DE19949637A1 - Noise detection and separation arrangement, having neural network which forms classification signal giving indication of type of acoustic signal - Google Patents

Abstract

The arrangement includes at least one normalization unit (13) which normalizes respectively an amplitude (12) of a certain frequency range from the spectrum of the acoustic signal (1). A differentiation arrangement (15) differentiates the normalized amplitude (14) to time, and forms an absolute value of the differentiated amplitude. A neural network (18) includes input neurones which receive the differentiated amplitude or its absolute value of the certain frequency range, and output neurones which form a classification signal (22) classifying the acoustic signal according to its frequency spectrum, giving an indication of the type of acoustic signal.

Description

The present invention relates to devices for noise detection and separation and noise monitoring and methods based thereon for noise monitoring and and forecast.

In the context of automatic sound measurements without human intervention in addition to the question of how loud it is in a particular location, the problem almost always identify the relevant noise-generating sources and determine their proportions. Just in obvious cases where a source of noise dominates, this is easy. Since the Sound immissions depend heavily on propagation conditions and times of day Shares in an immission location vary considerably.

In many cases, however, it is also not possible to clearly identify the source of the noise; she must first be searched extensively. Is a supposed cause then identified so can by means of a comparison between the respective situation at on and off the supposed source are attempted to verify the assumption about the noise source ren. This method quickly reaches its limits due to the variance mentioned.

So think about the fact that aircraft noise alongside the noise from a busy car Railway should be recorded, which does not fall silent even at night when the road is busy. To in In such a case, to be safe, it would be necessary here for at least the road to block a certain period of time, which is certainly only possible in exceptional cases.  

A shutdown, i.e. blocking the road for noise measurement purposes, is already out of the question here mostly out of proportionality reasons.

Even if the individual sound sources are identified, the pro often arises blem that from a variety of different sources only one very specific Sound source to be measured or monitored. However, this can only happen if the sound source to be monitored out of the entire acoustic signal knows and can be separated from the others.

It is therefore an object of the present invention to provide a device that he possible to automatically classify an acoustic signal according to its type and thus one to enable fully automatic noise detection, here for the purposes of loading Description and claim of the present invention under classification in particular the assignment of an acoustic signal to a source or a source type should be standing.

This object is achieved according to the invention by a device for noise detection in acoustic signals with at least one neural network with new input rons, output neurons and weighted connections between the Neurons, the input neurons each receiving the amplitude of a be agreed frequency band provided from the frequency spectrum of the acoustic signal are, and the output neurons generate at least one classification signal that the classified acoustic signal according to its frequency spectrum, the classifi cation signal or the classification signals provide information about the type of acoustic signal gives or give. So there is the classification signal, in which it provides information about the type of acoustic signal, in particular information about its source. Even the one gait neurons instead of receiving the amplitude of a certain frequency band the frequency spectrum of the acoustic signal for receiving its change over time to be provided. It is also possible that they both receive the respective Amplitude, as well as their derivation are provided. The device according to the invention is therefore able to detect the acoustic signal, especially after  Sources to classify. It has been shown that an artificial neural network the frequency spectrum of the acoustic signal is good with the appropriate training method is able to classify.

Classifying artificial neural networks, such as, are preferably suitable for this Single cluster learning networks, preferably those based on the Vector method Quantization work. Also multi cluster networks, preferably those that are self-learning and / or self-organizing (such as self-organizing cards) are suitable within the scope of the present invention. As an example, one for the present inventor suitable network types, here are the Kohonen networks.

For the present Erfin come as training methods for artificial neural networks a number of methods known to the person skilled in the art, for which purpose already several literature is available (see for example: Nauck, D. et al., Neuronal Networks, Braun Switzerland / Wiesbaden 1994). Generally speaking, the training process is based on appropriate change of the weights of the connections between the neurons until the neural network satisfies its classification task lend met. The training process usually takes place in one of the actual ones the supervised training phase preceding the operating phase, in which the network only learns is being trained and is not yet in operation. Beyond that, however, also exist Training methods that let the neural network learn unsupervised, which is also a be drive learning that enables continuous improvement of performance Network can contribute. As already mentioned above, the expert is here in the In individual cases, there are a number of options for training the artificial neural network that is open in different ways more or less quickly lead to the desired goal of a network that classifies the acoustic signal, since neural networks are generally well suited for the classification of signal patterns. Acoustic signals from the same source are often similar in their frequency specs as a signal signal that comes into question here for the task at hand ster. The neural network is capable of handling the amplitudes of the different frequencies bands of the acoustic signal to be identified is then received as one of the input signals  Sound source from which the acoustic signal originates, classification signal belonging to it testify and thus shed light on the type and in particular the source of the acoustic signal to give. The number and width of the frequency bands, as well as the area they cover ken, is to be adapted to the underlying problem. Numbers of 20 or better of 40 frequency bands have proven themselves.

Experiments have shown that even a very simple artificial neural network with only two layers and without a hidden layer is already suitable for fulfilling the task. The input neurons in the first layer receive the amplitudes of the frequency bands to be considered, 20 bands covering a range of 16 Hz-20 kHz, preferably 16 Hz to 16 kHz, having proven themselves here. One input per neuron can be provided in this first layer for each frequency band, ie for each amplitude to be received. The number of neurons of this first layer (input layer) is preferably chosen so that it corresponds to the number of frequency bands used. It has also proven to be useful not to weight the received amplitudes of the inputs of the neurons of the first layer, but to feed them to the input neurons unweighted. For the input neurons themselves, the identity function (Id) has proven to be suitable as a transfer function (transition function). The input neurons are then preferably each connected to all neurons of an output layer (output layer), the number of neurons in this layer, that is to say the number of output neurons, preferably being selected to be equal to the number of sounds to be recognized. Each output neuron then preferably itself outputs an output signal (classification signal), which provides information about the type of acoustic signal. The sigmoid function (σ function) has proven itself as a transition function for the output neurons. The connections between the neurons of the input layer and those of the output layer are preferably provided with weights ω _ij (as the weight of the connection from neuron no. I of the input layer to neuron no. J of the output layer).

In order to train the network, the following procedure has proven to be quite practical: At the beginning, the weights ω _{ij are} chosen randomly according to a Gaussian distribution. The spectrum of an acoustic signal is then presented to the network via the inputs of the input neurons, of which it is known what type it is. Now the classification signals (actual classification signals) generated by the output neurons are compared with the classification signals (target classification signals) expected for this acoustic signal and a measure of the quality of the signal classification is formed from the deviation. This is preferably done by using a fitness measure as follows,

where p is the number of classification signals generated. This fitness is all the more favorable the closer it is to 0 and optimal for the value 0, since then the target classification signals no longer differ from the actual classification signals. Now a single ω _ij (preferably random) is selected and its current value is changed randomly (according to a Gaussian distribution). Then the fitness is determined again. If the fitness of the network has become cheaper, the new value is retained and this new value for ω _{ij is} again changed randomly; this continues until there is no improvement in fitness; An appropriately chosen termination criteria for which different heuristics come into question (e.g. termination after a constant number (preferably ten) of attempts that have not resulted in improvement) serves to terminate the improvement process. Then the variance of the Gaussian distribution for the random change of the selected ω _{ij is} changed and the change process for the ω _ij and the associated checking process of the associated fitness of the network are run through again as described above. If it is no longer possible to achieve a fitness improvement with a change in the variance, the next weight ω _ij (preferably randomly) is selected and subjected to the same process, then again the next weight and so on until finally all weights ω _ij have been adjusted this way. This procedure is carried out for a number of representative representatives of the various classes of acoustic signals to be recognized and, as a result, a network is finally obtained which is able to generate classification signals for a given acoustic signal, which provide information about the type of acoustic signal give.

