EP3111667B1 - Method and system for automatic acoustic equalisation - Google Patents

Description

Field of the invention

The present invention relates to the field of sound signal processing.

The present invention relates more particularly to a method and an automated acoustic equalization system.

A case of use of the present invention is as follows: in the automotive field: a plurality of microphones are placed in a vehicle and pre-calibrated sound sequences are broadcast in the vehicle speakers. A system compares the sound signals emitted and the sound signals received and recorded. We deduce the "acoustic signature" of the passenger compartment of the vehicle. The user then defines a target acoustic signature curve, which is different from the vehicle's native acoustic signature. A second algorithm calculates digital filter coefficients so that, when these filters are applied before sound signals are broadcast in the vehicle speakers, the acoustic signature of the vehicle becomes the target sound signature curve, not the native acoustic signature of the vehicle.

In the context of the present invention, "IIR" or "Infinite Impulse Response" filters are used in English terminology.

More particularly, in the context of the present invention, so-called "Biquad" filters of the second order are used.

The method and the system according to the present invention concern the equalization of the amplitude of the frequency response of the passenger compartment.

State of the art

The document US2007 / 0025559 discloses an automated acoustic equalization system comprising the features of the preamble of the independent claims 1 and 7. It is known in the state of the art the French patent application No. FR 2 967 848 (Scientific and Technical Center for Building), which relates to a spectrum correction system intended in particular for a theater. This patent application of the prior art describes an electroacoustic system comprising a plurality of cells. In these cells are provided: an equalization device, a transmitter, a receiver, an amplification circuit for amplifying the signals from the receiver to said transmitter, and a computing element which will act, among other things, on the device of equalization. This technical solution of the prior art proposes to measure the emitting response, a theater observed at the receiver by using, in particular, a measurement signal supplied by a noise generator or any other measurement method making it possible to observe the system response in open loop. The equalizer device works to make said response come closest to a desired response. This technical solution of the prior art has for application the modification of the acoustic properties of theaters.

The prior art is also known from US Pat. US 6,721,428 B1 (Texas Instruments), an automatic speaker equalizer. This prior art US patent relates more particularly to a method for generating digital filters for equalizing a loudspeaker. First digital data is provided, for a tolerance interval for a tone-based target response response versus frequency for the loudspeaker. Second digital data is generated, for an actual acoustic signal response curve as a function of frequency for the loudspeaker. The first digital data is compared with the second digital data, and it is determined whether the actual response curve is in the tolerance range. If the actual response curve is not in the tolerance range, digital audio filters are iteratively generated, and the digital audio filters are applied to the second digital data to generate third digital data for a compensated response curve. . The frequency, gain and bandwidth of the digital audio filters are automatically optimized until the compensated response curve is within the tolerance range or a predetermined limit of the number of digital audio filters has been reached. of the two taking place first.

It is also known in the state of the art the scientific publication "Filter Design Method for Loudspeaker Equalization Based on IIR Parametric Filters" by German Ramos and Jose J. Lopez .

Presentation of the invention

The present invention provides a method for providing equalization of a signal by determining filter parameters to reduce the difference between the amplitude of a frequency response representing the acoustic signature of a set of speakers in their environment and a target sound signature curve.

• Mesure de N réponses impulsionnelles RI₁, RI₂, ..., RI_N après émission d'un signal sonore précalibré reçu par N microphones ;
• Calcul des N réponses fréquentielles correspondantes par Transformée de Fourier rapide ;
• Etablissement d'une moyenne M des N réponses fréquentielles ;
• Traduction en échelle fréquentielle logarithmique de ladite moyenne M des N réponses fréquentielles ;
• Interpolation d'une courbe de signature sonore cible C_ec à partir d'un certain nombre de points choisis par un utilisateur ;
• Traduction en échelle fréquentielle logarithmique de ladite courbe de signature sonore cible C_ec ;
• Comparaison de ladite réponse moyennée M et de ladite réponse cible C_ec, en calculant la différence entre ladite réponse moyennée M et ladite réponse de signature sonore cible C_ec ;
• Analyse de la courbe C_diff résultant de la différence entre ladite réponse moyennée M et ladite réponse cible C_{ec ;} et
• Détermination de paramètres de filtres pour la réduction de la différence entre ladite réponse moyennée M et ladite réponse cible C_ec en traitant tout d'abord les maxima locaux par ordre décroissant suivant leur gain, puis les minima locaux, et en réalisant des itérations successives ;

ledit procédé comportant en outre une étape d'optimisation des paramètres des filtres afin d'améliorer la performance du système.To this end, the present invention relates, in its most general sense, to an automated acoustic equalization method, characterized in that it comprises the following steps:

• Measuring N impulse responses RI ₁ , RI ₂ , ..., RI _N after emission of a precalibrated sound signal received by N microphones;
• Calculation of N corresponding frequency responses by Fast Fourier Transform;
• Establishment of an average M of N frequency responses;
• Logarithmic frequency scale translation of said average M of the N frequency responses;
• Interpolation of a target sound signature curve C _ec from a number of points selected by a user;
• Logarithmic frequency scale translation of said target sound signature curve C _ec ;
• Comparing said averaged response M and said target response C _ec, by calculating the difference between said averaged response M and said target sound signature response C _ec ;
• Analyzing the curve C _diff resulting from the difference between said averaged response M and said target response C _ec; and
• Determining filter parameters for reducing the difference between said averaged response M and said response target C _ec by first treating the local maxima in decreasing order according to their gain, then the local minima, and performing successive iterations;

said method further comprising a step of optimizing the parameters of the filters to improve the performance of the system.

Thus, the method according to the present invention makes it possible to obtain automated acoustic equalization thanks to a precise and optimized calculation of filter parameters.

Preferably, the interpolation step of the target sound signature curve is performed using the Hermite method.

Advantageously, said method further comprises a step of automatically optimizing the offset of the target response C _ec , repeated at each iteration.

Advantageously, said method further comprises a step of smoothing the N frequency responses.

Preferably, said method implements filters corresponding to the following types: "peak", "notch""high-shelf" and " low-shelf " depending on the shape of local maxima and local minima.

According to a particular mode of implementation, said method furthermore implements a global optimization algorithm to minimize the error.

• mesurer N réponses impulsionnelles RI₁, RI₂, ..., RI_N après émission d'un signal sonore précalibré reçu par N microphones ;
• calculer les N réponses fréquentielles correspondantes par Transformée de Fourier rapide ;
• établir une moyenne M des N réponses fréquentielles ;
• traduire en échelle fréquentielle logarithmique ladite moyenne M des N réponses fréquentielles ;
• interpoler une courbe cible C_ec à partir d'un certain nombre de points définis par un utilisateur;
• traduire en échelle fréquentielle logarithmique ladite courbe cible C_ec ;
• comparer ladite réponse moyennée M et ladite réponse cible C_ec, en calculant la différence entre ladite réponse moyennée M et ladite réponse cible C_ec ;
• analyser la courbe C_diff résultant de la différence entre ladite réponse moyennée M et ladite réponse cible C_{ec ;} et
• déterminer des paramètres de filtres pour la réduction de la différence entre ladite réponse moyennée M et ladite réponse cible C_ec en traitant tout d'abord les maxima locaux par ordre décroissant suivant leur gain, puis les minima locaux, et en réalisant des itérations successives ;

ledit procédé comportant en outre des moyens pour optimiser des paramètres des filtres afin d'améliorer la performance du système.The present invention also relates to an automated acoustic equalization system, characterized in that it comprises means for:

• measuring N impulse responses RI ₁ , RI ₂ , ..., RI _N after emission of a precalibrated sound signal received by N microphones;
• calculate the N corresponding frequency responses by Fast Fourier Transform;
• establish an average M of the N frequency responses;
• translating into a logarithmic frequency scale said average M of the N frequency responses;
• interpolating a target curve C _ec from a number of points defined by a user;
• translating into a logarithmic frequency scale said target curve C _ec ;
• comparing said averaged response M and said target response C _ec , by calculating the difference between said averaged response M and said target response C _ec ;
• analyzing the curve C _diff resulting from the difference between said averaged response M and said target response C _ec; and
• determining filter parameters for reducing the difference between said averaged response M and said target response C _ec by first processing local maxima in descending order according to their gain, then local minima, and performing successive iterations;

said method further comprising means for optimizing filter parameters to improve system performance.

