DE10139309B4 - Method of suppressing noise - Google Patents

Abstract

Method for suppressing noise with the following steps:
- obtaining an analytical signal from an input signal containing interference noise (S _in );
- Calculating an instant phase signal (IFI) from the analytical signal;
- Calculating an instant frequency signal (IFR) as a time derivative of the instant phase signal (IFI) divided by 2π;
- Forming a single delayed signal (S _del , IA _del , IFI _del , S _del1 , IA _del1 , IFI _del1 ) by delaying a signal (S, IA, IFI) from the group containing the input signal and signals derived therefrom a period of time (t _del ) that corresponds to your reciprocal of a simple integer multiple of the instant frequency signal (IFR);
Formation of a signal mean value (S _{a v} , IA _{a v} , IFI _{a v} ) from the signal (S, IA, IFI) and the simply delayed signal (S _del , IA _del , IFI _del , S _del1 , IA _del1 , IFI _del1 ) in order to obtain an output signal (S _out ) in which the noise is suppressed.

Description

The invention relates to a method for oppression of noise.

The development of hearing aids is have been perfected so far in recent years that technical Problems are almost impossible or insignificant. Still urgent However, the problem is with the amplification signals like this to process that the useful signals are transmitted with as little loss as possible and the interference signals as far as possible repressed become.

But also in other applications, such as in the area of message transmission over telephone lines or via Funk is the suppression of background noise an important topic.

A simple approach is by using high-pass filters, low-pass filters or bandpass filters attenuate certain frequency ranges in which a high proportion of interference signals is suspected. Due to the variety of possible interference signals, such methods have but only a limited use, and the useful signal, that usually a speech signal is distorted and disturbed.

Another difficulty is in that language is an extremely complex signal. Various models of speech generation are known, such as in J.L. FLANAGAN: "Speech Analysis, Synthesis and Perception "2nd ed, Springer Verlag, New York 1972. It defines a basic signal, which is either a series of There are impulses, as is the case with vowels, for example Noise, for example, in consonants such as "S" or "SCH". The pulse series defines the pitch and is often referred to as F0 (zero formant). Such a signal usually has numerous harmonic components up to very high frequencies. Breathing creates an additional Noise. The signals generated in this way continue to be articulated filtered. This changes the spectral form and the language arises. Derived from it has been tried to suppress noise to develop based on spectral analysis. That I however the language is constantly changing that is called Amplitude, frequency and spectra are not constant, they are Process limits. additional Difficulties arise, for example, from co-articulations, the one transition from one phoneme to another. In contrast to are usually disturbances relatively simpler signals, which by the way also for music true.

A basic account that is still applicable today, is in J.S. LIM, A.V. OPPENHEIM: "Enhancement and Bandwith compression of noisy speech "proceedings of IEEE Vol. 67, No. 12, December 1979. Furthermore, in recently, Time procedures like "Beam Forming "and" Blind Source Separation "meaning gained. However, such methods always require more than one microphone. The However, the present invention relates to methods that are also based on a signal obtained from a single microphone are applicable.

In practice, so-called "noise gates" are often used Basically represent one or more expanders connected in parallel. The input signal is amplified and several filters in parallel supplied and thereby into several frequency bands divided. In each channel after that. determined the amplitude by filtering the absolute value with a low pass filter to get the gain average energy or amplitude of the signal. This is followed by a nonlinear transformation, that of digital Signal processing also with a so-called "look up table", but also differently, for example through a closed function can be implemented. The so won The value is used to amplify the signal of the respective channel mitigate, this means, that in the simplest case, multiplication takes place. The on signals thus obtained from each channel are added to one Generate exit signal. An expansion of the signal can occur easily done this way by then when the energy, i.e. the amplitude of the signal, is low, the signal is reduced, whereas with a larger amplitude a reinforcement is made. Interference is therefore reduced in every frequency range Amplitude suppressed. However, such systems only work when the disturbance is relatively constant. On Another problem is that even quiet speech signals are suppressed. Furthermore. artifacts are generated during pauses in speech, sometimes are very annoying. Overall, mar can say that such systems are not satisfactory solution to suppress background noise can offer.

From the EP 542 710 A1 (RIBIC) is a method for processing signals in which an analytical signal is obtained from an input signal. An analytical signal is a complex signal whose imaginary component represents the Hilbert transform of the real component. The mathematical foundations of this are described in detail, for example, in RB RANDALL: "Frequency Analysis" BRÜL & KJAER, 1987. Various possibilities and circuits for obtaining the Hilbert signals are described in the published patent application. Due to the current possibilities of digital signal processing, it is possible to implement a Hilbert transformer in a relatively simple manner in order to obtain the real and the imaginary signal. For example, reference is made to SL HAHN: "Hilbert Transforms in Signal Processing" Artech House, 1996. Starting from the analytical signal consisting of the two Hilbert signals, or the real part and the imaginary part, a so-called instant amplitude can be determined signal can be calculated using the following formula (1): IA = (Re 2 + In 2 ) 1.2 (1) where Re denotes the real part of the analytical signal and Im the imaginary part of the analytical signal.