As already mentioned, alternative to the above explanations are also unsupervised, possibly in-company training methods, such as those used in the context of self-organizing maps (SOM ≈ self-organizing maps) in the implementation of the present invention. When training the self-organizing cards, an input vector X = (x ₁ ,..., X _n ) is preferably compared in parallel with all weight vectors W _j = (w _1j , _... , W _nj ). The Euclidean norm, or particularly preferably the dot product, is preferably used as a metric for comparison

Use, the latter being particularly suitable for normalized vectors. The neuron, whose weights (its weight vector), in the context of the selected metric, comes closest to or is most similar to the input vector X, is now selected in order to make its weight vector even more similar to the input vector. This is preferably done by adding or subtracting a variable fraction of the difference vector X - W _j to the weight vector W _j .

To further improve the classification ability of acoustic signals, the neuro nale network also have input neurons, each for receiving the after Time (at least once) derived amplitude (or amplitudes) of a certain Fre quenzband from the frequency spectrum of the acoustic signal or its absolute amount totals (or their absolute amounts) are provided. This takes into account the observation that acoustic signals are not only recognized by their static frequency pattern can. Rather, it was found that the chronological course of this pattern again to have specifics that lead to a classification and thus to an identification lead of the acoustic signal or can at least contribute. Thus in this embodiment the acoustic signal not only according to its frequency spectrum,  but also u. U. also classified according to its change in time. Appropriate Patterns can also be observed for the higher derivatives over time.

In a preferred embodiment of the device for noise detection in acousti signals according to the present invention is also a prerecognizer device provided, which in turn provides means, each for receiving the amplitude a certain frequency band from the frequency spectrum of the acoustic signal NEN, and means, each for receiving the time-derived amplitude of a certain frequency band from the frequency spectrum of the acoustic signal or their Absolute amount serve, the prerecognition device at least one prerecognition generated signal, which the acoustic signal according to the categories sound; Whistle and No-sound classified, and so gives further information about the type of acoustic signal and the neural network is activated by at least one of the prerecognition signals, if the acoustic signal from the pre-detection device as such of the category No sound was classified.

A pre-detection device offers the advantage of pre-detection of sounds and Whistle tones from the acoustic signal, which is also for further noise detection is far more important than that of better identification of so-called sound-containing noises serves. Noises are always sound i. S. of DIN 45 681 (determination of the Tonhal activity of noises and determination of a sound surcharge for the assessment of noises chimission), if they contain a sound, the level of which corresponds to the level of the noise components the frequency group around the sound frequency without the sound by less than 6 dB below Such noise emissions are i. d. Usually particularly uncomfortable for humans and therefore u. U. the application of their level by a surcharge, the so-called called the tone surcharge, "which is a correction quantity to the equivalent continuous sound level, which at the formation of assessment levels to take into account the particular annoyance of clayey Noise is applied (see also VDI 2058 sheet 1) "(see above: DIN 45 681, p. 3, point 2.2).

A prerecognition signal can then always come from the prerecognition device be witnessed, which classifies the acoustic signal according to the category of sound, if  at least one amplitude of a certain frequency band from the frequency spectrum the acoustic signal is above a preselected tone threshold.

A prerecognition signal can also always be sent by the prerecognition device be witnessed, which classifies the acoustic signal according to the whistle category, if at least one amplitude of a specific frequency band derived over time the frequency spectrum of the acoustic signal or its absolute amount above one preselected whistle threshold.

A particularly preferred embodiment of the device for noise detection in acoustic signals according to the present invention, is characterized in that a pre-recognition signal is always generated by the pre-recognition device, which classifies the acoustic signal according to the sound category if at least one Amplitude of a specific frequency band from the frequency spectrum of the acoustic Signal is above a preselected tone threshold and then always a precedent Identification signal is generated, which is the acoustic signal according to the category whistle classified if at least one amplitude derived from time of a certain one Frequency band from the frequency spectrum of the acoustic signal or its absolute is above a preselected whistle threshold and always then Pre-detection signal is generated, which the acoustic signal according to the category no Sound classified if the acoustic signal is neither classified according to the category of sound nor according to the Category whistling is classified.

A fuzzy logic can also be provided as a pre-detection device, which in turn Provides means, each for receiving the amplitude of a specific frequency band serve from the frequency spectrum of the acoustic signal, and means, each for Reception of the amplitude of a specific frequency band derived according to time serve the frequency spectrum of the acoustic signal or its absolute amount, wherein the amplitudes received or derived amplitudes or their absolute amounts each fuzzyfected, d. H. according to a membership function on linguistic Va riable, and the fuzzy logic using an inference machine the so fuz  amplified and / or time-derived amplitudes or their abs amount by means of rules stored in a knowledge base for at least one preliminary work inferencesignal, which the acoustic signal according to the categories tone, whistle and No-sound classified, and so provides information about the type of acoustic signal.

Means are preferably provided in the fuzzy logic for receiving the maxima len amplitude and those for receiving the second largest amplitude each from one A large number of amplitudes belonging to certain frequency bands from the frequency Spectrum of the acoustic signal serve, as well as means for receiving the maximum after the time derived amplitude or its absolute amount from a variety of be agreed frequency bands belonging to derived amplitudes or their absolute amounts from the frequency spectrum of the acoustic signal, the amplitudes thus received or derived amplitudes or their absolute amounts are fuzzyfied, d. H. correspond are mapped to a linguistic variable according to a membership function.

The above embodiments according to the invention of a device for noise detection in acoustic signals, which use fuzzy logic as a prerecognition device, have proven particularly useful in experiments. Due to the easily changeable assignment functions of the input values to the linguistic variables used and the also easily changeable and expandable rule base for the fuzzy logic inference machine, they enabled the prerogative recognition device to be adapted particularly flexibly to the respective application. For a better understanding, reference is made to the exemplary embodiment discussed in relation to FIG. 2.

In a further embodiment of a device according to the invention for noise a differentiation device is provided in acoustic signals in turn provides means, each for receiving the amplitude of a certain Fre quenzband serve from the frequency spectrum of the acoustic signal, and means that derive or derive this amplitude or amplitudes at least simply over time and further output means which are the amplitude or derivative thus derived according to time th amplitudes of a certain frequency band from the frequency spectrum of the acousti  output signal. Such a differentiation device serves in some cases management forms of the present invention of supplying the central evaluation devices such as prerecognizer and artificial neural network the time-derived values of the frequency spectrum of the battery to be classified signal.

Preferably, the differentiation device furthermore has means that are based on the following the time-derived amplitude or the time-derived amplitudes each Determine current absolute amount (without negative sign) and output this or these.

The differentiating device can also operate in clocked mode and the amplitude or amplitudes of a certain frequency band from the frequency spectrum of the acousti received signal intermittently, the means for deriving the amplitude or Amplitudes after the time have a buffer, each of the amplitude or Contains amplitudes of the previous measure and thus the derivation according to time by subtracting the current from the previous amplitude or amplitudes is formed in cycles. Such a Dif that works intermittently by means of simple subtraction Referencing device represents a particularly easy to implement according to the invention great variant.

Furthermore, the inventive device for noise detection in acoustic Signals have a renormalization device, which provides means, each for Receiving the amplitude of a certain frequency band from the frequency spectrum of the Acoustic signal serve and means which renormie this amplitude or amplitudes ren that the sum of all amplitudes after renormalization is 1 and the ampli in pairs are in the same relationship to each other before the renormalization as after the renormalization, whereby further output means are provided which the so renormalized amplitude or amplified amplitudes of a certain frequency band output from the frequency spectrum of the acoustic signal. It was found that it is sufficient in certain cases, even advantageous, if not the level of the individual frequencies of the frequency band under consideration as amplitudes  but, as amplitudes, only their intensities as an expression of their relative ver relationships to each other can be used. This is done through the use of a renaissance achieved device that maintains the ratios of the amplitudes to each other Total of all, however, renormalized to 1.

