DE102013011761A1 - Motor vehicle has estimating unit and background noise spectrum unit that are designed to open dynamic filter with low background noise and close with strong background noise - Google Patents

Abstract

The motor vehicle has a hands-free system that is provided with a microphone for detecting a disturbed speech signal (x). The hands-free system is connected to signal processing unit for noise reduction of the background noise of the disturbed speech signal. A filter unit configured for generating a hands-free frequency response. An estimating unit (10) of a noise spectrum and background noise spectrum unit (11) for a speech signal are designed to open dynamic filter (7) with a low background noise and close with a strong background noise. An independent claim is included for a method for generating frequency response for hands-free for motor vehicle.

Description

The invention relates to a motor vehicle with a hands-free device, which comprises at least one microphone for detecting a disturbed speech signal x, which is connected to a signal processing unit for noise reduction of the background noise n of the disturbed speech signal x, which includes a filter device for generating a handsfree frequency response, and a method for generating a frequency response for the hands-free device in response to the background noise n.

Hands-free devices in motor vehicles are well known. As a rule, such hands-free devices require a, connected to a microphone signal processing unit for noise reduction. At the same time, the signal processing unit allows echo cancellation to enable hands-free operation. A known signal processing unit for noise reduction is in 7 shown. The microphone picks up the disturbed speech signal x, which corresponds to the sum of speech signal s and background noise n, ie x = s + n. In a filter device, the disturbed speech signal x is filtered. With 3 designated first filter means H _ref generates the handsfree reference frequency response. For this frequency response there is a recommendation from the International Telecommunication Union (ITU_T P.1110 of December 2011). After this first filter device, the actual noise reduction takes place with 4 designated second filter. This noise reduction is flawed because the background noise can only be estimated in rough approximation. The background noise estimate may differ greatly from the current background noise n and the difference is audible as musical noise (also known as musical tones). Usually in another, with 5 third filter reduced artifacts of the second filter. At the exit 6 the signal processing unit results in a background noise reduced signal y = s '+ n'.

In motor vehicles, the low frequencies below 500 Hz are particularly disturbed at higher speeds of the motor vehicle or fan noise, so that the handsfree reference frequency response is set by the filter device H _ref to attenuate the low frequencies.

The handsfree reference frequency response for vehicle handsfree according to the recommendation ITU_T P.1110 is in 8th shown. The tolerance band T determined by the standard is defined between 100 Hz to 8000 Hz. In this tolerance band T is an example of a handsfree frequency response of the filter device H _{ref is located} . 8th shows the filter magnitude ie level values above the frequency i.

The frequencies below 500 Hz are strongly attenuated in order to relieve the background noise reduction or to reduce coarse estimation errors. On the other hand, however, the speech signal s _{ref is} strongly affected by the filter device H _ref , since the low-frequency speech components are reduced, which is why the sound of the speech is unnaturally thin.

The object of the invention is to provide a motor vehicle with a hands-free device, which outputs a better quality voice signal.

The invention results from the features of the independent claims. Advantageous developments and refinements are the subject of the dependent claims. Other features, applications and advantages of the invention will become apparent from the following description and the explanation of embodiments of the invention, which are illustrated in the figures.

The object is achieved with a motor vehicle with a hands-free system, in which a filter H _dyn for determining a filter characteristic H _dyn (i) an estimation unit of the noise spectrum N _est (i) and a unit for a reference spectrum S _ref (i) for a speech signal s , which assigns a level value to each frequency i, and which are designed so that the filter H _dyn opens at a low background noise n and closes at a strong background noise n. Opening or closing means in the present context that an attenuation of predetermined frequencies is reduced or increased. Compared to the prior art embodiment, the filter H _dyn replaces the first filter H _ref . The fact that the filtering is set to be noise-dependent, a strong attenuation of the low frequencies is only when it is needed, that is, when actually occur in this frequency range strong interference. If there is little or no interference, the filter leaves the low frequencies largely undamped. If only slight noises are generated in a driving situation or if the vehicle is even standing, then a very high voice quality with low frequency components can be achieved. The low speech frequencies contribute to the "warmth" and naturalness of the speech signal.