The instant amplitude signal represents a value that represents the current magnitude. The magnitude is the vector length for complex signals, the amplitude of the input signal in the time domain is the instantaneous value of the real part of the analytical signal. An instant phase signal is calculated in an analogous manner according to the following formula (2): IFI = anctan (Im / Re) (2) where IFI represents a value that can be regarded as the current phase of the signal.

From the EP 542 711 A1 a method is known with which audio signals can be processed in order to improve the function of hearing aids. An input signal is used to generate an analytical signal, from which an instant amplitude signal is calculated. This instant amplitude signal is used as a manipulated variable in order to appropriately amplify the input signal or one of the Hilbert signals, so that signal compression is achieved. The instant amplitude signal is therefore only used to process the input signal accordingly. However, since the delay of the instant amplitude signal and thus; controlled signal do not match, a completely satisfactory solution can not be achieved.

Also the US 4,495,643 A (ORBAN) and the US 6,205,225 B (ORBAN) show methods that first generate an analytical signal using a Hilbert transformation. In the above circuits, however, the Hilbert signals are filtered before further processing, so that a real instant amplitude signal cannot be obtained. Signal peaks can be limited with such methods, but it is not possible to effectively suppress overall noise.

Object of the present invention is to provide a method of the type described above at de m noise effectively suppressed can be. In particular, such a method should be easy Adjustability and adaptation to various environmental conditions enable, and in the case of hearing aids too the specific hearing loss of the respective person is. In particular, the harmonic components of the input signal be highlighted.

According to the invention, the steps of the claim 1 performed.

By the method according to the invention it is possible in a simple way recognize and reinforce harmonious rows. The amount of time that the Reciprocal of the instant frequency signal is the instantaneous Vibration duration. The signal that is delayed can be the input signal be yourself, or it can be and signals coming from the Input signal are derived. In the case of claim 4 are this is the instant amplitude signal and the instant phase signal.

In the simplest case, the signal in general, or the instant amplitude signal and the instant phase signal in particular, delayed by exactly this amount of time, as is the case in claims 3, 7 or 8 is recorded. In this way it is achieved that a Signal that primary consists of a fundamental tone and harmonic overtones, a very high one Correlation between the instant amplitude signal and the delayed instant amplitude signal and also has a high correlation between the instant phase signal and the delayed Has instant phase signal. By forming the mean amplitude and positive phase interference is therefore achieved. For others non-harmonic signals there is no such correlation, so that the averaging results in a weakening. This way, speech signals that are essentially harmonic Signals exist, preferably amplified.

According to claim 2 or 5, it is possible several delay elements to connect in series and thereby increase the selectivity. at the use of a single delay element becomes harmonic Signals, i.e. those for there is a high correlation, amplified by the addition by a factor of 2. For not correlated signals the factor corresponds only to the root of 2, so that the signal / speed ratio with simple delay around 3 dB improved. If according to claim 2 for example four delay elements connected in series, there is an improvement of 12 dB reached.

A summation like a moving average can be achieved according to claim 6. Depending on the choice of the constant factor k, a higher or lower selectivity reached. It applies k <1, being a higher one Value of k a larger number of delay elements the above variant equivalent.

The value of k or the number of delay elements is capped because voice signals are not exactly harmonic and each period is slightly different from the previous or subsequent one. The further the periods, the greater the differences apart.

From the same G and can be one minor Averaging According to claim 12 lead to an improvement in sound quality.

In the embodiment variant of claims 3, 7 and 8, all harmonics of the input signal are amplified, which is generally desirable. Alternatively, however, only half the period can also be used for the delay according to claims 9 and 10. In this case, when the delayed signal is added to the undelayed signal, the frequencies IW, 3 IW, 5 IW, etc. are suppressed and the remaining harmonics are amplified. The effect corresponds to a comb filter. Even though the fundamental vibration is extinguished, speech signals are easy to understand. Alternatively, the delayed signal can be subtracted from the undelayed signal, thereby zeroing out the frequencies 2 IW, 4 IW, 6 IW etc. arise.

Another improvement in the distinction can according to claim 11 can be achieved. You can the individual harmonics are specifically extracted from the signal using filters become.