The device for noise detection in acoustic signals according to the front lying invention further on a background noise elimination device point, which provides means, each for receiving the amplitude of a certain Fre quenzband serve from the frequency spectrum of the acoustic signal, and means that each for receiving the background noise amplitude of a certain frequency band from the frequency spectrum of the respective Hin belonging to the acoustic signal serve background noise, as well as means that depend on the respective amplitude of a th frequency band from the frequency spectrum of the acoustic signal, the amplitude of the corresponding frequency band from the frequency spectrum of the acoustic signal deduct the relevant background noise and continue output means that remove the amplitude or the background noise thus freed Amplitudes of a certain frequency band from the frequency spectrum of the acoustic Output signals.

The background noise elimination device can operate in cyclical mode, which receives the respective mutually corresponding amplitudes in cycles and subtracts from each other in cycles.

Preferably, the background noise eliminator receives the one to the rear signals associated with background noise from a device for background noise determination. This improves the quality of the work of the invention direction for noise detection in acoustic signals once again considerably, since it does almost every realistic acoustic environment has its own background noise to then hide it from the background noise eliminator leave what the device for background noise detection to solve the Pro  Consider the automatic noise detection in acoustic signals to a considerable extent contributes.

The background noise elimination device is based on a good determination of the Background noise instructed to make this uncomfortable for noise detection Remove the disturbance from the acoustic signal as effectively as possible. So it's up give the device used for this purpose for background noise determination as possible to provide powerful background noise detection.

This is done by a device for background noise detection in acoustic Si gnalen reached, each the amplitude of a certain frequency band from the Fre Frequency spectrum of the background noise belonging to the acoustic signal determined, it has means, each for receiving the currently measured NEN amplitude of a certain frequency band from the frequency spectrum of the acousti signal and serve to determine the background noise from one to Be drive start highly selected fictitious start background noise amplitude such ge happens that the respective background noise amplitude after each of time Time measurement of the currently prevailing amplitude of the associated frequency bandes is corrected to the last measured value whenever this measured value is smaller than the value of the background noise Amplitude, preferably plus a certain constant premium, was and output means are also provided, which the determined background noise Amplitudes of a certain frequency band from the frequency spectrum of the acoustic Output signals.

In this case, the device preferably increases before each measurement taking place from time to time of the currently prevailing amplitude, the respective one determined up to this point in time Background noise amplitude around a certain fixed surcharge.

In a preferred embodiment, background noise detection works direction in clock mode and gives the respective background noise amplitude or respective background noise amplitudes clockwise.  

All of the above embodiments of a background noise detection device the background noise of the respective acoustic can quickly and reliably Determine signal. The device starts in the event that it actually has levels and the determination of the background noise not only receives intensities starting from assuming a fairly high fictitious starting value of preferably 200 dB, the then quickly to the actually measured minimum level of the received acousti signal drops. When determining the measured up to a certain point in time It is a minimal background noise. The mini measured so far may be helpful paint level (or more generally: amplitude) only by a new minimum level (or more generally: minimum amplitude) as the new background noise level (or general mine: background noise amplitude) to replace if the new minimum level (or more general: minimum amplitude) the background noise level valid until then (or more generally: background noise amplitude) at least by a certain amount (about one Constant), preferably by 0.5 dB, falls short-term smaller undershoot not to be considered.

Furthermore, it makes sense to design the device so that the currently determined one Background noise amplitude from device to device from time to time Dimension is increased so that the device is able to move itself again registering background noise. The surcharge from time to time ensures that the determined background noise level gradually from egg Once you have reached the valley, as it were, as long as it is in the background has become.

The background noise detection device preferably operates in parallel to the the components of the noise detection device. This means that the noise Detection device only after a certain start-up phase from the background Noise detection device supplied with correct data about the background noise becomes. Experience has shown that this takes about 5 minutes. After that, it can continue in parallel work and the sound detection meanwhile always with updated data about supply the pending background noise.  

In a further embodiment of the device for noise In acoustic signals, a spectrum analysis device is provided, which again means for receiving a single amplitude over time provides acoustic signal and has other means to this only over time running amplitude of an acoustic signal in Am over time plitudes of certain frequency bands from the frequency spectrum of the acoustic signal to disassemble and spend. For this purpose, a Fast Fourier Transformation device (FFT device) used that be such a decomposition enables particularly effective. The disassembly device is used to disassemble the aku signal into a certain number of frequency bands, which, as already mentioned was mentioned, their number and frequency range according to the respective conditions of the to be identified to be identified acoustic signal.

A filter, preferably a bandpass filter, can also be used in the invention moderate noise detection device are used, the means for receiving a only provides an amplitude of an acoustic signal over time and this only outputs amplitude over time only to the extent that it is the acoustic signal essentially only within a certain frequency band preferably in a range from about 16 Hz to 20 KHz, particularly preferably in one Corresponds to a range of approximately 16 Hz to 16 kHz. This makes it possible, for example Limit the evaluation of the acoustic signal to the audible range.

A filter can also be used, which also prefers a decibel rating as an A-weighting, the received amplitude and the weighted Outputs amplitude again. Such filters are commercially available as so-called A filters and with their A-rating they support differently sensitive human hearing feel bill at different frequencies.

Another preferred embodiment of a device for noise detection in Acoustic signals are characterized in that they continue to reclassify tion device, which means for receiving the classification signal  and / or the prerecognition signal, the reclassification device generates and outputs at least one noise class signal, which is the acoustic one Assigns signal to one or more noise categories. Such a reclassification device is always advantageous if there are several classified signals that in turn should be grouped together again, for example according to their origin, but without these groups being classified according to acoustic criteria from the outset left.

For example, in the context of noise assessments, groups of acoustic signals based on the relevant statutory regulations are of concern as follows:

• - Road and rail noise due to the regulations in the 16th regulation to the Federal Immission Control Act
• - Aircraft noise due to the Aircraft Noise Protection Act
• - Commercial noise due to TA noise
• - Sports noise due to the regulations in the 18th Ordinance on Federal Mission Protection Act
• - Leisure noise due to state regulations, (see e.g. the sample Administrative Ordinance for TA Noise)
• - Noises from nature, the extraneous noises i. See the above present regulations

Here, different noises classified individually must each match the above Categories are assigned that are technically based solely on their acoustic structure cannot be classified as such. For example, that of Automobi sound emissions from motorbikes, motorcycles or rail vehicles Road and rail noises, but they each have their own acoustic pattern on which each can be recognized individually. In their entirety, these show Categories however i. d. Usually, there are no acoustic similarities to be considered  Road and rail noise can be identified directly. However, this is by means of possible of the reclassification device according to the invention, which allows it to be classified individually to map specific signals to new classes, in particular also to upper groups and so on to get the desired classification result.

Another particularly preferred embodiment of a device for noise Identification in acoustic signals according to the present invention is thereby characterized records that the device further comprises a marking device that the only time-running, possibly frequency-filtered and / or A-weighted Amplitude corresponding to the temporally associated noise class signal and / or Pre-detection signal, preferably by means of a color associated with the respective signal marked in a graphical representation of the time course of the amplitude, and Means for outputting the signal marked in this way are provided. This allows the signal run according to the type of acoustic signal z. B. in a graph on the Time can be marked in color. The color of the curve can then be seen the origin of the signal at the marked (colored) points, what the viewer is extremely vivid and informative.

Preferably, the device according to the invention for noise detection in acoustically signals also a pulse detection device, which means for reception the only frequency-filtered and / or A- provides evaluated amplitude and monitors this signal for a pulse, whereby the pulse detection device generates at least one pulse detection signal as long as the only time-filtered and possibly frequency-filtered and / or A- evaluated amplitude has a pulse, preferably exceeding a Limit by the time derivative or its absolute amount of the only over the Time-running, possibly frequency-filtered and / or A-weighted amplitude Pulse detection is used. This can be done in the case of using a marker device the only frequency-filtered and / or time-passed, possibly A-weighted amplitude corresponding to the associated pulse detection signal with priority over the associated noise class signal or previous  identification signal, preferably by means of a Far associated with the pulse detection signal be marked in a graphical representation of the time course of the amplitude the. The pulse detection serves to identify pulse-like courses of the acoustic Signals that are separated from other acoustic signals and also as an impulse signals should be recognized.