Due to the dynamic adaptation of the filter to the actual noise spectrum N _est (i) of the background noise, a high attenuation occurs at low frequencies, which are below 500 Hz, in order to suppress driving noise. At the damping is waived, if strong Background noise is not present. By such a method, the use of the frequency band for wide-band telephony to z. B. 80 Hz possible.

In a preferred embodiment, the filter H _dyn for determining a filter characteristic H _dyn (i) comprises a computing unit for determining a further filter, preferably a Wiener filter, which is connected to the estimation unit for the background noise spectrum N _est (i) to be determined and the unit for the reference spectrum S _ref (i) for the speech signal s is connected. Advantageously, the other filter for setting the handsfree frequency response is based on the Wiener filter design. The Wiener filter is calculated from the two spectra: the estimated spectrum N _est (i) of the background noise n and the reference spectrum S _ref (i) of the language s. Under a Wiener filter is generally understood an algorithm in which the minimization of the mean square error of the (here background noise) is sought. By means of the estimation unit, the occurring background noise n is estimated. In this case, for example, methods are used which automatically follow the minimum of the input signal of the signal processing unit and thereby detect the background noise n. Further details of this calculation will be explained later.

In one variant, the estimation unit recognizes in each case a minimum of a volume of the disturbed speech signal x in the speech pauses and thus uses the speech pause for background noise estimation. This achieves a good approximation to the actual background noise n.

In one variant, the unit for a reference spectrum for the speech signal s assigns to each frequency i a constant level value or each frequency i one but different level value, which preferably corresponds to one of a male and / or female averaged speech spectrum. The selected function of the reference spectrum for the speech signal s provides a largely faithful reproduction of speech.

A development of the invention relates to a method for generating a frequency response for hands-free devices, in particular for a motor vehicle, in which in a known manner background noise frequencies which are underlaid a speech signal s, filtered according to a predetermined characteristic during transmission. According to the invention, an estimated value N _est (i) of a spectrum of the background noise n and a reference characteristic S _ref (i) of a spectrum of a speech signal s are used to determine the filter characteristic H _dyn (i). The reference characteristic S _ref (i) assigns a constant noise level to each frequency i, thus minimizing artifacts.

The dynamic filter characteristic H _dyn (i) allows for high damping in the event of heavy driving noise. The attenuation is omitted if strong background noise is not present. By such a method, the use of the frequency band for white-band telephony is possible up to 80 Hz, for example.

In an embodiment, the filter characteristic H _dyn (i) for i <i _{0 is raised} with low background noise and lowered at high background noise and is preferably kept constant in the wider frequency range for i> i ₀ . This method makes it possible to dynamize the filter characteristic H _dyn (i) in a frequency range i below a selectable limit frequency i ₀ , ie to set a filter characteristic H _dyn (i) dependent on the background noise n for each frequency i. Hereby a realistic reproduction of speech signals s is possible at low background noise n.

In one embodiment, to determine a filter characteristic H _dyn (i) a reference characteristic S _ref (i) is used, which is determined from a speech reference _spectrum , wherein the reference characteristic S _ref (i) is adjusted by an estimate of the actual speech volume. Including an estimated speech volume leads to a realistic reproduction of speech signals.

In a further embodiment, the values of the reference characteristic S _ref (i) and / or the estimated values N _est (i) are weighted. The weighting of the speech spectrum or of the noise spectrum makes it possible to improve the subjective hearing of the users of the hands-free device. Alternatively or in addition to the weighting, the spectra are modified via a root function. By means of this root function, large and small values are weighted in a different relation to each other, which can also improve the subjective hearing impression.

In a further refinement, the standardized characteristic speech spectrum, ie an "artificial-voice" spectrum, is used as the reference characteristic S _ref (i). "Artificial-Voice" spectra are known for male and female speakers, and as an alternative, the mean of the two spectra can be used. The use of Artificial Voice spectra further enhances the listening experience of reproduced speech.

Further advantages, features and details will become apparent from the following description in which - where appropriate, with reference to the drawings - at least one embodiment is described in detail. Described and / or illustrated features may form the subject of the invention itself or in any meaningful combination, optionally also independent of the claims, and in particular may also be the subject of one or more separate application / s. The same, similar and / or functionally identical parts are provided with the same reference numerals.

