Method for optimising the frequency response of at least two loudspeakers placed in a specific environment

Abstract

The invention relates to a method for optimising the frequency response at a listening point (P) of at least two loudspeakers (23, 24) placed in a specific environment (20), the method comprising: – a step of centring the sound scene, in which step the two loudspeakers are configured to simultaneously emit sound signals; – a tonal correction step in which the two loudspeakers are configured to simultaneously emit sound signals played sequentially in pairs with a frequency distance less than or equal to one octave; and – a step of calculating corrections to be applied to the loudspeakers (23, 24) to optimise the frequency response of the loudspeakers depending on the modifications made in the step of centring the sound scene and in the tonal correction step.

Description

[0001] Description

[0002] Title of the invention: METHOD FOR OPTIMIZING THE FREQUENCY RESPONSE OF AT LEAST TWO SPEAKERS PLACED IN A SPECIFIC ENVIRONMENT

[0003] technical field

[0004] The technical field of the present invention relates to the correction of the frequency response of a speaker placed in a specific environment and, more specifically, a method for optimizing the frequency response at a listening point of at least two speakers placed in said environment.

[0005] This technique is particularly applicable to improving the listening experience in various environments such as home living rooms, recording studios, or vehicle interiors. It aims to correct undesirable effects caused by the acoustic characteristics of the environment, such as asymmetry, resonances due to acoustic modes, and the proximity effects between speakers and the walls of a room or vehicle, which can impair the fidelity of sound reproduction.

[0006] Previous art

[0007] The frequency response of a room is influenced by many factors, including the placement of the speakers within the listening environment. Indeed, the proximity of the speakers to the walls, ceiling, and floor can significantly alter the frequency response, particularly in the low frequencies, due to acoustic reflections that induce standing waves at the room's resonant frequencies. These reflections lead to both constructive and destructive interference. These interactions between the sound waves emitted by the speakers and the room surfaces can result in localized increases or decreases in the level of certain frequencies, thus creating peaks and dips in the frequency response perceived by the listener.These effects are often undesirable because they can alter the fidelity of sound reproduction and affect the tonal balance as well as the clarity of the soundstage.

[0008] To estimate the frequency response of a room, several dimensional analysis methods have been developed. One of the earliest methods is that described by Roy F. Allison in "The Influence of Room Boundaries on Loudspeaker Power Output," published in JAES Volume 22 Issue 5, pp. 314-320, in June 1974. This method uses an analytical expression to estimate, among other things, the increase in low-frequency level due to the proximity of walls and the floor. While this approach is useful for obtaining a quick estimate, it can be limited by its generalization and may not take into account the specific characteristics of each room.

[0009] The Source Image method is a technique used to model the acoustic response of a room. It involves simulating virtual sound sources (images) reflected by the room's surfaces to predict the sound field within the room. This method can provide accurate results, but it can become complex and computationally expensive as the number of reflections to consider increases, and it is limited to simple geometries, typically rectangular rooms.

[0010] Ray tracing, also known as ray tracing in English-language literature, is another acoustic modeling technique that simulates the path of sound waves in a room by tracing the path of the rays from the source to the listening position. This method takes into account multiple reflections from the room's surfaces and can provide a detailed representation of the spatial distribution of sound pressure. However, this method is often difficult to implement because the computation time is relatively long, especially for rooms with complex geometries, and it requires the intervention of a professional / acoustician to perform the simulation. Indeed, the user of this method must be trained to define the room's geometry and the acoustic absorption and diffusion properties of the materials.

[0011] FEM, for "Finite Element Method" in the Anglo-Saxon literature, and BEM, for "Boundary Element Method" in the Anglo-Saxon literature, are numerical modeling techniques that allow a detailed analysis of the acoustic response taking into account the complete geometry of the room and the properties of the materials.

[0012] These methods are very accurate but still require significant computing power and are often reserved for professional applications due to their complexity and the time required to perform the simulations.

[0013] In conclusion, these dimensional analysis methods are often complex to implement, and the actual room surfaces are generally simplified to limit the computation time for room simulation. Consequently, these dimensional analysis methods typically provide inaccurate results.