Brief description of the drawings

• la Figure 1 illustre les différentes étapes du procédé selon la présente invention ;
• la Figure 2 représente la courbe de signature sonore cible C_ec au sens de la présente invention, les réponses fréquentielles dérivées des N mesures de réponses impulsionnelles, ainsi que la moyenne M ; et
• la Figure 3 illustre la détection et le classement des maxima locaux (« pics ») et des minima locaux (« creux »).

The invention will be better understood by means of the description, given below purely for explanatory purposes, of one embodiment of the invention, with reference to the figures in which:

• the Figure 1 illustrates the different steps of the process according to the present invention;
• the Figure 2 represents the target sound signature curve C _ec within the meaning of the present invention, the frequency responses derived from the N impulse response measurements, as well as the average M; and
• the Figure 3 illustrates the detection and ranking of local maxima ("peaks") and local minima ("troughs").

DETAILED DESCRIPTION OF THE EMBODIMENTS OF THE INVENTION

The Figure 1 illustrates the different steps of the process according to the present invention.

• Dans un premier temps, on mesure N réponses impulsionnelles RI₁, RI₂, ..., RI_N après émission d'un signal sonore précalibré reçu par N microphones.

The automated acoustic equalization method according to the present invention comprises the following steps:

• In a first step, N impulse responses RI ₁ , RI ₂ , ..., RI _N are measured after transmission of a precalibrated sound signal received by N microphones.

Then the corresponding N frequency responses are computed by Fast Fourier Transform.

Then, an average M of the N calculated frequency responses is established.

A logarithmic frequency scale translation of said average M of the N frequency responses is performed.

A target sound signature curve C _ec is interpolated and then translated into a logarithmic frequency scale.

Next, the averaged response M and the target sound signature response C _ec are compared by calculating the difference between the averaged response M and the target response C _ec .

The curve C _diff resulting from the difference between said averaged response M and said target response C _ec is analyzed.

Finally, filter parameters are determined for reducing the difference between said averaged response M and said target response C _ec by firstly processing the local maxima in descending order according to their gain, then the local minima, and performing successive iterations.

The method according to the present invention further comprises a step of optimizing the parameters of the filters in order to improve the performance of the system.

Frequency responses can be averaged "standard" (that is, with identical weights), or with different weights.

The Figure 2 represents the target sound signature curve C _ec within the meaning of the present invention, the frequency responses derived from the N impulse response measurements, as well as the average M. In the context of the present invention, a comparison of the averaged response is carried out. M and the target response C _ec , by calculating the difference between the averaged response M and the target response C _ec .

The Figure 3 illustrates the detection and ranking of local maxima ("peaks") and local minima ("troughs"). In accordance with the present invention, the local maxima (peaks) are first processed in descending order according to their gain, then the local minima (troughs) are processed in ascending order. This makes it possible to determine filter parameters for reducing the difference between the averaged response M and the target response C _ec . Successive iterations are performed.

It has been shown in scientific studies that it is better to first equalize the peaks, then the troughs. Indeed, the human ear is more sensitive to peaks than hollows.

• La courbe cible et la réponse fréquentielle moyenne sont recalculées sur une échelle logarithmique afin d'approximer la résolution non uniforme du système auditif. Ceci est réalisé par une fonction de lissage qui rééchantillonne la réponse fréquentielle sur une échelle logarithmique avec par exemple une résolution fréquentielle de 1/48 octave.
  1. i) La bande de fréquence d'optimisation « FreqRange » est appliquée comme un vecteur de poids FreqWeight qui vaut 0 en dehors de la bande de fréquence et 1 à l'intérieur de la bande de fréquence.
  2. ii) La valeur initiale de l'offset « Offset » (en dB) est calculée comme la valeur moyenne de la réponse fréquentielle moyenne dans la bande de fréquence d'égalisation : $$\mathit{Offset}=\mathit{mean}\left(C_{\mathit{ec}}\left(n_i:n_f\right)\right)$$
     
     où ni et n_f sont respectivement le premier et le dernier des points de fréquence de la bande de fréquence logarithmique d'égalisation.
  3. iii) L'algorithme d'optimisation consiste à trouver l'offset optimal, Offset, qui minimise l'erreur entre M et C_ec (=Shape+Offset), définie comme suit : $$e_{\mathit{mean}}=\frac{1}{n_f-n_i+1}{\displaystyle \sum _{k=n_i}^{n_f}\left|M\left(f_k\right)-C_{\mathit{ec}}\left(f_k\right)\right|}$$
     