Show it:

1 A first embodiment of a filter H _{dyn according} to the invention for the dynamic generation of the hands-free frequency response,

2 an embodiment of the signal processing _unit according to the invention with the filter H _dyn according to 1 in combination with a well-known noise reduction,

3 A second embodiment of the signal processing unit with the filter H _dyn according to 1 in combination with a well-known noise reduction,

4 Exemplary embodiments for the method according to the invention, with 3 examples of characteristic curves generated from filter H _dyn in comparison to a characteristic curve generated from the filter H _ref according to ITU recommendation,

5 a first embodiment of the arrangement of the filter H _dyn for the handsfree response according to 1 in a handsfree,

6 a second embodiment of the arrangement of the filter H _dyn for the handsfree response according to 1 in a handsfree,

7 a schematic diagram of a signal processing unit for noise reduction according to the prior art,

8th the history of the state-of-the-art hands-free frequency response.

1 shows a first embodiment of the dynamic filter H _dyn 7 for adjusting the frequency response of the background noise, in which the noise-controlled frequency response is calculated and applied to the disturbed speech signal x = s + n. The input signal of the dynamic filter H _dyn 7 is a disturbed speech signal x coming from the microphone 2 is recorded. The disturbed speech signal x comprises a speech signal s and a background noise n. An input stage of this filter H _dyn 7 is through a pre-filter 8th educated. Here, the frequency response deviates from the ITU_T P.1110 recommendation to a frequency response H0. The frequency response H0 can usually be the neutral, linear frequency response, which corresponds to a straight line over the frequency axis.

The pre-filter 8th with the frequency response H0 can take over the function of a desired frequency response, z. B. in addition to the neutral, linear frequency response for better speech intelligibility to have a slightly increasing frequency response. Furthermore, H0 can have a frequency response that strongly attenuates very low frequencies below 80 Hz in order to filter out the unimportant voice components under 80 Hz. In a simplified embodiment of the dynamic filter H _dyn 7 can the pre-filter 8th also omitted.

After setting the desired frequency response H0, this input signal becomes another unit 12 supplied to provide the filter characteristics H _dyn (i). To automatically determine a dynamized frequency response and associated filter characteristic H _dyn (i), an estimate of the spectrum of the background noise N _est (i) is required, which is in the estimation unit 10 provided. The background noise is estimated by methods such as those used in the known standard noise reduction method for the noise reduction filter 4 in 7 be used. Such noise reduction methods are for example from EP 2 151 821 A1 known. These methods automatically follow the minimum of the input signal x and thereby can detect the background noise.

Another variable for determining the filter H _dyn for generating the dynamic frequency response is a reference spectrum S _ref (i) for the speech signal s, which is in the unit 11 provided. In the arithmetic unit 9 a Wiener filter is determined and the filter characteristic H _dyn (i) in the further unit 12 provided. This results H _dyn (i) = 1 - N _est (i) / (N _est (i) + S _ref (i)).

In this case, spectral quantities are used, where i denotes the frequency. In the proposed arrangement, the lower frequencies of the background noise n are to be filtered below, for example, 500 Hz. The calculation can thus take place up to a frequency i <i ₀ , where i ₀ corresponds to 500 Hz, for example. For higher frequencies i> i ₀ , H _dyn (i) = 1 is set. In the spectral notation results for the output signal y Y (i) = X (i) × H _dyn (i).

In spectral notation, the capital letters X and Y are used instead of the small letters for x and y. The speech spectrum S _ref (i) is assumed to be constant for a given frequency, which is why no "musical" artifacts occur. Thus, a spectrum is set for the language as S _ref (i). The further unit 12 to provide the filter characteristics H _dyn (i) thus changes its filter characteristics only very slowly, adapting to the estimated background noise n. This estimation of the background noise n takes place with a large time constant, for example of the order of one second. In contrast, the filters of noise reduction in hands-free systems known from the prior art are subject to artifacts due to a considerably higher frequency of change (rate of change is in the range of 20 ms).

As a suitable value for S _ref (i), the standardized mean speech spectrum of the so-called "artificial voice" can be used. Artificial voice spectra are known for male and female speakers. Here can z. For example, the mean of these two spectra, consisting of male and female speakers, may be used. The spectrum of the mean value is described by means of a characteristic curve which in each case assigns a level value or a filter magnitude to a frequency i.