[0014] To improve measurement accuracy, various microphone-based measurement methods are also known, as described in patent application WO 9210876. These methods use microphones placed at different locations in the room to capture the acoustic response to specific sound signals emitted by the loudspeakers. Among the most common techniques is impulse response measurement, which determines how the emitted sound is modified by the room environment. This measurement can be performed using a sliding sinusoidal signal or impulse noise, such as Dirac noise.

[0015] Another commonly used method is near-field measurement, where the microphone is placed in the immediate vicinity of the loudspeaker to minimize the influence of the room and obtain a more accurate measurement of the loudspeaker's direct response. This measurement can then be combined with modeling to extrapolate the far-field response.

[0016] Automatic calibration systems with built-in microphones are also popular in consumer audio equipment. These systems emit test signals and use the pressure responses captured by the microphones to automatically adjust speaker parameters, such as volume levels, equalization, and delay, to optimize the sound response for the specific acoustics of the room.

[0017] These measurement methods using microphones are generally more accurate than dimensional analysis methods, but they require specialized equipment that can be expensive. Furthermore, using a microphone to measure acoustic response has a major drawback: due to the presence of standing waves at the room's natural frequencies, the pressure level varies significantly throughout the space, with areas where the pressure level is very strongly attenuated, or even zero. To overcome this phenomenon, it is possible to calculate an average of the frequency response from several measurement points.

[0018] Therefore, a measurement at a single point in the room can overestimate the importance of these modes, degrading the response near the microphone's listening position. This means the measurement may not be representative of the actual listening experience in different parts of the room, potentially leading to acoustic correction that is not optimal for the entire listening space. Alternatively, using multiple measurement points increases the time required for these microphone measurement methods.

[0019] Furthermore, these measurements with microphones do not take into account the subjective auditory perception of the listener, which can vary depending on many factors, including listening position and individual characteristics of the listener.

[0020] Furthermore, it is possible to specifically consider the listener's acoustic perception to refine the acoustic adaptation. As described in patent EP3614380, psychoacoustic tests can be performed to assess the listener's auditory perception abilities in a specific room, typically an anechoic chamber. These tests allow for the determination of the individual's perception of frequencies and sound levels.

[0021] Similarly, in order to adapt the audio processing to the listener's hearing, it is possible to take into account their perception of the location of a sound source by using HRTF filters, for "Head-Related Transfer Function" in the Anglo-Saxon literature.

[0022] These approaches can significantly improve the listening experience by personalizing acoustic correction based on the listener's auditory responses.

[0023] However, these approaches aim to take into account the listener's perception in order to correct it, for example, in cases of age-related high-frequency hearing loss, but they do not allow for correction of the specific characteristics of the room, since tests are typically conducted in an anechoic chamber or with headphones. Furthermore, this procedure is complex to implement because the listener must have access to a non-reverberant environment, typically an anechoic chamber or high-quality headphones, and undergo several listening tests, which can be lengthy.

[0024] To account for room characteristics and individual listener perception, US2017373656 recommends simultaneously measuring the listener's hearing capabilities in an ideal room and measuring the room's acoustic response with a microphone. This dual measurement allows for optimizing loudspeaker response by integrating both the listener's hearing abilities and the room's acoustic response.

[0025] However, this solution combines the disadvantages of measurement solutions using microphones, risk of overestimating the importance of modes at the listening point and need for a specific device, and the disadvantages of a prior measurement of a listener's hearing abilities in a non-reverberant environment.

[0026] The technical problem of the invention is therefore to determine how to optimize the frequency response, at a listening point, of at least two speakers placed in a specific environment, with simple measures to implement and taking into consideration the hearing capabilities of a listener.

[0027] Description of the invention

[0028] To address this technical problem, the invention proposes to generate specific stimuli from the speakers of the enclosures in a specific environment so that the listener can define the corrections to be applied to the speakers of the enclosures to improve the frequency response of the environment, while taking into consideration the hearing capabilities of the listener.

[0029] To do this, two distinct steps are implemented: a first step of centering the sound scene, and a second step of tonal correction.