The device according to the invention for noise detection in acoustic Si gnalen have a signal analysis device, which in turn means for receiving the marked only those that are frequency-filtered and / or that run over time A-weighted amplitude, the signal analysis device with the Ge noise class detection signal or pulse detection signal marked amplitude analyzed for patterns to be recognized. The signal analysis device can in particular re as fuzzy logic, as a neural network or as a combination thereof. Such an additional analysis device also allows more complex acoustic processes identify signals over time, such as a periodic one Noise.

In general, in addition to the devices according to the invention mentioned above, that they are of course particularly good on a suitably equipped Have the computer implemented using suitable measurement and control software. In this case, the individual devices are then program-controlled Machine implemented as a technical process, controlled by the respective program.

The device for noise detection according to the invention in any case offers a lot of advantages in its application in the field of technical acoustics. It can be used wherever it is not only the extent of a sound immission that is important, but also the determination of its causes. In particular, however, it also enables the extent of a sound immission to be assigned to the associated source. In this context, a particularly important area of application of the present invention is to receive special attention below:
In the recent past, the operators of wind turbines have not only seen the wind they want in their systems, but also the increasing rejection of the residents of such systems, as they feel annoyed by the noise emissions from these systems. Correspondingly, the administrative regulations for the Federal Immission Control Act (cf. § 48 BImSchG) in the sixth general administrative regulation (Technical Instructions for Protection against Noise - TA Lärm, GMBI 1998, 503ff.) Have been drafted in such a way that the interests of the residents of such systems, which are worth protecting, are also summarized Is taken into account. According to point 6.8 of the aforementioned TA Lärm, the assessment level, which significantly influences the operating time of such a system, is formed according to its appendix: Determination of the noise immissions for the TA Lärm (cf.GMBI 1998, 510ff.) Determination by forecast is permitted. According to the state of the art, this procedure (prognosis) is the normal case today, since the technical requirements are usually missing for a measuring procedure that rewards TA Noise according to Section 6.9 with a flat-rate discount for the assessment level of 3 dB (A).

For this it would be necessary that the sound sources of the wind turbine, ie in particular the noise emitted by the wind turbines from the background and interference noise separate. The main noise is the noise of the to call the measuring microphone impinging wind, which is a very considerable extent can accept. The wind causes a deeper freer at lower speeds Quent often sound-like noise and in the case of high wind speeds a high frequency Quent annoying whistle. If you want a measuring method instead of a forecasting method to monitor the noise emissions of a wind turbine and thus in the You have to enjoy the flat-rate 3 dB (A) discount from the assessment level make sure not to measure the wind noise, but the noise emissions of the Wind turbine. Thus, the wind noise must be masked out for the measurement the.  

According to the state of the art, there are two options:
The first is, similar to how it is common in radio and television broadcasts in the outdoor area, to put a so-called windball made of fine-pored foam over the microphone in order to dampen the noise of the wind hitting the microphone. This method is also effective up to a certain wind speed (about 3 meters per second), but unfortunately fails at higher speeds, where the noise is particularly unpleasant.

As an alternative or in addition, the wind can be operated using a noise pause method are faded away, in which the acoustic signal from the microphone is only for the measurement is taken into account if the wind speed also measured un is below a certain value. However, this procedure is only for gusty wind suitable of the maximum measuring speed at which the device for measurement is still provides usable measurement results at least from time to time. At constant tem wind of higher speed, i.e. from 3 meters per second, where the windball this procedure also fails.

Since now for the monitoring of a wind turbine, which is just a lot and if possible constant wind from a certain speed to their economical operation assuming that the above measurement methods have proven unsuitable, it is necessary here a new solution, namely a measuring device that can be used even at higher constant Wind speeds is capable of that from the wind turbine Measure the sound level with the wind impingement noises suppressed.

This is achieved by a device for measuring the sound level of wind power were, which is characterized in that they have a device for noise detection in has acoustic signals, which originated in the environment of wind turbines evaluating acoustic signal and classified so that the noise detection before Direction about the type of acousti coming from the environment of wind turbines rule signal at least insofar as that the acoustic signal at least according to the categories wind turbine sound on the one hand and no wind turbine sound  on the other hand is classified, the acoustic signal only to the extent that it measures the Sound level is taken into account as it is in the device for noise detection acoustic signals is classified as wind turbine sound.

The device according to the invention for measuring the sound level of is preferably Wind turbines characterized in that they have a device for noise has detection in acoustic signals according to the present invention, wel evaluates an acoustic signal originating from the surroundings of wind turbines and whose output neurons generate at least one classification signal, which or which about the type of acoustic originating from the environment of wind turbines Signal gives or give information at least to the extent that the acoustic signal at least least according to the categories wind turbine sound and no wind turbine Sound, on the other hand, is classified, with the acoustic signal only to this extent for measurement of the sound level is taken into account as it is in the device for noise detection acoustic signals is classified as wind turbine sound.

In a further preferred embodiment according to the present invention, classifi adorns the device for measuring the sound level of wind turbines from Wind-related noises also in their closer nature as wind-sound low frequency and high-frequency wind sound.

When using a pre-detection device in the noise detector according to the invention nungseinrichtung this can be recorded in the vicinity of the wind turbine acoustic signal also preferably according to the categories of sound, preferably wind Classify tone, whistle, preferably wind whistle, and no-sound.

It is of course also possible to decide whether the acoustic signal is to be measured should be taken into account, the noise class signal of a possibly used Um to use classification device.  

The signal then actually measured and taken into account is preferably determined using a Averaging device averaged, preferably averaged over the long term, this averaging is particularly preferably carried out smoothly.

The mean value formed in this way can then be monitored by a monitoring device exceeding a limit value can be monitored.

With the device described above for measuring the sound level of wind power plants can therefore take a TA noise compliant monitoring measurement is not affected by wind noise.

This helps without the wishes of a progno that may be inaccurate in individual cases calculation to be able to operate a wind turbine economically.

Advantageously, the assessment level for such an installation can also be around a pau shell reduction of 3 dB (A) can be reduced, which further reduces their downtimes who helps and thus also contributes to their profitability.

Furthermore, the present invention enables by providing its devices gene but also other inventive methods that improve the operation of Serve wind turbines. For example, it is desirable to have the local sound level the annoyance, usually to be recorded in built-up areas, so that an exact stop say about the harassment that actually occurs there.

This could be achieved by measuring devices at all of these locations of the present invention and hereby measures the level occurring there. This however fails i. d. R. the high costs for the large number of then necessary measuring stations.

Taking into account the state of the art, such as VDI 2714 (sound propagation in Freien) and also DE 197 40 783 A1, EP 0 831 310 A2 (each method and device it is to determine additional meteorological attenuation of sound propagation) however possible using the measuring device according to the invention presented here  measurements at only one location in the area and yet for all other locations to determine the sound level exactly.

This is done according to the invention by methods for determining sound sources a wind turbine emitting sound level at a freely selectable first location (Be location) by measuring the sound level at a second location (measuring location) with a Device for measuring the sound level of wind turbines according to the present Invention, the distance between the respective sound source and the measurement site, the Distance between the respective sound source and the destination, the amount of each due sound source, the height of the measuring point, the height of the destination, the wind speed, the wind direction, the humidity, the temperature, the angle between the wind direction and the direct connection distance from the respective sound source to the measuring point, the angle between the wind direction and the direct connection route from the respective sound source to the destination, the stability factor and the meteo rological temperature gradient can be determined and from this the sound level at the loading location is determined.