It is also possible to estimate the actual speech volume and slowly track it. Here, female and male speakers can be recognized and the artificial voice spectrum can be switched accordingly. Furthermore, an estimated value of the actual speech signal s could be determined in order to use this as a reference spectrum S _ref (i).

An embodiment for the further unit 12 to provide the filter characteristics H _dyn (i) for generating the dynamic frequency response results from the fact that z. B. Spectral weight functions are introduced to improve the subjective hearing. With a weight function Gn (i) for the sound spectrum and a weight function Gs (i) for the speech spectrum H _dyn (i) = 1 - Gn (i) × N _est (i) / (Gn (i) × N _est (i) + Gs (i) × S _ref (i)).

Alternatively, a modification with a root function is possible. In this embodiment, large and small perturbations in a different relation to each other are weighted, which improves the subjective listening experience. In this case follows H _dyn (i) = 1 - sqrt (N _est (i) / N _est (i) + S _ref (i)).

Then the filter characteristic H _dyn (i) of the filter H _dyn 7 generated frequency response advantageously temporally smoothed. H _{dyn_smooth} (i, t) = 0.7 × H _{dyn_smooth} (i, t-1) + 0.3 × H _dyn (i, t).

Where t is the current discrete time, where t-1 is the value that preceded by one clock. For filtering then H _{dyn_smooth is} used.

Other characteristics or extensions of the characteristic are also possible, for. For example, combinations of weight functions and exponential functions over individual variables or the entire equation.

An example of the use of the noise-controlled filter H _dyn 7 in the known per se signal processing unit 1 for noise reduction of background noise is in 2 shown. The noise-controlled filter is H _dyn 7 a second filter 4 for noise reduction and a third filter 5 preceded by the reduction of artifacts, which is why an output signal at the output 14 of y = s '+ n'.

An alternative is in 3 see where the noise-controlled filter H _dyn 7 the second filter 4 and third filter 5 is downstream.

4 shows examples of the course of the frequency response generated by the filter H _dyn , wherein the filter amplitude that is a noise level above the frequency i is stored. In this case, three different noise situations are considered as an example. In the filter characteristic H _dyn 1, a situation is considered in which the motor vehicle runs at a speed of 100 km / h. The filter characteristic H _dyn 2 is relevant when the motor vehicle is traveling at 140 km / h. In an air flow with a fan, such as an air conditioner, which is set at 7 fan speeds to the maximum fan speed 7, a filter characteristic H _dyn 3 is effective. For comparison with the prior art, the filter characteristic H _{ref was} plotted. In the prior art, the constant filter H _ref would always be used in all three exemplary noise situations. In this 4 it is clear that in the range below 200 Hz almost a gain of almost 10 decibels between the standard characteristic H _ref and the background noise-controlled filter characteristic H _dyn 1 is achieved, which is an improvement in favor of the natural speech signal s. On the other hand, the strong low-frequency noise below 200 Hz is more strongly attenuated in the case of a very strong fan noise compared to the characteristic H _ref with the filter characteristic H _dyn, thereby further reducing the disturbance.

In 5 is an example of a hands-free system noise reduction unit 1 and echo cancellation unit 19 shown. The noise reduction unit 1 comprising the dynamic filter H _dyn 7 , the second filter 4 for noise reduction and the third filter 5 to reduce artifacts are the echo cancellation unit 19 downstream.

Alternatively, in 6 shown that the noise-controlled filter H _dyn 7 according to 1 in the echo cancellation unit 19 is inserted. It is advantageous here that the echo canceller 16 can work with less interference by low-frequency noise. To the convergence of the echo canceller 16 not interfering with the slow change from the noise-controlled frequency response of the filter H _dyn becomes a copy 18 from the noise-controlled filter H _dyn 7 in the reference branch after the echo canceller 16 used. The second filter 4 for noise reduction and the third filter 5 to reduce artifacts are the echo cancellation unit 19 downstream.