Speech signals generally indicate strong harmonic components, but also noise components. Vice versa have interference signals generally great Noise components, but certain machine noises are harmonic Know components. Not about harmonic interference signals during pauses in speech in unwanted Way to reinforce a speech recognition can be carried out which the signal processing controls.

The invention further relates to a Signal processing apparatus for performing one of the above Method.

As a result, the present Invention based on the embodiments shown in the figures explained in more detail.

Show it

1 and 2 Block diagrams of circuits used in the present invention

3 to 6 Block diagrams of various embodiments of the invention, and

7 and 8th Diagrams which show a noisy input signal or an output signal which has been obtained with the method according to the invention.

In 1 A general circuit is shown in which an instant amplitude signal IA, an instant phase signal IFI and an instant angular frequency signal IW are obtained from an input signal. In a first block 1 the input signal S is _converted into an analytical signal which consists of a real part Re and an imaginary part Im. Since the real part and the imaginary part have a constant phase difference of ^π / _Z , the imaginary part Im represents the Hilbert transform of the real part Re. These signals Re and Im are therefore also referred to as Hilbert signals. Ways of obtaining the analytical signal are in the EP 542 710 A. described. In essence, can be subjected to the input signal S _in a Hilbert transform, for example, to the imaginary to come in. Since the Hilbert transformation is associated with a delay), the input signal S _{in must} also be delayed in order to obtain the real part Re. An alternative possibility is to convert the input signal S _in different by two all-pass filters in Hilbert two signals. A further possibility for obtaining the analytical signal is to obtain a complex spectrum of the input signal S _in by means of a Fourier transformation, to shift all lines by ^π / ₂ and to reset the signal back to the time domain by inverse transformation. Due to the possibilities of digital signal processing, it is unproblematic to receive such an analytical signal in a suitable manner. In the blocks 2 and 3 the instant amplitude signal IA and the instant phase signal IFI are obtained according to formulas (1) and (2) above. By time derivative of the instant phase signal IFI in block 4 the instant angular frequency signal IW can be formed. It must be noted that the signals IA, IFI and IW are, apart from the delay due to the Hilbert transformation, real-time parameters that are free of averaging or delays. The constant amplitude IA is always non-negative, whereas the instant angular frequency signal IW is not necessarily positive. Since the instant phase signal essentially defines an angle, it can be restricted to a range between 0 and 2π or to a range between -π and π by so-called wraping.

In 2 are the individual processing steps of 1 in a single block 5 summarized to simplify the presentation in the following.

When switching from 3 is the instant amplitude signal IA in a total of three delay elements 6a . 6b . 6c delayed in order to obtain a single-delayed instant amplitude _signal IA _del1 , a double-delayed instant amplitude _signal IA _del2 and a three-time delayed instant amplitude _signal IA _del3 . By adding IA, IA _del1 , IA _del2 and IIA _del3 in block 8th and division by four in block 10 an arithmetic mean of these signals IA, IA _del1 , IA _del2 and IA _{del3 is} generated, which is referred to as the amplitude mean IA _{a v} . In the above averaging, not all signals IA, IA _del1 , IA _del2 and IA _{del3 have} to be evaluated equally; for example, the instant amplitude _signal IA can be weighted more strongly than the single-delayed instant amplitude _signal IA _del1 and this in turn stronger than twice delayed instant amplitude _signal IA _del2 and so on. This way, a weaker but more robust noise reduction reached.

In the same way, the instant phase signal IFI is divided into a total of three delay elements 7a . 7b . 7c delayed to obtain a single-delayed instant phase signal IFI _del1, a double-delayed instant phase signal IFI _del2 and a three-time delayed instant phase signal IFI _del3 . An analogous way is in the blocks 9 and 11 a phase mean IFI _{a v is} formed.

In the block 12 the output signal S _{out is} calculated according to the following formula (3). S out = IA a v · Cos (IFI a v ). (3)

In the block 13 the time period t _{del is} calculated according to formula (4) from the instant frequency signal IFR: t del = 1 / I - R = 2π / IW, (4) where IW represents the instant angular frequency signal, which is the time derivative of the instant phase signal IFI.

The time period t _del is accordingly continuously updated and sent to the delay elements 6a . 6b . 6c respectively. 7a, 7b, 7c transmitted to delay the signals IA and IFI du -ch.