With regard to this method, the values of the stability factor and the meteorological temperature gradient (also called temperature factor), which occur in addition to the generally known variables, require explanation:
The stability factor, also designated by the letter m, is a constant that depends on atmospheric stability. According to TA Luft, it takes on the following values:

Table 1 Stability classes according to TA Luft and the stability factor m

The frequency with which the wind direction and wind speed in the 6 stability classes appear, stand for a grid of 30 × 30 km for Germany at the German Wet Service available. Similar models are found all over Europe, so in Frank rich with the Pasquill classes or in Great Britain with the so-called Turner classes to disposal. The method according to the present invention is thus in its own right application is not limited to Germany, but can be used wherever a system is similar Lich the TA Luft is created. This is already the case for Europe. Stability classes I and II describe stable weather conditions with ground inversions that Classes IV and V unstable weather conditions with warmed floors, i.e. H. in sunlight. Classes III.1 and III.2 describe the transition between stable and unstable weather were.

After a publication by G. Manier, "Dispersion classes and temperature gradients ten ", Meteorologische Rundschau 28, 6-11 (March 75) on TA Luft can also be based on the stability class of the temperature gradient (denoted by dtm) specified in is usually given as a temperature change between 0 and 100 m and that of fol the following table:

Table 2 Stability classes according to TA Luft and the temperature gradient dtm

Regarding the further details that a determination of the sound level in a be agreed distance from a sound source especially considering the me On the one hand, those skilled in the art are able to make the geological conditions possible on the VDI 2714 and on the other hand to DE 197 40 783 A1 and EP 0 831 310 A2 (each method and devices for determining additional meteorological attenuations of sound levels spread).  

The devices according to the invention for measuring the sound level of wind turbines First of all, the sound level of the sound source in question enables without Determine noise and background noise at the measurement location. Under consideration of predefined geometric and meteorological conditions can then be used for example by means of the aforementioned methods according to VDI 2714 and / or DE 197 40 783 A1 and EP 0 831 310 A2 to those originating from the sound source Sound levels are closed. It is then possible (again, for example, with means of the aforementioned methods according to VDI 2714 and / or DE 197 40 783 A1 or EP 0 831 310 A2) to determine the sound level at the destination without it needs its own measurement.

The present invention also enables a method for predicting the occurrence of a Wind turbine originating averaged, preferably long-term averaged sound gels at a freely selectable forecast location during a forecast period, whereby this based on a measurement of the sound level at a second location (measurement location) with a Device for measuring the sound level of wind turbines according to the present Invention and a forecast model with assumptions about the development of Windge Speed, wind direction, air humidity, temperature, stability factor and meteorological temperature gradients occurs during the forecast period by using the The progression over time of the progression of the expected ge at the forecast location averaged, preferably long-term averaged sound level according to the above benen method is determined, the local quantities to be determined for Windge Speed, wind direction, air humidity, temperature, stability factor and meteorological temperature gradient is not determined, but instead the forecast model for are taken from the forecast period.

Such a forecasting method in turn enables a method to optimize the Operation of a wind turbine, whereby for one or more freely selectable locations (Pro gnose locations) one or more assumptions about the operating periods of the individual sound sources len of the wind turbine (wind turbines) are made during the forecast period and based on this the expected average, preferably long-term average  Development of the sound level at the forecast locations according to the above predicted procedures, and then finally the operating periods of the individual sound sources of the wind turbine (wind turbines) for actual operation be determined during the forecast period, which with the greatest possible energy production by the wind turbine below a level that cannot be exceeded Limit for the averaged, preferably long-term averaged sound level at the respective forecast locations.

This allows the operation of a wind turbine from an economic point of view can be further optimized by the downtimes of the plant or individual wind turbines This can be optimized to a minimum without compromising the interests of the To affect residents of such facilities.

The present invention also enables noise monitoring in certain areas, in which noise sources are located (e.g. commercial noise areas) in such a way that the noise in the Immission area in question (e.g. in the area of adjoining residential buildings) is identified and the noises measured in this way on the basis of those occurring in parallel Emission measurements in the area where the noise is generated, the corresponding Is assigned to causers. The methods and devices of the present The invention allows a sufficiently reliable separation of the Ge to be identified noises from other noises, such as those from public road traffic.

For this purpose, a central evaluation unit is provided which, preferably online, with the Emission and immission (outside) measuring points is connected.

If noises are now identified, the level of which can be expected, based on the prevailing propagation conditions, that the immission guide value may be exceeded - or has already occurred - appropriate alarm messages are automatically sent to the person causing the noise, preferably the following ones contain defined information or parts thereof:

• - Most likely source of noise  
• - Position of the source
• - level of sound immission
• - Date and Time
• - an alarm text

The points to be alerted are preferably operated by the respective operators of the Noise sources specified. You should be available at all times. After receiving one Countermeasures can then be taken, such as organizational or technical measures be taken to achieve noise reduction.

The alarm messages preferably appear in a control center, for example at a shift office ter. However, they can also be sent by fax or e-mail or directly to act on controls or regulations, such as for switching light or Signal signs are used that allow noise reduction.

Such noise monitoring (or noise-reducing interference) is preferred way to monitor larger commercial areas with a variety of commercial Noise sources (noise sources), here the monitoring of port facilities with the Handling facilities as noise sources is a particularly preferred area of application.

In the following, exemplary embodiments to be understood in a non-restrictive manner are based on the Drawing discussed. In this shows:

Fig. 1 shows an inventive apparatus for noise identification in the block diagram,

Fig. 2 is a Vorerkennungseinrichtung as fuzzy logic executed also in the block diagram,

Fig. 3 is a schematic overview of a wind turbine with surrounding residential buildings, and

Fig. 4 shows the structure of a noise monitoring system according to the invention in a schematic overview.

Fig. 1 shows an inventive device for noise detection 1 in block diagram. In this illustration, various sources of noise, such as a car 2 , a locomotive 3 and a twittering bird 4 can be seen.

The acoustic signal picked up by all of these noise sources by means of a microphone 5 is first fed in the device 1 to a so-called A-filter 6 which, on the one hand, only allows the audible range of 16 Hz-16 KHz to pass from the frequency spectrum of the signal and thereby subjecting the signal to a so-called A-weighting, which it evaluates according to the different subjective human perceptions for different frequencies. The signal evaluated in this way then leaves the A filter 6 again.

The signal is now fed from the A filter 6 to a spectrum analysis device 7 , which uses a Fast Fourier Transformation to split the signal into a number n different levels (P) 8 for the n frequency bands.

These n different levels (P) 8 are now supplied on the one hand to a background noise elimination device (HEL) 9 and on the other hand to a background noise detection device (HEM) 10 which determines the background noise level (H) 11 for the levels (P) 8 , which are also fed to the background noise elimination device (HEL) 9 . The background noise elimination device (HEL) 9 now subtracts from the levels (P) 8, the background noise level (H) 9 and outputs the result as a corrected level (K) 12 again.

The corrected levels (K) 12 are fed to the renormalization device (REN) 13 , which renormalizes them and outputs them again as renormalized intensities (R) 14 .

The renormalized intensities (R) 14 are in turn fed to a differentiating device (DIFF) 15 , which they derive according to the time and then form the absolute amount from the result. The result of differentiated, standardized intensities (dR) is then output again and forms the signal vector 16 together with the standardized intensities (R).

The signal vector 16 then serves both the pre-detection device (VOR) 17 and the neural network (NNN) for signal classification 18 as an input variable.

The pre-detection device (VOR) 17 now determines from this whether it is a whistle tone, a tone or no tone and accordingly outputs the pre-recognition signals tone 19 , whistle tone 20 or no tone 21 .

The neural network (NNN) for signal classification 18 classifies the acoustic signal with the aid of the signal vector 16 and generates various classification signals 22 that classify the acoustic signal according to its frequency spectrum and / or its change over time.

The classification signals 22 are supplied to a reclassification device (UMK) 23 , which combines them into three groups of noise classes and, depending on the association of a classification signal with these groups, generates one of three noise class signals 24 .