[0030] Thus, the invention relates to a method for optimizing the frequency response, at a listening position, of at least two loudspeakers placed in a specific environment. The method comprises:

[0031] - a soundstage centering step in which:

[0032] The two speakers are configured to simultaneously emit sound signals transmitted to the input of each speaker; and

[0033] . for each sound signal, a listener provides an assessment of the sound balance between the two speakers by possibly modifying the gain of one of the speakers in order to position the virtual source, generated by the combination of the speakers, relative to the listening point;

[0034] - a tonal correction step, performed after the soundstage centering step, in which:

[0035] The two speakers are configured to simultaneously emit sound signals transmitted to the input of each speaker, the sound signals being played sequentially in pairs with a frequency difference of less than or equal to one octave; and

[0036] For each pair of sound signals, a listener provides an assessment of the sound intensity between the two signals by indicating a change so as to perceive the same sound intensity between the two signals; and

[0037] - a step of calculating the corrections to be applied to the speakers to optimize the frequency response of the speakers according to the modifications made in the sound scene centering step and the different signals made in the tonal correction step.

[0038] For the purposes of this invention, the position of the "virtual source" corresponds to the position in the listening space from which the sound appears to originate for the listener positioned at the listening point. Indeed, with at least two speakers, it is possible to adjust the gain and / or the delay between the sound signals transmitted to the speakers to virtually shift the perceived origin of the sound for the user.

[0039] The soundstage centering step minimizes the interaural level difference, also known as ILD. This interaural level difference is a measure of the perceived difference in sound intensity between a listener's two ears. This difference is due to the acoustic characteristics of the environment, such as room asymmetry and the listener's position relative to the speakers, resonances caused by acoustic modes, and the proximity effects between the speakers and the walls of a room or vehicle, as well as the acoustic shadow effect of the head, which partially blocks sounds arriving from certain directions, thus reducing their intensity for the ear opposite the sound source.

[0040] Tonal correction refers to adjustments made to compensate for variations in the perceived loudness of sounds based on their frequency. When sounds reach the listener's ears, their intensity and perceived loudness can vary due to the acoustic characteristics of the environment and the effects of the head, ears, and body on sound waves. These effects can alter the levels of some frequencies more than others, potentially leading to distortion in sound perception. Preferably, this step is implemented with sound signals played sequentially with a frequency interval of half an octave.

[0041] The invention therefore makes it possible to optimize the frequency response of speakers according to the acoustic characteristics of the environment and the acoustic perception of the listener.

[0042] To apply the corrections, the input signal of each speaker is preferably filtered using FIR filters or HR filters, preferably 2nd order HR filters, such as BIQUAD filters.

[0043] Alternatively, it is possible to perform analog filtering based on the corrections to be applied calculated to optimize the frequency response of the speakers, for example from analog BIQUADs, multi-feedback cells, Sallen & Key cells or even gyrators.

[0044] Furthermore, the corrections applied to loudspeakers to optimize their frequency response can be applied to both the dips and peaks of the room's frequency response. Preferably, the amplified room frequencies, i.e., the peaks, are treated with a limiting factor, such that the total amplitude of the correction resulting from the sum of the various filters never falls below a limit, for example, -10dB. This limitation can potentially be asymmetrical, with the correction of the attenuated room frequencies, i.e., the dips, using an amplification that does not exceed a high limit, for example, +6dB, in order to limit the excursion of the loudspeaker diaphragms and avoid potential distortion.Thus, contrary to the teachings of patent US2017373656, which proposes only addressing the "dips" in the frequency response, the invention can be used in this embodiment to improve the frequency response by also addressing the "peaks." This embodiment is advantageous because the ear is more sensitive to "peaks" than to "dips," and it is simpler to remove energy than to add it. In a preferred embodiment, the correction applied has a zero mean, so the correction does not alter the total energy.

[0045] Furthermore, with regard to the sound signals used in the sound scene centering stage and in the tonal correction stage, they preferentially follow the following equation:

[0046] [Math 1] with:

[0047] N corresponding to the total number of sinusoids contained in the frequency band; n corresponding to the index of the sinusoid contained in the frequency band; an corresponding to an amplitude factor of the sinusoid with index n; fn corresponding to the frequency of the sinusoid with index n;

[0048] <f> n corresponds to the angle at the origin of the sinusoid with index n;

[0049] Tn corresponds to the damping time constant of the sinusoid with index n; and

[0050] W(t) corresponding to a weighting window.

[0051] In a preferred embodiment, the total number N of sinusoids contained in each band is equal to 5. This value makes it possible to limit the number of sinusoids to be processed in real time while offering sufficient accuracy for optimizing the frequency response.