     Avec C_ec =Offset+Shape
  Ceci est réalisé avec l'algorithme d'optimisation, qui calcule de façon itérative l'erreur e_mean, et recherche l'offset optimal dans un intervalle de +/- 100 dB autour de la valeur initiale. En plus de la minimisation de l'erreur e_mean, une contrainte est ajoutée au problème d'optimisation, afin de limiter le gain des pics à G_max en dB. Il est défini comme suit : $$\underset{n_i\le k\le n_f}{\max }\left|M\left(f_k\right)-C_{\mathit{ec}}\left(f_k\right)\right|\le G_{\mathit{max}}$$
  

In one embodiment, optimizing the offset of the target curve is performed as follows:

• The target curve and the average frequency response are recalculated on a log scale to approximate the nonuniform resolution of the auditory system. This is achieved by a smoothing function that resamples the frequency response on a logarithmic scale with for example a 1/48 octave frequency resolution.
  1. i) The "FreqRange" optimization frequency band is applied as a FreqWeight weight vector that is 0 outside the frequency band and 1 within the frequency band.
  2. ii) The initial Offset value (in dB) is calculated as the average value of the average frequency response in the equalization frequency band: $$\mathit{Offset}=\mathit{mean}\left({\mathrm{VS}}_{\mathit{ec}}\left({\mathrm{not}}_i:{\mathrm{not}}_f\right)\right)$$
     
     where ni and n _f are respectively the first and the last of the frequency points of the logarithmic equalization frequency band.
  3. iii) The optimization algorithm consists in finding the optimal offset, Offset, which minimizes the error between M and C _ec (= Shape + Offset), defined as follows: $$e_{\mathit{mean}}=\frac{1}{{\mathrm{not}}_f-{\mathrm{not}}_i+1}{\displaystyle \underset{k={\mathrm{not}}_i}{\overset{{\mathrm{not}}_f}{\Sigma }}\left|M\left(f_k\right)-{\mathrm{VS}}_{\mathit{ec}}\left(f_k\right)\right|}$$
     
     With C _ec = Offset + Shape
  This is done with the optimization algorithm, which iteratively calculates the error e _mean , and looks for the optimal offset within +/- 100 dB around the initial value. In addition to minimizing the e _mean error , a constraint is added to the optimization problem, in order to limit the peak gain to G _max in dB. It is defined as follows: $$\underset{{\mathrm{not}}_i\le k\le {\mathrm{not}}_f}{\max }\left|M\left(f_k\right)-{\mathrm{VS}}_{\mathit{ec}}\left(f_k\right)\right|\le {\mathrm{BOY\; WUT}}_{\mathit{max}}$$
  

Parameters and number of filters are optimized using an algorithm. The parameters f, Q and G (respectively central frequency, quality factor and gain of the biquads) are optimized from intervals of values that can be predefined by a user, and the ranges of values of Q and G may depend on frequency. Thus, for example in high frequencies, the low gain filters are more easily eliminated because they are not perceptible.