The noise class signals 24 , as well as the prerecognition signals 19 , 20 are fed to a marking device (MAR) 25 , which thus marks the level received by the A filter 6 in accordance with these signals 19 , 20 , 24 . This marker can then, for. B. as a different colored representation of the A-weighted level that has left the A filter 6 .

A pulse detection device (IDET) 26 is also arranged in front of the marking device (MAR) 25 , which also receives the level output by the A filter 6 and tests for the presence of a pulse. If such a pulse is present, it outputs a pulse detection signal 27 for this time, which is also fed to the marking device (MAR) 25 . The marking device (MAR) 25 thus marks the level received by the A filter 6 primarily over the markings caused by the noise class signals 24 or prerecognition signals 19 , 20 .

The marked level signal 28 is finally output by the marking device (MAR) 25 and can then be displayed, for example, on a display 29 in accordance with the amplitude of its respective level and its marking.

Fig. 2 shows a Vorerkennungseinrichtung 17 as fuzzy logic executed also in the block diagram. Which abge of amplitudes 31 and of absolute values with respect to time initiated amplitudes 32 composing signal vector 16 is analyzed in its amplitude portion 31 by a first maximum-determining device 34, which Determined lunsgvorrichtung 34, both the maximum amplitude and the second largest amplitude of the n amplitudes ( A ₁ ... A _n ) 31 each generated as an output signal. A second maximum determination device 33 analyzes the absolute amounts of the n amplitudes derived from time (dA ₁ ... DA _n ) 32 and outputs their maximum.

The maxima determined in this way are now fed to the pre-detection device 17 , which is designed here as fuzzy logic, the maximum amplitude from the plurality of amplitudes 32 being fed to a first fuzzification device (fuzzy max _A1 ) 30 as a receiving means of the device 17 . The second largest amplitude from the plurality of amplitudes 31 is fed to a second fuzzification device (fuzzy Ma x _A2 ) 35 as a receiving means. Finally, the absolute value of the maximum time-derived amplitude from the plurality of time-derived n amplitudes (dA ₁ ... DA _n ) 32 is fed to a third fuzzification device (fuzzy max _dA ) 36 as receiving means.

The fuzzy devices depict the received amplitudes or the absolute absolute amount of an amplitude derived over time by means of membership functions in accordance with the degree of their membership in a specific fuzzy set on linguistic variables 37 . The choice of the course of the corresponding auditory functions according to the respective task is to be determined experimentally.

These linguistic variables 37 , which now have membership grade values, are processed (inferred) by the fuzzy logic with its inference machine in accordance with the rules stored in a rule base 38 . As a result of this Inferierens He will then receive results of which are defuzzified in a Defuzzyfizietungsmodul 38 a and then leave this as Vorerkennungssignale 19, 20, 21, the processing the Vorerkennungseinrich finally outputs.

Fig. 3 is a schematic overview illustration showing a wind power plant with surrounding residential buildings. 39 The wind turbine consists of wind turbines 40 , which represent the sound sources of the wind turbine, and from a measuring station at a measuring point 41 , shown here by means of a microphone, with which the measuring point 41 is used. The measurement signal is fed to a device according to the invention for measuring the sound level of wind turbines 42 , which now makes it possible to first determine the sound level of the sound source 40 in question without interference and background noise at the measurement location 41 . Taking into account the previously determined geometric and meteorological conditions, it is then possible, for example, to infer the sound level emanating from the sound sources 40 using the method according to VDI 2714 and / or DE 197 40 783 A1 or EP 0 831 310 A2. This then makes it possible (again, for example, by means of the aforementioned method according to VDI 2714 and / or DE 197 40 783 A1 or EP 0 831 310 A2) because noise levels at the residential areas are determined as destination 39 without it being one own measurement required.

Fig. 4 shows the construction of a noise monitoring system of the invention as such is suitable for monitoring of port facilities, in schematic overview. It is based on the networking of so-called emission and immission measurement units.

Claims (42)

1. Device for noise detection in acoustic signals ( 1 ) with at least one neural network ( 18 )

• Input neurons, output neurons and weighted connections between the neurons,
• wherein the input neurons are each provided for receiving the amplitude ( 31 ) of a specific frequency band from the frequency spectrum of the acoustic signal, and
• the output neurons generate at least one classification signal ( 22 ) which classifies the acoustic signal according to its frequency spectrum,