[0052] If the sound level at the listening position (or its order of magnitude) is known, it is recommended to adjust the amplitude of each sinusoid to follow the ISO loudness perception curve or the one defined by Fletcher & Munson. If the sound level at the listening position is unknown, it is recommended to refer to the ISO 80 dB SPL curve. This standardized curve provides a reliable reference for adjusting sound levels.

[0053] The frequencies of the sinusoids, fn, are preferentially spaced logarithmically on either side of the central frequency fo, with a step between two consecutive frequencies equal to:

[0054] [Math 2] fn J fn-1 where:

[0055] 1 / K is the width of each band expressed in octaves;

[0056] N is the number of frequencies in each band and / ? is a factor preferably between 1 and 1.5.

[0057]

[0001] By preferentially choosing 1 / K = , N = 5 and [3 = 1.33, the five frequencies of the central frequency band fo are respectively: fo / 1.2 for the first; fo / 1.1 for the second; fo for the third; fo 1.1 for the fourth; and fo x 1.2 for the fifth.

[0058] This frequency distribution allows a wide range of frequencies to be covered while maintaining consistency with the center frequency.

[0059] The original angle, <p n , is preferably zero. This simplification reduces the complexity of the calculations without significantly affecting the accuracy of the optimization.

[0060] The damping of the sound signal, Tn, is preferably constant. For the sound scene centering stage, T n The time interval (Tn) can be between 50 and 200 ms, for example, 100 ms. For the pitch correction stage, Tn is preferably infinite. These values ​​allow for precise adjustment of the temporal characteristics of the audio signals. Finally, the weighting window, W(t), is a volume increase / decrease function, also called "fade-in / fade-out" in English-language literature, with a value of zero at the beginning and zero at the end of the signal. This function smooths the transitions of the audio signals, thus avoiding discontinuities that could generate unwanted noise.

[0061] In addition to the soundstage centering and tonal correction stages, it is also advantageous to modify the frequency response using a dimensional adaptation stage that includes:

[0062] - a sub-step of retrieving the dimensions of the environment and the position of the speakers;

[0063] - a substep for estimating the frequency response of the environment based on the dimensions of the environment and the position of the loudspeakers; and

[0064] - a sub-step of possible modification of the correction to be applied to the speakers in order to compensate for the increases or decreases of certain frequencies estimated in the frequency response of the environment, carried out before the tonal correction step.

[0065] Thus, the dimensional adaptation step is performed before the tonal correction step, either before or after the soundstage centering step. Preferably, this dimensional adaptation step is performed before the soundstage centering step.

[0066] According to one embodiment, said step of calculating the frequency response of the speakers is carried out according to the different modifications applied in the steps of centering the sound scene and tonal correction and by anchoring the corrections of the lowest and highest frequency bands on the frequency response determined in said dimensional adaptation step.

[0067] Furthermore, the said dimensional adaptation step can be carried out analytically, by the ray tracing method, the image source method, the FEM method or the BEM method.

[0068] The substep of retrieving the dimensions of the environment and the position of the loudspeakers can be carried out using an interface where the listener enters measured distance values. More generally, the soundstage centering and tonal correction steps can also be implemented via a mobile application, on a phone or tablet, or as computer software, so as to easily collect listener feedback.

[0069] Brief description of the drawings

[0070] The method of implementing the invention, as well as the resulting advantages, will become clear from the following embodiments, given by way of example but not limitation, supported by the attached figures in which:

[0071] Figure 1 is a perspective view of an auditor implementing the optimization method according to one embodiment of the invention;

[0072] Figure 2 is a schematic representation of the optimization method according to a first embodiment of the invention.

[0073] Figure 3 is a schematic representation of the optimization method according to a second embodiment of the invention;

[0074] Figure 4 is a schematic representation of the optimization method according to a third embodiment of the invention;

[0075] Figure 5 is a schematic representation of the dimensional adaptation step according to one embodiment of the invention;

[0076] Figure 6 is a frequency representation of the sound response of a room at a listening point for two sound signals from the two speakers and the associated average response;

[0077] Figure 7 is a frequency representation of the preliminary correction curve of the sound response of a room, with and without limitation;

[0078] Figure 8 is a frequency representation of a regularization factor for optimizing the sound response of a room;

[0079] Figure 9 is a frequency representation of two correction curves for optimizing the sound response of a room, with and without regularization;