4 shows an old-native embodiment, in which only one delay element 6 respectively. 7 is provided, and in which the delayed signal IA _del or IFI _{de l is} multiplied by a constant factor k and to one of the delay element 6 respectively. 7 upstream adder 16 respectively. 17 is returned to receive the signals IA _av and IFI _av . In this way, a kind of moving average of the signals IA or IFI with the one- and multiple-delayed signals is formed. For the constant factor k, 0 <k <1 applies, a larger value of meaning no greater weighting of the delayed signals. In the blocks 18 respectively. 19 the signals IA _{a v} and IFI _{a v} are normalized by multiplication by (1 - k) and then the block 12 supplied to calculate the output signal S _out .

The circuit according to 5 has the following differences from the circuit of 3 on:
First in block 6 a signal 5 delayed, which corresponds directly to the input signal S _in to obtain a delayed signal S _del . In itself, the block could 6 the input signal S _in be fed directly. However, this is not the case in the embodiment variant shown, since there is a signal here 5 that is delayed from the analytical signal in block 12 is re-synthesized according to Formula 3. This has the advantage that the delay caused by the Hilbert transformation in the calculation of the instant frequency signal IFR is compensated for. There is also the possibility of signal processing in the frequency domain, such as compression or expansion.

Another difference of this embodiment variant to that described above is that the signal S is delayed by a time period t _del which corresponds to the reciprocal of the double instant frequency IFR, as shown in the formula (4a): t de l = 1 / (2 IFR) = π / IW, (4a)

This means that the signal 5 is only delayed by half the current oscillation period. In the block 20 the delayed signal S _del can be multiplied by a constant factor, for which -1 ≤ 0 ≤ 1 applies in this case. As described above, for k = 1 the fundamental and the harmonics are canceled with the triple, five times, etc. frequency, while for k = -1, the harmonics with the double, four times, etc. frequency are canceled.

In the variant of 6 the signal corresponding to the input signal S _in parallel with the filters 21a . 21b . 21c fed that a combination of the blocks 6 and 8th the 5 correspond. In the block 13b values for the time period t _{del are} calculated which correspond to the respective reciprocal of a simple integer multiple of the instant frequency IFR, as shown in the formula (4b): t del = 1 / (n IFR) = 2π / (π IW), (4b) where n = 1, 2, 3 ... applies. In this way it is possible to target individual harmonics from the signal 5 to extract and in an adder 22 to compose an output signal S _out .

In the 7 a spectrum of a noisy signal is shown. The harmonic components are clearly recognizable as peaks.

After the method according to the invention has been carried out, a signal having a spectrum according to 8th reached. It can be clearly seen that the noise components are significantly reduced, which improves the signal / noise ratio accordingly.

Claims (14)