wherein the classification signal or the classification signals give or give information about the type of acoustic signal. 2. Device for noise detection in acoustic signals according to claim 1, characterized in that the neural network ( 18 ) also has input neurons, each for receiving the time-derived amplitude ( 32 ) of a specific frequency band from the frequency spectrum of the acoustic signal or its respective absolute amount are provided. 3. Device for noise detection in acoustic signals according to claim 2, characterized in that it further comprises a pre-detection device ( 17 ),
which in turn provides means which are used to receive the amplitude ( 31 ) of a specific frequency band from the frequency spectrum of the acoustic signal,
and means which each serve to receive the amplitude ( 32 ) of a specific frequency band derived from the time from the frequency spectrum of the acoustic signal or its absolute amount,
wherein the prerecognition device generates at least one prerecognition signal ( 19 , 20 , 21 ) which classifies the acoustic signal according to the categories tone ( 19 ), whistle tone ( 20 ) and no-tone ( 21 ), and thus further information about the type of acoustic Si gnals there
and the neural network is activated by at least one of the prerecognition signals if the acoustic signal has been classified by the prerecognition device as one of the no-sound category ( 21 ). 4. A device for noise detection in acoustic signals according to claim 3, characterized in that the prerecognition device always generates a prerecognition signal which classifies the acoustic signal according to the category sound ( 19 ) if at least one amplitude of a certain frequency band from the Fre quency spectrum of the acoustic signal is above a preselected tone threshold. 5. A device for noise detection in acoustic signals according to claim 3 or 4, characterized in that a pre-identification signal is always generated by the prerecognition device, which classifies the acoustic signal according to the category whistle ( 20 ) if at least one amplitude derived over time a certain frequency band from the frequency spectrum of the acoustic signal or its absolute amount is above a preselected whistle threshold. 6. A device for noise detection in acoustic signals according to claim 3, characterized in that a prerecognition signal is always generated by the prerecognition device, which classifies the acoustic signal according to the category sound ( 19 ) if at least one amplitude of a certain frequency band from the Fre quency spectrum of the acoustic signal is above a preselected tone threshold
and whenever a pre-detection signal is generated which classifies the acoustic signal according to the whistle category ( 20 ) if at least one time-derived amplitude of a specific frequency band from the frequency spectrum of the acoustic signal or its absolute amount is above a pre-selected whistle threshold
and a prerecognition signal is generated which classifies the acoustic signal according to the no-sound category ( 21 ) when the acoustic signal is neither classified according to the sound category nor according to the whistle category. 7. A device for noise detection in acoustic signals according to claim 3, characterized in that a fuzzy logic is provided as the pre-detection device ( 17 )
in turn provides means ( 30 , 35 ) which each serve to receive the amplitude of a certain frequency band from the frequency spectrum of the acoustic signal,
and means ( 36 ), each of which serves to receive the amplitude of a particular frequency band derived from the time from the frequency spectrum of the acoustic signal or its absolute amount,
the amplitudes ( 31 ) thus received or derived amplitudes or their absolute amounts ( 32 ) each being fuzzified, that is to say mapped to a linguistic variable ( 37 ) according to a membership function, and
the fuzzy logic uses an inference machine to infer the fuzzy amplitudes ( 31 ) and / or the amplitudes ( 32 ) derived according to time or their absolute amount using rules ( 38 ) stored in a knowledge base to form at least one prerecognition signal ( 19 , 20 , 21 ), which classifies the acoustic signal according to the categories sound, whistle and no sound, and thus provides information about the type of acoustic signal. 8. A device for noise detection in acoustic signals according to claim 7, characterized in that a fuzzy logic is provided as the pre-detection device ( 17 )
in turn provides means ( 30 , 35 ) which serve to receive the maximum amplitude from a plurality of amplitudes belonging to certain frequency bands from the frequency spectrum of the acoustic signal, and
further serve to receive the second largest amplitude from a plurality of amplitudes belonging to certain frequency bands from the frequency spectrum of the acoustic signal,
and means ( 36 ) which serve to receive the maximum amplitude derived from time or its absolute value from a plurality of derived amplitudes belonging to certain frequency bands or their absolute amounts from the frequency spectrum of the acoustic signal,
wherein the amplitudes ( 31 ) received or derived amplitudes ( 32 ) or their absolute amounts are fuzzyfied, ie are mapped to a linguistic variable ( 37 ) according to a membership function, and
the fuzzy logic uses an inference machine to infer the so fuzzy amplitudes and / or the amplitudes ( 32 ) derived from the time or their absolute amount by means of rules ( 38 ) stored in a knowledge base to at least one prerecognition signal ( 19 , 20 , 21 ), which or which classifies or classifies the acoustic signal according to the categories sound, whistling and no sound, and thus provides information about the type of acoustic signal. 9. Device for noise detection in acoustic signals according to one of claims 1 to 8, characterized in that it further comprises a differentiation device ( 15 ),
which in turn provides means which are used to receive the amplitude ( 31 ) of a specific frequency band from the frequency spectrum of the acoustic signal,
and means which at least simply derive or derive this amplitude or amplitudes ( 31 ) over time and
furthermore output means which output the amplitude or amplitudes ( 32 ) of a specific frequency band derived from the frequency spectrum of the acoustic signal. 10. Device for noise detection in acoustic signals according to claim 9, characterized characterized in that the differentiating device further comprises means which the time-derived amplitude or the time-derived amplitudes determine the respective absolute amount (without negative sign) and this or this issues. 11. A device for noise detection in acoustic signals according to claims 9 or 10, characterized in that the differentiating device operates in clocked mode and receives the amplitude or amplitudes of a specific frequency band from the frequency spectrum of the acoustic signal
and that the means for deriving the amplitude or amplitudes according to the time have a buffer, each of which contains the amplitude or amplitudes of the preceding clock and thus the derivation according to the time is formed in cycles by subtracting the current from the previous amplitude or amplitudes. 12. Device for noise detection in acoustic signals according to one of claims 1 to 11, characterized in that it further comprises a renormation device ( 13 ),
provides the means which are used to receive the amplitude ( 31 ) of a specific frequency band from the frequency spectrum of the acoustic signal,
and means which renormalize this amplitude or amplitudes ( 31 ) such that the sum of all amplitudes ( 31 ) after the renormalization is equal to 1 and the amplitudes ( 31 ) are in pairs in relation to one another before the renormalization as after the renormalization and
furthermore output means which output the renormalized amplitude or renormalized amplitude of a specific frequency band from the frequency spectrum of the acoustic signal. 13. Device for noise detection in acoustic signals according to one of claims 1 to 12, characterized in that it further comprises a background noise elimination device ( 9 ),
which in turn provides means which are used to receive the amplitude ( 31 ) of a specific frequency band from the frequency spectrum of the acoustic signal,
and means which each serve to receive the background noise amplitude ( 11 ) of a certain frequency band from the frequency spectrum of the background noise belonging to the acoustic signal,
and means which subtract the background noise amplitude ( 11 ) of the corresponding frequency band from the frequency spectrum of the respective background noise belonging to the acoustic signal from the respective amplitude ( 31 ) of a specific frequency band from the frequency spectrum of the acoustic signal,
furthermore output means which output the amplitude or liberated amplitudes of a certain frequency band from the frequency spectrum of the acoustic signal, which are thus freed from background noise. 14. A device for noise detection in acoustic signals according to claim 13, characterized characterized in that the background noise eliminator in the clock mode works and receives the respective mutually corresponding amplitudes in cycles and subtracts from each other in cycles. 15. A device for noise detection in acoustic signals according to claim 13 or 14, characterized in that the device further comprises a device for determining background noise ( 10 ) according to claim 30 to 33. 16. Device for noise detection in acoustic signals according to one of claims 1 to 15, characterized in that it further comprises a spectrum analysis device ( 7 ) which
Provides means for receiving a single amplitude of an acoustic signal over time
and further means for dividing and outputting this single amplitude of an acoustic signal over time into amplitudes ( 31 ) of specific frequency over time, from the frequency spectrum of the acoustic signal. 17. A device for noise detection in acoustic signals according to claim 16, characterized in that as a means for breaking down a single amplitude over time into each of the time amplitudes ( 31 ) of certain frequency bands a Fast Fourier transforming device (FFT- Device) ( 7 ) is used. 18. A device for noise detection in acoustic signals according to one of claims 16 or 17, characterized in that it further comprises a filter ( 6 ), preferably a bandpass filter ( 6 ), the means for receiving a single amplitude over time provides an acoustic signal and outputs this single amplitude over time only to the extent that it essentially corresponds to the acoustic signal only within a certain frequency band. 19. A device for noise detection in acoustic signals according to claim 18, characterized in that the filter ( 6 ) outputs the amplitude of the received acoustic signal essentially only in a range from about 16 Hz to 20 KHz. 20. Device for noise detection in acoustic signals according to claim 18 or 19, characterized in that the filter ( 6 ) additionally performs a decibel evaluation, preferably an A-evaluation (A-filter) ( 6 ), of the received amplitude and the outputs the weighted amplitude again. 21. Device for noise detection in acoustic signals according to one of claims 1 to 20, characterized in that it further comprises a reclassification device ( 23 ) which has means for receiving the respective classification signals ( 22 ) and / or respective prerecognition signals, the Reclassification device ( 23 ) generates and outputs at least one noise class signal ( 24 ) which assigns the acoustic signal to one or more noise categories. 22. A device for noise detection in acoustic signals according to claim 21, insofar as this is referred back to one of claims 16 to 20, characterized in that the device further comprises a marking device ( 25 ) which is the only time-running, possibly frequency-filtered and / or A-weighted amplitude according to the temporally associated noise class signal ( 24 ) and / or prerecognition signal ( 19 , 20 ), preferably by means of a color associated with the respective signal in a graphical representation of the time course of the amplitude, marked, and Means for outputting the signal marked in this way are provided. 