[0080] Figure 10 is a frequency representation of two correction curves for optimizing the sound response of a room, a regularized theoretical curve and a curve representing the equivalent implementation using a filter;

[0081] Figure 11 is a top view of a listener implementing the sound scene centering step at a first listening point according to one embodiment of the invention;

[0082] Figure 12 is a top view of a listener implementing the step of centering the sound scene at a second listening point according to an embodiment of the invention;

[0083] Figure 13 is a schematic representation of the sound scene centering step according to one embodiment of the invention;

[0084] Figure 14 is a schematic representation of five weighted frequency bands in the sound scene centering stage according to one embodiment of the invention;

[0085] Figure 15 is a schematic representation of five limited frequency bands in the soundstage centering stage according to one embodiment of the invention; and

[0086] Figure 16 is a schematic representation of the tone correction step according to one embodiment of the invention.

[0087] Detailed description of the invention

[0088] Figure 1 illustrates a listener (21) using a mobile phone (22) to adjust two speakers (23, 24) in a specific environment (20), typically a room represented by a room with walls. Two speakers, the left speaker

[0089] (23) and the right speaker (24) are placed in this room. The left speaker (23) has a tweeter (25) and a woofer (26), and the right speaker

[0090] (24) also includes a tweeter (27) and a woofer (28). Of course, the invention can be implemented with more than two speakers and a variable number of speakers per speaker.

[0091] The listener's perceptual characteristics, as well as the geometry of the room and the position of the speakers (23, 24), may cause the listener (21) to not hear certain sounds generated by the speakers (23, 24) correctly at the listening point (P) due to an alteration of the frequency response of the speaker at the listening point due to the environment and the listener's perceptual characteristics.

[0092] The invention proposes to optimize the frequency response at the listening point (P) using one of the optimization methods illustrated in Figures 2 to 4. In a first, optional step, shown only in Figures 3 and 4, a dimensional adaptation step (30) is implemented. In this dimensional adaptation step (30), the listener (21) is asked to provide at a minimum the position of the loudspeakers (25-28) and the dimensions of the environment (20), and optionally their position (P) in the room via the mobile phone interface (22). This information can be obtained by measurements or by specifying the speaker model in order to extract technical information, such as the distance between the bass speakers (26, 28) and the room floor.To achieve this, the graphical interface allows the listener (21) to select speakers from a list of predefined speakers in order to directly retrieve their dimensions. Furthermore, the user can be guided by graphical interfaces containing explanations for measuring and entering the speaker positions. The distances (d1) and (d2) illustrate the offsets of the left speaker relative to the back wall and the side wall. Similarly, the offsets of the right speaker (24) relative to the back wall and the side wall are obtained from measurements (d3) and (d4).

[0093] Thus, by retrieving the dimensions of the environment (20) and the position of the loudspeakers (25-28), the dimensional adaptation step (30) can implement sub-steps (310-317) illustrated in Figure 5. The first sub-step (310) allows the technical information described previously to be entered.

[0094] Next, a substep (311) determines the theoretical frequency response for the left speaker (Lb) and the frequency response for the right speaker (Rb). Furthermore, an optional substep (312) calculates the average frequency response (Ab) by averaging the two frequency responses, allowing the same correction to be applied to both speakers (23-24). These frequency responses from substeps (311-312) are illustrated, for example, in Figure 6.

[0095] This substep of estimating the frequency response of the environment can be calculated analytically, by the ray tracing method, the image source method, the FEM method or the BEM method.

[0096] Following the determination of the average frequency response (Ab), or simply the responses (Lb, Rb), it is possible to modify the sounds generated by the speakers to compensate for the dips and peaks detected in the average frequency response (Ab). To do this, the inverse frequency response curve can be directly calculated in a substep (313). This original correction curve (Ori) is illustrated in Figure 7. However, to avoid saturation problems, it is preferable to limit certain corrections, particularly in amplification, as this risks saturating the speakers or the digital or analog components associated with them. To achieve this, the frequencies of the room to be amplified can be treated with a limiting coefficient in a substep (314), in order to obtain the limited correction curve (Lim), illustrated in Figure 7.