$$-G\times 0.9\le G_{\mathit{opt}}\le G\times 1.1\mathrm{}si\mathrm{}G\le 0\mathrm{}et\mathrm{}-G\times 1.1\le G_{\mathit{opt}}\le G\times 0.9\mathrm{}si\mathrm{}G>0$$

$$\mathit{QRange}\left(1\right)\le Q_{\mathit{opt}}\le \mathit{QRange}\left(2\right)$$

$$\mathit{TargetGain}-100\le \mathit{TargetGain}{g}_{\mathit{opt}}\le \mathit{TargetGain}+100$$

où f_c et G sont respectivement la fréquence centrale et le gain d'un filtre biquad modélisant le nième pic, et QRange est la plage de valeurs admissibles du facteur de qualité Q.In one embodiment, the objective is to find the optimal parameters (fc _opt , G _opt , Q _opt ) of a filter and the optimal offset of the target curve Offset _opt. The limits of the parameters are determined like this: $$\max \left(\frac{f_{\mathrm{vs}}}{2^{\frac{1}{12}}},\mathit{FreqRange}\left(1\right)\right)\le f_{{\mathrm{vs}}_{\mathit{Opt}}}\le \min \left(f_{\mathrm{vs}}\times 2^{\frac{1}{12}},\mathit{FreqRange}\left(2\right)\right)$$

$$-\mathrm{BOY\; WUT}\times 0.9\le {\mathrm{BOY\; WUT}}_{\mathit{Opt}}\le \mathrm{BOY\; WUT}\times 1.1\mathrm{}si\mathrm{}\mathrm{BOY\; WUT}\le 0\mathrm{}et\mathrm{}-\mathrm{BOY\; WUT}\times 1.1\le {\mathrm{BOY\; WUT}}_{\mathit{Opt}}\le \mathrm{BOY\; WUT}\times 0.9\mathrm{}si\mathrm{}\mathrm{BOY\; WUT}>0$$

$$\mathit{QRange}\left(1\right)\le Q_{\mathit{Opt}}\le \mathit{<figure-callout\; id="2"\; label="QRange"\; filenames="imgf0002.png"\; state="\{\{state\}\}">QRange</figure-callout>}\left(2\right)$$

$$\mathit{TargetGain}-100\le \mathit{TargetGain}{\mathrm{boy\; Wut}}_{\mathit{Opt}}\le \mathit{TargetGain}+100$$

where f _c and G are respectively the center frequency and the gain of a biquad filter modeling the nth peak and QRange is the range of admissible values of the quality factor Q.

A post-optimization process is performed. This post-optimization process consists of reclassifying the filters by increasing frequency and reoptimizing the coefficients. If a filter is canceled during this process, a new peak / dip is searched in order to output the maximum number of filters. The optimization process is implemented until the maximum number of filters is reached.

In one embodiment, the interpolation step of the target curve is performed using the Hermite method.

In one embodiment, the method according to the present invention further comprises a step of automatically optimizing the offset of the target response C _ec , repeated at each iteration.

In one embodiment, the method according to the present invention further comprises a step of smoothing the N frequency responses.

In the context of the process according to the present invention, filters corresponding to the following types are used: "peak","notch","high-shelf" "And" low-shelf "depending on the shape of local maxima (peaks) and local minima (troughs).

In some cases, it is better to choose a peak type filter. In other cases, it is better to choose a "shelf" filter. The selection of the filter is made according to whether or not a certain threshold is exceeded by the quality factor.

In one embodiment, the method according to the present invention furthermore implements a global optimization algorithm to minimize the error.

• mesurer N réponses impulsionnelles RI₁, RI₂, ..., RI_N après émission d'un signal sonore précalibré reçu par N microphones ;
• calculer les N réponses fréquentielles correspondantes par Transformée de Fourier rapide ;
• établir une moyenne M des N réponses fréquentielles ;
• traduire en échelle fréquentielle logarithmique ladite moyenne M des N réponses fréquentielles ;
• interpoler une courbe cible C_ec à partir d'un certain nombre de points définis par un utilisateur;
• traduire en échelle fréquentielle logarithmique ladite courbe cible C_ec ;
• comparer ladite réponse moyennée M et ladite réponse cible C_ec, en calculant la différence entre ladite réponse moyennée M et ladite réponse cible C_{ec ;}
• analyser la courbe C_diff résultant de la différence entre ladite réponse moyennée M et ladite réponse cible C_{ec ;} et
• déterminer des paramètres de filtres pour la réduction de la différence entre ladite réponse moyennée M et ladite réponse cible C_ec en traitant tout d'abord les maxima locaux par ordre décroissant suivant leur gain, puis les minima locaux, et en réalisant des itérations successives ;

ledit système comportant en outre des moyens pour optimiser des paramètres des filtres afin d'améliorer la performance du système.The present invention also relates to an automated acoustic equalization system, comprising means for:

• measuring N impulse responses RI ₁ , RI ₂ , ..., RI _N after emission of a precalibrated sound signal received by N microphones;
• calculate the N corresponding frequency responses by Fast Fourier Transform;
• establish an average M of the N frequency responses;
• translating into a logarithmic frequency scale said average M of the N frequency responses;
• interpolating a target curve C _ec from a number of points defined by a user;
• translating into a logarithmic frequency scale said target curve C _ec ;
• comparing said averaged response M and said target response C _ec , by calculating the difference between said averaged response M and said target response C _ec;
• analyzing the curve C _diff resulting from the difference between said averaged response M and said target response C _ec; and
• determining filter parameters for reducing the difference between said averaged response M and said target response C _ec by first processing the local maxima by order decreasing according to their gain, then the local minima, and realizing successive iterations;

said system further comprising means for optimizing filter parameters to improve system performance.

The invention is described in the foregoing by way of example. It is understood that the skilled person is able to realize different variants of the invention without departing from the scope of the patent.

Claims (7)

1. Method for automatic acoustic equalization, comprising the steps of:

   • measuring N impulse responses RI₁, RI₂, ..., RI_N following the emission of a pre-calibrated sound signal received by N microphones;

   • calculating N corresponding frequency responses using fast Fourier transforms;

   • establishing a mean M of the N frequency responses;

   • transposing said mean M of the N frequency responses onto a logarithmic frequency scale;

   • interpolating a target sound signature curve C_ec from a specific number of points defined by a user;

   • transposing said target sound signature curve C_ec onto a logarithmic frequency scale;

   • comparing said mean response M and said target response C_ec by calculating the offset between said mean response M and said target sound signature response C_ec;

   • analyzing the curve C_diff resulting from the offset between said mean response M and said target response C_ec; and characterized in that it comprises the following step:

   • determining filter parameters for reducing the offset between said mean response M and said target response C_ec by firstly processing the local maxima in descending order according to the gain thereof, then the local minima, and by carrying out successive iterations;

   said method further including a step of optimization of the parameters of the filters in order to improve the performance of the system.
2. Automated acoustic equalization method according to Claim 1, characterized in that the step of interpolation of the target sound signature curve is carried out by means of the Hermitian method.
3. Automated acoustic equalization method according to Claim 1 or 2, characterized in that it further includes a step of automatic optimization of the offset of the target response C_ec repeated on each iteration.
4. Automated acoustic equalization method according to Claim 1, 2 or 3, characterized in that it further includes a step of smoothing of the N frequency responses.
5. Automated acoustic equalization method according to at least one of the preceding claims, characterized in that it employs filters corresponding to the following types: "peak", "notch", "high-shelf' and "low-shelf' as a function of the shape of the local maxima and the local minima.
6. Automated acoustic equalization method according to at least one of the preceding claims, characterized in that it employs a global optimization algorithm to minimize the error.
7. Automated acoustic equalization system, including means for:

   • measuring N impulse responses RI₁, RI₂, ..., RI_N following the emission of a pre-calibrated sound signal received by N microphones;

   • calculating N corresponding frequency responses using fast Fourier transforms;

   • establishing a mean M of the N frequency responses;

   • transposing said mean M of the N frequency responses onto a logarithmic frequency scale;

   • interpolating a target sound signature curve C_ec from a specific number of points defined by a user;

   • transposing said target sound signature curve C_ec onto a logarithmic frequency scale;

   • comparing said mean response M and said target response C_ec by calculating the offset between said mean response M and said target sound signature response C_ec;

   • analyzing the curve C_diff resulting from the offset between said mean response M and said target response C_ec; and characterized in that it includes means for:

   • determining filter parameters for reducing the offset between said mean response M and said target response C_ec by firstly processing the local maxima in descending order according to the gain thereof, then the local minima, and by carrying out successive iterations;

   said system further including means for optimization of the parameters of the filters in order to improve the performance of the system.

Images