Although the invention has been further illustrated and explained in detail by way of preferred embodiments, the invention is not limited by the disclosed examples, and other variations can be derived therefrom by those skilled in the art without departing from the scope of the invention. It is therefore clear that a multitude of possible variations exists. It is also to be understood that exemplified embodiments are really only examples that are not to be construed in any way as limiting the scope, applicability, or configuration of the invention. Rather, the foregoing description and description of the figures enable one skilled in the art to practice the exemplary embodiments, and those skilled in the art, having the benefit of the disclosed inventive concept, can make various changes, for example, to the function or arrangement of individual elements recited in an exemplary embodiment, without Protected area, which is defined by the claims and their legal equivalents, such as further explanation in the description.

LIST OF REFERENCE NUMBERS

1

Signal processing unit for noise reduction

2

microphone

3

Filter device H _ref

4

second filter for noise reduction

5

third filter for noise reduction

6

Output dynamic filter H _dyn

7

dynamic filter H _dyn

8th

prefilter

9

Calculation unit for the determination of another filter (Wiener Filter)

10

Estimation unit for background noise spectrum

11

Unit for a reference spectrum

12

further unit for providing the filter characteristics H _dyn (i)

14

Output signal processing unit for noise reduction

15

Output of the handsfree system

16

echo canceller

17

speaker

18

Copy from the sound-controlled filter 7

19

Echo cancellation unit

QUOTES INCLUDE IN THE DESCRIPTION

This list of the documents listed by the applicant has been generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Cited patent literature

• EP 2151821 A1 [0031]

Claims (10)

Motor vehicle with a hands-free device, which is a microphone ( 2 ) for detecting a disturbed speech signal x, which is connected to a signal processing unit ( 1 ) for a noise reduction of the background noise n of the disturbed speech signal x, which comprises a filter device ( 3 ) for generating a hands-free frequency response, characterized in that the filter device as a dynamic filter H _dyn ( 7 ) is designed to determine a filter characteristic H _dyn (i) an estimation unit of a noise spectrum N _est (i) ( 10 ) and a reference _spectrum unit S _ref (i) ( 11 ) for a speech signal s, which assigns a level value to each frequency i, and which are designed such that the filter H _dyn ( 7 ) opens at a low background noise and closes at a strong background noise. Motor vehicle according to claim 1, characterized in that the filter H _dyn ( 7 ) for determining a filter characteristic H _dyn (i) a computing unit ( 9 ) for determining a further filter, preferably a Wiener filter, which is connected to the estimation unit ( 10 ) for the background noise spectrum to be determined and the unit for the reference spectrum ( 11 ) is connected to the speech signal s. Motor vehicle according to one of claims 1 or 2, characterized in that the estimation unit ( 10 ) each evaluates a minimum of a volume of the disturbed speech signal x during pauses in speech. Motor vehicle according to one of claims 1 to 3, characterized in that the unit for a reference spectrum ( 11 ) assigns to the speech signal s every frequency i a constant level value or to each frequency i different level values which preferably correspond to one of a male and / or female averaged speech spectrum. Method for generating a frequency response for hands-free devices, in particular for a motor vehicle, in which the disturbed voice signal x, are filtered according to a predetermined filter characteristic during transmission, characterized in that for determining the filter characteristic H _dyn (i) an estimated value N _est (i) a spectrum of the background noise n and a reference characteristic S _ref (i) of a spectrum of a speech signal s is used. A method according to claim 5, characterized in that the filter characteristic H _dyn (i) for i <i _{0 is raised} with low background noise and lowered at high background noise and preferably kept constant in the wider frequency range for i> i ₀ . A method according to claim 5 or 6, characterized in that for determining the filter characteristic H _dyn (i) a reference characteristic S _ref (i) is used, which is determined from a speech reference _spectrum , wherein the reference characteristic S _ref (i) by an estimate of the actual Voice volume is adjusted. Method according to one of claims 5 to 7, characterized in that to determine the filter characteristic H _dyn (i) values of the reference characteristic S _ref (i) and / or the estimated values N _est (i) are weighted. Method according to one of claims 5 to 8, characterized in that the filter characteristic H _dyn (i) is modified via a root function. Method according to one of Claims 5 to 9, characterized in that a "artificial-voice" spectrum is used as the reference characteristic S _ref (i).

Images