Method for suppressing interference noise with the following steps: - obtaining an analytical signal from an input signal (S _in ) containing interference noise; - Calculating an instant phase signal (IFI) from the analytical signal; - Calculating an instant frequency signal (IFR) as a time derivative of the instant phase signal (IFI) divided by 2π; - Forming a single delayed signal (S _del , IA _del , IFI _del , S _del1 , IA _del1 , IFI _del1 ) by delaying a signal (S, IA, IFI) from the group containing the input signal and signals derived therefrom a time period (t _del ) that your reciprocal of a simple corresponds to integer multiples of the instant frequency signal (IFR); Formation of a signal mean value (S _{a v} , IA _{a v} , IFI _{a v} ) from the signal (S, IA, IFI) and the simply delayed signal (S _del , IA _del , IFI _del , S _del1 , IA _del1 , IFI _del1 ) in order to obtain an output signal (S _out ) in which the noise is suppressed. A method according to claim 1, characterized in that further multiple delayed signals (S _de2 , IA _de2 , IFI _de2 , S _de3 , IA _de3 , IFI _de3 ) are formed by the signal (S, IA, IFI) several times in succession for a period of time (t _del ) is delayed, which corresponds to the reciprocal of a simple integer multiple of the instant frequency signal (IFR), and that the signal mean (S _av , IA _av , IFI _{a v} ) from the signal (S, IA, IFI) and the single and the multiple delayed signals (S _{del 1} , IA _del1 , IFI _del1 , S _del2 , IA _del2 , IFI _{del 2} , S _del3 , IA _del3 , IFI _{del 3} ) are formed. Method according to one of claims 1 or 2, characterized in that the single delayed signal (S _del , IA _del , IFI _del ; S _del1 , IA _del1 , IFI _{del 1} ) by delaying the signal (S, IA, IFI) by one Time period (t _del ) is obtained, which corresponds to the characteristic value of the instant frequency signal (IFR). Method according to Claim 1, characterized by the following steps: - obtaining an analytical signal from an input signal (S _in ); - calculating an instant amplitude signal (IA) from the analytical signal; - Calculating an instant phase signal (IFI) from the analytical signal; - Calculating an instant frequency signal (IFR) as a time derivative of the instant phase signal (IFI) divided by 2π; - Forming a single delayed instant amplitude _signal (IA _del ; IA _del1 ) by delaying the instant amplitude _signal (IA) by a time period (t _del ), which corresponds to the reciprocal of a simple integer multiple of the instant frequency signal (IFR); - forming a singly delayed instant phase signal (IFI _del; IFI _del1) by delaying the instant the phase signal (IFI) for the time (t _{del) which} corresponds to the reciprocal value of a simple integer multiple of the instant frequency signal (IFR); - forming an amplitude center value (IA _{a v)} of the instant signal amplitude (IA) and the singly delayed instant amplitude Slgnal (IA _del; IA _del1); - Forming a phase average (IFI _{a v} ) from the instant phase signal (IFI) and the single-delayed instant phase signal (IFI _del ; IFI _del1 ); - Linking the mean amplitude value (IA _av ) with the phase mean value (IFI _av ) to an m output signal (S _out ) in which the noise is suppressed. A method according to claim 4, characterized in that further multiple delayed instant amplitude signals (IA _del2 ; IA _del3 ) are formed in that the instant amplitude _signal (IA) is delayed several times in succession by a time period (t _del ) which corresponds to the reciprocal of a simple Integer multiples of the instant frequency signal (IFR) corresponds to the fact that multiply delayed instant phase signals (IFI _del2 ; IFI _del3 ) are formed by delaying the instant phase signal (IFI) several times in succession by a time period (t _del ) that is the reciprocal a simple integer multiple of the instant frequency signal (IFR) corresponds to the fact that the mean amplitude value (IA _av ) from the instant amplitude signal (IA) and the single and multiple delayed instant amplitude signals (IA _del ; IA _del1 ; IA _del2 ; IA _del3 ) is formed and that the phase mean (IFI _{a v} ) from the instant phase signal (IFI) and the single and the multiple delayed instant phase ignalen (IFI _del ; IFI _del1 ; IFI _del2 ; IFI _del3 ) is formed. Method according to Claim 4, characterized in that the mean amplitude value (IA _{a v} ) is obtained by adding the instantaneous amplitude signal (IA _del ), which has been multiplied by a constant factor (k), to the instant amplitude signal (IA) and that the phase mean value (IFI _av ) is obtained by adding the simply delayed instant phase signal (IFI _del ) multiplied by a constant factor (k) to the instant phase signal (IFI). Method according to one of claims 4 to 6, characterized in that the simply delayed instant amplitude _signal (IA _del ; IA _del1 ) is obtained by delaying the instant amplitude _signal (IA) by a time period (t _{d e l} ) which is the reciprocal of the instant frequency signal (IFR). Method according to one of claims 4 to 6, characterized in that the simply delayed instant phase signal (IFI _del ; IFI _del1 ) is obtained by delaying the instant phase signal (IFI) by a time period (t _del ) which corresponds to the reciprocal of the instant -Frequency signal (IFR) corresponds. Method according to one of claims 4 to 6, characterized in that the simply delayed instant amplitude _signal (IA _del ; IA _del1 ) is obtained by delaying the instant amplitude _signal (IA) by a time period (t _del ) which is half the reciprocal of the Instant frequency signal (IFR) corresponds. Method according to one of claims 4 to 6, characterized in that the simply delayed instant phase signal (IFI _del ; IFI _del1 ) is obtained by delaying the instant phase signal (IFI) by a time period (t _del ) which is half the reciprocal of the Instant frequency signal (IFR) corresponds. Method according to one of claims 4 to 6, characterized in that that several signal mean values are obtained in parallel with one another, which are combined into an output signal. Method according to one of claims 1 to 11, characterized in that the instant frequency signal or the time duration (t _del ), which corresponds to the reciprocal of a simple integer multiple of the instant frequency signal, are averaged over time before the delays are carried out. Method according to one of claims 1 to 12, characterized in that that in addition a speech recognition process is carried out. Device for suppressing interference noise for carrying out a method according to one of claims 1 to 13, with a circuit for obtaining an instant frequency signal (IFR) from an input signal (S _in ), characterized in that a simply delayed signal (S _del , IA _del , IFI _del ; S _del1 , IA _del1 , IFI _{del 1} ) by delaying a signal (S, IA, IFI) from the group containing the input signal and signals derived therefrom by a time period (t _del ) which is formed corresponds to the reciprocal of a simple integer multiple of the instant frequency signal (IFR), and that a signal mean (S _av , IA _av , IFI _av ) from the signal (S, IA, IFI) and the single delayed signal (S _del , IA _del , IFI _del ; S _del1 , IA _del1 , IFI _{del 1} ) is formed.