[0097] In addition to this amplitude limitation, it is also possible to apply a frequency-dependent regularization factor (a), for example, with a greater gain at low frequencies than at high frequencies. This regularization factor (a) is calculated, for example, in substep (315) and illustrated in Figure 8. Figure 9 illustrates the regularized correction curve (Reg) following the application of this regularization factor (a) to the limited correction curve (Lim) in substep (316).

[0098] In this substep (316), this regularized correction curve (Reg) can be adjusted to prevent the corrections from being too locally focused in frequency, which can lead to a sound often described as "artificial," and also to limit the number of correction filters. Furthermore, each dip and peak must conventionally be addressed by a filter, which is costly in terms of digital resources or components when the filters are analog. For example, by using at least one LowShelf filter and a least-squares method to define the filter's amplitude, frequency, and quality factor, it is possible to obtain a dimensional matching correction curve (Cor) from the regularized correction curve (Reg), as illustrated in Figure 10. Of course, other methods alternative to the least-squares method can be used to converge on the filtering parameters.

[0099] This dimensional adaptation correction (Cor) curve can be applied to the loudspeakers in a substep (317) by modifying the audio signal transmitted to the loudspeakers (23, 24). Preferably, each loudspeaker (25-28) is associated with a filter, either digital or analog, to apply the determined correction. Preferably, for a given channel, the filter is configured to apply the same correction to all loudspeakers (25-28) in the same loudspeaker (23-24). Alternatively, it is also possible to apply a filter to each loudspeaker (25-28).

[0100] With or without the dimensional adaptation correction (Cor), the invention proposes to calculate a correction taking into account the actual perception of the listener (21) at the listening point (P). To do this, the optimization method incorporates a sound scene centering step (31).

[0101] Soundstage centering describes a process that involves placing a virtual sound source (XX1) in space by modifying the gains and / or delays applied to the speakers (23-24). More specifically, as illustrated in Figures 11 and 12, a sound is considered centered when the virtual source (XX1), generated simultaneously by the left (23) and right (24) speakers, is perceived on the axis (XX2) in front of the user (21). In the present invention, the user is preferably located between the two speakers (23) and (24).

[0102] As illustrated in Figure 13, this process begins with step (40) where the frequency band with index s is fixed at s=Smax. Of course, it is possible to start with a completely different frequency band as long as the test is carried out on all bands.

[0103] Next, the initialization step (41) creates a table of provisional gains in dB corresponding, in the case of a pair of stereo speakers, to a matrix of dimensions [2xSmax] initially with a value of zero. For example, if 7 bands are tested, the table then takes the initial form of:

[0104] [Math 3]

[0105] / Gain_Lb\ / O 0 0 0 0 0 0\ \Gain_Rb) \0 0 0 0 0 0 0 / '

[0106] In step (42), the index band stimulus was reproduced simultaneously by both speakers with initially the same level.

[0107] In step (43), the listener (21) checks the perceived centering level of the virtual source (XX1). If the virtual source (XX1) is not located on the axis (XX2), the listener (21) gives the information (44) to the application (22) specifying on which side the image of the virtual source (XX1) is off-center, which leads to a change in the gain (Gain_Lb) of the left speaker, in step (45), or in the gain (Gain_Rb) of the right speaker, in step (46), modifying the initial gain matrix (Gain_Lb, Gain_Rb).

[0108] When the listener (21) has checked the centering level for a frequency, a test is performed to verify whether all the expected frequency bands s have been tested, in step (47). If not, the index band s is reduced by 1, in step (48), and the process is repeated for the next band.

[0109] Once all frequencies have been tested, the applied gains are weighted in step (49) and the applied corrections are limited in amplitude in step (50). The weighting (I3>) can be constant or frequency-dependent according to the following equation:

[0110] [Math 4]

[0111] In the example in Figure 14, five frequency bands are shown on the curve (Ori) with drifts observed for the s=2 and s=3 bands. The frequency bands after weighting are shown on the curve (Reg) and reveal that the weighting has decreased the amplitude of the reported displacements.

[0112] Figure 15 illustrates the influence of the amplitude limiting factor (Lim) on the frequency bands shown in Figure 14. After limiting, on the bands shown on the curve (RegLim), no amplitude crosses the lower limiting threshold (Lim). Of course, the limiting threshold (Lim) can also correspond to a positive amplitude threshold.