23. A device for noise detection in acoustic signals according to one of claims 16 to 20 or claim 22 or claim 21 insofar as this is related to one of claims 16 to 20, characterized in that it further comprises a pulse detection device ( 26 ) Which provides means for receiving the only time-running, possibly frequency-filtered and / or A-weighted amplitude and monitors this signal for a pulse, the pulse detection device generating at least one pulse detection signal ( 27 ) as long as the only one over the Time-running and possibly frequency-filtered and / or A-weighted amplitude has a pulse. 24. Device for noise detection in acoustic signals according to claim 23, characterized in that the pulse detection device ( 26 ) has means for pulse detection, the time derivative or its absolute amount of the only time-running, possibly frequency-filtered and / or A-weighted Monitor the amplitude for exceeding a limit value and, in the event of such an overshoot, generate the pulse detection signal ( 27 ), and means are provided for outputting this signal. 25. A device for noise detection in acoustic signals according to claim 23 or 24, insofar as these are related to claim 22, characterized in that the marking device ( 25 ) corresponds to the only time-running, possibly frequency-filtered and / or A-weighted amplitude the temporally associated pulse detection signal ( 27 ) with priority over the associated Ge noise class signal ( 24 ) and / or pre-detection signal ( 19 , 20 ), preferably by means of egg ner the pulse detection signal color in a graphical representation of the time course of the amplitude, marked. 26. Device for noise detection in acoustic signals according to one of claims 23 to 25 insofar as these are referred back to claim 22, characterized in that it further comprises a signal analysis device which has means for receiving the marked only time-running, possibly frequency has filtered and / or A-weighted amplitude, the signal analysis device further analyzing the amplitude ( 28 ) marked with the noise class detection signal or pulse detection signal for patterns to be recognized. 27. The device for noise detection in acoustic signals according to claim 26, characterized characterized in that the signal analysis device is implemented as a fuzzy logic. 28. Device for noise detection in acoustic signals according to claim 26, characterized characterized in that the signal analysis device is implemented as a neural network. 29. Device for noise detection in acoustic signals according to claim 26, characterized characterized in that the signal analysis device as a combination of a fuzzy Logic is realized with a neural network. 30. A device for determining background noise ( 10 ) in acoustic signals, each of which determines the background noise amplitude ( 11 ) of a specific frequency band from the frequency spectrum of the background noise associated with the acoustic signal, wherein it
Has means which are used to receive the respectively currently measured amplitude ( 31 ) of a specific frequency band from the frequency spectrum of the acoustic signal, and
the background noise is determined based on a fictitious start background noise amplitude selected at the start of operation such that the respective background noise amplitude ( 11 ) after a measurement of the currently prevailing amplitude ( 31 ) of the associated frequency band from time to time in each case is corrected to the last measured value if this measured value was smaller than the value of the background noise amplitude ( 11 ) determined up to this point in time, and
output means are also provided which output the determined background noise amplitudes ( 11 ) of a specific frequency band from the frequency spectrum of the acoustic signal. 31. A device for determining background noise ( 10 ) in acoustic signals, each of which determines the background noise amplitude ( 11 ) of a specific frequency band from the frequency spectrum of the background noise associated with the acoustic signal, wherein it
Has means which are used to receive the respectively currently measured amplitude ( 31 ) of a specific frequency band from the frequency spectrum of the acoustic signal, and
the background noise is determined based on a fictitious start background noise amplitude selected at the start of operation such that the respective background noise amplitude ( 11 ) after a measurement of the currently prevailing amplitude ( 31 ) of the associated frequency band from time to time in each case is corrected to the last measured value if this measured value was less than the value of the background noise amplitude ( 11 ) determined up to this point in relation to a certain constant surcharge, and
output means are also provided which output the determined background noise amplitudes ( 11 ) of a specific frequency band from the frequency spectrum of the acoustic signal. 32. Device for background noise determination ( 10 ) in acoustic signals according to claim 30 or 31, characterized in that the device before each measurement of the currently prevailing amplitude ( 31 ) from time to time the respective background noise amplitude determined up to this point in time ( 11 ) increased by a certain fixed surcharge. 33. Device for background noise detection ( 10 ) in acoustic signals according to one of claims 30 to 32, characterized in that the background noise detection device ( 10 ) operates in clocked mode and the respective background noise amplitude ( 11 ) or the respective background noise amplitudes ( 11 ) clockwise from there. 34. Device for measuring the sound level of wind turbines ( 42 ), characterized in that it has a device for noise detection in acoustic signals ( 1 ) according to one of claims 1 to 29, which evaluates an acoustic signal originating from the surroundings of wind turbines and whose output neurons generate at least one classification signal ( 22 ), which or which provides information about the type of acoustic signal originating from the environment of wind turbines at least to the extent that the acoustic signal at least according to the categories of wind turbine sound and On the other hand, no wind turbine sound is classified, the acoustic signal being taken into account for measuring the sound level only insofar as it is classified as wind turbine sound by the device for noise detection in acoustic signals. 35. Apparatus for measuring the sound level of wind turbines ( 42 ) according to claim 34, characterized in that the output neurons generate at least one classification signal ( 22 ), which or which information about the type of the acoustic signal originating from the environment of wind turbines, at least to this extent gives or give that the acoustic signal is classified at least according to the categories wind turbine sound on the one hand and wind sound low frequency or wind sound high frequency on the other hand, the acoustic signal being taken into account only to the extent that it measures the sound level the device for noise detection in acoustic signals is classified as wind turbine sound. 36. Device for measuring the sound level of wind turbines ( 42 ) according to claim 34, insofar as this is referred back to claims, according to which the device for measuring the noise level of wind turbines has a device for noise detection in acoustic signals according to claim 3 or a claim referring back to , characterized in that the output neurons generate at least one classification signal ( 22 ) which provides information about the type of acoustic signal originating from the environment of wind turbines, or that the acoustic signal at least according to the categories of wind turbine sound On the one hand and no wind turbine sound is classified on the other hand and the prerecognition device generates at least one prerecognition signal ( 19 , 20 , 21 ), which is the acoustic signal according to the categories of sound, preferably wind sound, whistling sound, preferably wind whistling sound, and No sound classification rt, wherein the acoustic signal is only taken into account for the measurement of the sound level insofar as it is classified by the device for noise detection in acoustic signals as wind turbine sound. 37. Device for measuring the sound level of wind turbines ( 42 ), characterized in that it has a device for noise detection in acoustic signals according to claim 21 or a claim related thereto, this device evaluating an acoustic signal originating from the environment of wind turbines and the Reclassification device generates at least one noise class signal ( 24 ) therefrom, which classifies or classifies the acoustic signal at least as wind turbine sound or no wind turbine sound, the acoustic signal being taken into account only to the extent that the sound level is measured by the Device for noise detection in acoustic signals is classified as wind turbine sound. 38. Device for measuring the sound level of wind turbines ( 42 ) according to any one of claims 34 to 37, characterized in that an averaging device for averaging, preferably long-term averaging of the noise level measured during operation of the wind power plant is provided, only those during Operation of the wind turbine measured levels are taken into account by the averaging device for forming the agent, which result from an acoustic signal which was classified by the device for noise detection in acoustic signals as wind turbine sound. 39. Device for measuring the sound level of wind turbines ( 42 ) according to claim 38, characterized in that a monitoring device is also provided, which monitors the sliding averaging of the sound level measured during operation of the wind turbine for exceeding a limit value. 40. Method for determining the sound sources ( 40 ) of a wind power plant, the sound level at a freely selectable first location (destination) ( 39 ) by means of measurement of the sound level at a second location (measuring location) ( 41 ) with a device ( 42 ) one of claims 34 to 39, wherein the distance between the respective sound source ( 40 ) and the measuring location ( 41 ), the distance between the respective sound source ( 40 ) and the destination ( 39 ), the height of the respective sound source ( 40 ), the height of the measuring location ( 41 ), the height of the destination ( 39 ), the wind speed, the wind direction, the air humidity, the temperature, the angle between the wind direction and the direct connection path from the respective sound source ( 40 ) to the measuring location ( 41 ) , the angle between the wind direction and the direct link from the respective sound source ( 40 ) to the destination ( 39 ), the stability factor and the me teorological Temperaturg be determined radially and from this the sound level at the destination ( 39 ) is determined. 41. Method for forecasting the averaged, preferably long-term averaged sound level originating from a wind power plant at a freely selectable forecast location ( 39 ) during a forecast period, this being based on a measurement of the sound level at a second location (measurement location) ( 41 ) with a device ( 42 ) according to any one of claims 34 to 39 and a forecast model with assumptions about the development of wind speed, wind direction, air humidity, temperature, stability factor and meteorological temperature gradient during the forecast period by the time course of the The forecast location ( 39 ) of the expected average, preferably long-term average sound level is determined by the method according to claim 40, the variables to be determined there for wind speed, wind direction, air humidity, temperature, stability factor and meteorological temperature gradient not ht determined, but instead are taken from the forecast model for the forecast period. 42. Method for optimizing the operation of a wind turbine, one or more assumptions being made about the operating periods of the individual sound sources ( 40 ) of the wind turbine (wind turbines) for one or more freely selectable locations (forecast locations) ( 39 ) during the forecast period and based on this the respectively expected, preferably long-term averaged development of the sound level at the forecast locations ( 39 ) are predicted according to the method according to claim 41, and then finally the operating periods of the individual sound sources ( 40 ) of the wind power plant (wind turbines) for the actual operation during the forecast period, which are below the limit for the averaged, preferably long-term averaged sound level at the respective forecast locations ( 39 ) with the greatest possible energy production by the wind turbine.