could be used as filter spectra, the spectrum of the fingerprint of the quietest tone and the spectrum of the fingerprint of the loudest tone, each weighted with about 50%.

The invention is not limited in its embodiment to the above-mentioned preferred embodiments. Rather, a number of variants is conceivable that make use of the inventive arrangement and the method according to the invention and the corresponding computer program and the corresponding computer-readable storage medium even with fundamentally different types of use.

LIST OF REFERENCE NUMBERS

100

spectral fingerprint

104

input

106

step

108

medium spectrum

110

step

112

Power density spectra

114

step

116

step

118

Constant-Q spectrum

120

step

122

step

200

source signal

202

Spectral representation of the source signal

204

spectral fingerprint of the source signal

206

inverted spectral fingerprint of the source signal

208

normalized source signal

210

Spectrum of the normalized source signal

212

Spectrum of the spectral fingerprint of the target signal

214

Spectrum of the target signal

216

transformed source signal

218

Spectrum of the transformed source signal

Claims (27)

Method for processing audio data, wherein in a first step removes at least a first spectrum of at least one first spectral property from the spectrum of the audio data and in a second step, the audio data is impressed on at least one second spectral property, the at least one second spectral property corresponding to at least one second spectrum. A method of processing audio data, wherein at least a first spectrum removes at least a first spectral property from the spectrum of the audio data, after removal of the at least one first spectrum, transforming the audio signal corresponding to the spectrum of the audio data resulting from the removal of the at least one first spectrum, and one or more of the at least one first spectral property and / or at least one second spectral property is impressed on the transformed audio signal. A method according to claim 1 or 2, characterized in that the audio signal is divided into temporal sections and for at least a portion of the temporal sections in each case a section-specific first and / or second spectrum of the first and second spectral property is defined. A method according to claim 3, characterized in that the temporal sections overlap. Method according to one of the preceding claims, characterized in that to remove the at least one first spectrum, the inverted at least one first spectrum is used as a filter spectrum. Method according to one of the preceding claims, characterized in that for impressing the at least one first spectrum and / or the at least one second spectrum, the corresponding spectra are used as filter spectra. A method according to claim 5 or 6, characterized in that the filter is generated by the inverted at least one first spectrum and / or the at least one first spectrum and / or the at least one second spectrum are transformed into the time domain. A method according to claim 7, characterized in that it is the filter is a linear phase filter. Method according to Claim 8, characterized in that a minimal-phase filter is produced by cepstral transformation of the linear-phase filter. A method according to claim 8, characterized in that at least a part of the spectra to be recorded is weighted. Method according to one of the preceding claims, characterized in that the at least one first spectrum and / or the at least one second spectrum is determined by analyzing audio data. A method according to claim 11, characterized in that the analysis comprises a detection of harmonics. A method according to claim 11 or 12, characterized in that the analysis comprises an analysis in the spectral range. Method according to one of the preceding claims, characterized in that the determination of the at least one first and / or the at least one second spectrum comprises an interpolation of maxima of a spectral representation of the audio data. A method according to claim 14, characterized in that at least one power density spectrum is generated before the interpolation from the spectral representation of the audio data. A method according to claim 15, characterized in that an averaging of the at least one power density spectrum or a part of the power density spectra is carried out. Method according to one of claims 14 to 16, characterized in that the spectral representation is converted before or after the detection of the harmonics in a constant-Q spectrum. A method according to claim 17, characterized in that the conversion into the constant-Q spectrum after a return transfer of the at least one power density spectrum is carried out in a middle spectrum. Method according to one of claims 5 to 18, characterized in that the filtering takes place at least in part by means of rapid folding in the overlap-add and / or overlap-save method. Arrangement with at least one chip and / or processor, wherein the arrangement is set up such that a method according to one of claims 1 to 19 can be executed. Arrangement according to claim 20, characterized in that the arrangement comprises at least one filter for removing the at least one first spectrum and / or at least one filter for impressing the at least one first spectrum and / or the at least one second spectrum. Arrangement according to claim 21, characterized in that the outputs of the at least one filter for removing the at least one first spectrum and the inputs of the at least one filter for impressing the at least one first spectrum and / or the at least one second spectrum are connected via a bus. Arrangement according to one of claims 20 to 22, characterized in that the arrangement comprises a sampler. Arrangement according to claim 23, characterized in that stored in the sampler first and / or second spectra of the first and second spectral characteristic, which are assigned to samples of the sampler. Computer program that allows a data processing device, after it has been loaded into storage means of the data processing device to perform a method according to any one of claims 1 to 19. A computer-readable storage medium having stored thereon a program that allows a data processing device, after being loaded into storage means of the data processing device, to perform a method according to any one of claims 1 to 19. Method in which a computer program according to claim 25 is downloaded from an electronic data network, such as for example from the Internet, to a data processing device connected to the data network.

Images