[0113] At the end of this sound scene centering step (31), the gains (Gain_Lb, Gain_Rb) are therefore obtained and stored (51), possibly with the corrections from steps (49) and (50), to center the virtual source (XX1).

[0114] Following the soundstage centering step (31), the optimization method incorporates a tonal correction step (32), as illustrated in Figure 16. The process begins with step (60) where the index band s is fixed at Smax. Of course, it is possible to start with a completely different frequency band as long as the test is performed on all bands. The initialization step (61) then creates a table of provisional gains to be applied to both speakers simultaneously, corresponding to a vector of dimensions [IxSmax] initially with a value of zero. For example, if 7 bands are tested, the table then takes the initial form of:

[0115] [Math 5]

[0116] Gain G = [00 0 0 00 0] dB

[0117] Next, at least two signals emitted sequentially from successive frequency bands are transmitted simultaneously to the left (23) and right (24) speakers during step (62). Depending on the frequency band emitted, the level of each speaker or each signal is individually pre-modified according to the gain table defined during the soundstage centering step (31).

[0118] In step (63), the listener checks, through a dedicated interface of the application (22), whether the perceived sound intensity between at least two successive bands is close or identical.

[0119] The pitch correction stage can be implemented with sound signals played sequentially, preferably spaced at a frequency distance equal to half an octave.

[0120] If an imbalance in sound level or intensity is detected (64), a gain adjustment is made either when the first signal (65) is played or when the second signal (66) is played. This gain modification is saved in the initial gain table (Gain_G) at the index of the modified band.

[0121] After the user has verified and confirmed that the sound level of the two bands is similar or equal (63), a test is performed to check if all the expected bands have been tested in step (67). If not, the band with index s is reduced by 1 in step (68) and the process begins again.

[0122] Once all frequency bands have been tested, the gains (Gain_G) are "reconstructed" in pairs in step (69).

[0123] The construction of these gains is, for example, carried out from the highest frequency band to the lowest frequency band according to the following formula:

[0124] [Math 6]

[0125] Alternatively, it is also possible to construct the gains from the lowest frequency band to the highest frequency band using the following formula:

[0126] [Math 7]

[0127] In another variant, it is also possible to construct the pairwise gains from a reference band S according to the formula:

[0128] [Math 8]

[0129] The gains in dB are then weighted with a factor of 5, ranging from 0.1 to 1, in step (70). An offset is preferably applied to all the gains so that the correction results in zero average energy, in step (71). It should be noted that this step (71) is optional, as it is not necessary for someone with very good critical hearing.

[0130] Finally, in a step (72), the gains (Gain_G) are "anchored" to the correction from the dimensional matching step (30) at the lowest and highest frequency bands. Anchoring consists of setting the lowest and highest gains to the values ​​determined during the dimensional matching step (30), and then recalculating the intermediate gains based on these extreme values.

[0131] In practical terms, this means that the first and last gains are fixed relative to those obtained in the dimensional adaptation step (30), while the gains between these two extremes are adjusted relative to the difference between the gains (Gain_G) taken two at a time. The intermediate gains can be recalculated either by linear interpolation, by uniformly distributing the total difference between the extreme values, or by non-linear adjustment methods such as quadratic interpolation. Finally, the corrections are limited (73) with one of the maximum positive and negative gain thresholds, for example -10 and +10 dB.

[0132] At the end of this tonal correction step (32), the gains (Gain_G) are therefore obtained and stored (74), possibly with the corrections from steps (70) to (73).

[0133] As illustrated in Figures 2 to 4, the soundstage centering (31) and tonal correction (32) steps are used to feed a calculation step (33) of the corrections to be applied to the speakers (23, 24).

[0134] In this calculation step (33), the final corrections are defined by merging the gain tables / matrices (Gain_Lb, Gain_Rb, and Gain_G). Thus, the final gain (G_L) to be applied to the left speaker (23) and the final gain (G_R) to be applied to the right speaker (24) can be calculated according to the equation:

[0135] [Math 9]

[0136] Finally, in the application step (34) of the corrections calculated in the calculation step (33) of the corrections to be applied to the speakers (23, 24), a synthesis of the correction response is performed to convert the gain table, containing discrete values, into implementable filters. The synthesis is done using a series of filters, for example FIR or HR type filters, preferably second-order HR filters, such as BIQUAD filters, associated with each loudspeaker.

[0137] In some embodiments, the application step (34) uses at least one LowShelf type filter or one Peak EQ type filter on the calculated corrections.

[0138] The invention optimizes the frequency response of the speakers (23, 24) according to the dimensions of the environment, the positions of the loudspeakers, and the listening position. It compensates for increases or decreases in the levels of certain frequencies estimated in the frequency response of the environment and the listener's acoustic perception.

[0139] The main advantages of this invention are the improvement in sound quality perceived by the listener and the precise adaptation of speakers to the specific environment. It solves problems of sound imbalance and tonal variations, thus offering an optimal listening experience.< / f>

Claims

Demands 1. A method for optimizing the frequency response at a listening point (P) of at least two loudspeakers (23, 24) placed in a specific environment (20), said method comprising: - a soundstage centering step (31) in which: . the two speakers (23, 24) are configured to simultaneously emit sound signals transmitted to the input of each speaker (23, 24); and . for each sound signal, a listener (21) provides an appreciation of the sound balance between the two speakers (23, 24) by possibly modifying the gain of one of the speakers (23, 24) so ​​as to position the virtual source (XX1), generated by the association of the speakers (23, 24), relative to the listening point (P); - a tonal correction step (32), carried out after the sound scene centering step (31), in which: The two speakers (23, 24) are configured to simultaneously emit sound signals transmitted to the input of each speaker (23, 24), the sound signals being played sequentially two at a time with a frequency difference of less than or equal to one octave; and For each pair of sound signals, a listener (21) provides an assessment of the sound intensity between the two signals by reporting a change so as to perceive the same sound intensity between the two signals; and - a calculation step (33) of the corrections to be applied to the speakers (23, 24) to optimize the frequency response of the speakers (23, 24) according to the modifications made in the sound scene centering step (31) and the different signals made in the tonal correction step (32).

2. A frequency response optimization method according to claim 1, wherein the method also includes a dimensional adaptation step (30) comprising: - a sub-step of retrieving the dimensions of the environment (20) and the position of the speakers (25-28); - a substep for estimating the frequency response of the environment (20) depending on the dimensions of the environment (20) and the position of the loudspeakers (25-28); and - a sub-step of modifying the correction to be applied to the speakers (23, 24) so ​​as to compensate for the increases or decreases of certain frequencies estimated in the frequency response of the environment (20), carried out before the tonal correction step (32).

3. Frequency response optimization method according to claim 2, wherein said step of calculating the corrections to be applied to the speakers (23, 24) to optimize the frequency response of the speakers (23, 24) is carried out as a function of the different modifications applied in the sound scene centering (31) and tonal correction (32) steps and by anchoring the corrections of the lowest and highest frequency bands on the frequency response determined in said dimensional adaptation step (30).

4. Frequency response optimization method according to claim 2 or 3, wherein said dimensional adaptation step (30) is carried out by the ray tracing method, the image source method, the FEM method or the BEM method.

5. Frequency response optimization method according to any one of claims 1 to 4, wherein the method also includes an application step (34) of the corrections calculated in the calculation step (33) of the corrections to be applied to the speakers (23, 24) to optimize the frequency response of the speakers (23, 24).

6. Frequency response optimization method according to claim 5, wherein the application step (34) uses at least one LowShelf type filter or one Peak EQ type filter on the calculated corrections.

7. Frequency response optimization method according to claim 5 or 6, wherein the loudspeakers are associated with FIR filters or HR filters, preferably 2nd order HR filters, such as BIQUAD filters, so as to apply the calculated corrections.

8. A method for optimizing the frequency response according to any one of claims 1 to 7, wherein the sound signals used in the sound scene centering step (31) and in the tonal correction step (32) follow the following equation: [Math 1] with: N corresponding to the total number of sinusoids contained in the frequency band; n corresponding to the index of the sinusoid contained in the frequency band; an corresponding to an amplitude factor of the sinusoid with index n; fn corresponding to the frequency of the sinusoid with index n; n corresponds to the angle at the origin of the sinusoid with index n; Tn corresponding to the damping time constant of the sinusoid with index n; and W(t) corresponding to a weighting window.

9. Frequency response optimization method according to any one of claims 1 to 8, wherein the pitch correction step is implemented with audio signals played sequentially with a frequency gap equal to half an octave, j

Images