FR3087599A1 - METHOD FOR CLASSIFYING THE NOISE LEVEL OF AN AUDIO SIGNAL - Google Patents

Abstract

There is disclosed a method of classifying the noise level of an audio signal for testing the quality of radio reception of a motor vehicle audio system under test. The method comprises the following steps, performed by a processing unit: receiving an audio signal called a test audio signal (207); generating from the test audio signal a feature vector (208) having a determined number of components corresponding to respective determined characteristics of the test audio signal; and, determining, based on the components of the characteristic vector of the test audio signal and further based on a prediction model (206) associating given components of characteristic vectors with given labels characterizing the level of noise of a corresponding audio signal, of a label (209) characterizing the noise level of the test audio signal.

Description

Description Title of the invention: Level classification method. of noise of an audio signal Technical field [00011 The present invention relates generally to the performance test of the radio reception system of a motor vehicle, and more particularly to the classification of the noise level of a captured audio signal. by the radio reception system.

PRIOR ART [00021 Motor vehicles are generally equipped with radio reception systems, for the entertainment or information of the driver and passengers (for example for listening to music or information stations), and / or for reception. infomaation of. services (eg traffic information), andiou for vehicle communication with fixed or mobile equipment.

[00031 These systems make it possible to capture radiofrequency signals emitted by surrounding sources, and which carry information intended to be received by the driver and any passengers of the vehicle.

These are for example FM radio type signals which are then converted into audio signals to be transmitted to on-board speakers of the vehicle, which in turn retransmit an audio signal perceptible by the driver or the passengers.

[00041 In addition to the quality. intrinsic to the components of such systems, their performance can be linked to various factors: layout of the components within the vehicle, vehicle running speed, use environment (urban or peri-urban environment, mountain area, for example), etc.

This is why it is essential to be able to test the performance of these systems and in particular to be able to test them in real operating conditions.

Indeed, ultimately, the quality of the audio signal perceived by the driver is a direct reflection of the level of performance of the radio reception system when the vehicle is used in real conditions.

[00051 Document EP 1093245 discloses a method for testing the quality of the components of a radio reception system of a vehicle.

This method is based on the use of two test vehicles each comprising a component to be tested and its own recording device.

The two vehicles travel on the same road and acquire signals during their journey which are compared to deduce therefrom the level of performance of the system under test, [0006; In general, the known approaches aimed at testing the performance of a radio reception system in a prototype vehicle involve several operators, or testers.

These people sit in several vehicles, and then listen, analyze and compare the quality of an audio signal received in a vehicle under test with that of a signal received in a reference vehicle which is traveling at the same time on the same portion of the vehicle. road.

The performances are therefore established on the basis of the subjective perceptions of the testers located in the vehicle under test, which are evaluated and compared in real time with the subjective perceptions of the testers located in the reference vehicle, all these testers communicating with each other in real time, for example by Walkie-Talkie or by mobile phone.

[0007] Such an approach has significant limitations.

First of all, it assumes a great homogeneity of the respective perception of the audio quality by the different testers.

The relevance of the comparison may otherwise be insignificant.

In addition, there is a need to train personnel with the necessary technical skills to analyze such signals.

Such training can be complex and costly.

In addition, the resources mobilized are significant.

Typically, for testing an FM radio antenna, two vehicles are used and each must drive about 75 km for about an hour to get relevant results.

Finally, such tests are only carried out in real time.

No recording of the tested signals is made.

Therefore, the tests are not reproducible.

They are only carried out once.

If they are repeated, for example after having modified the implantation of a particular component of the reception chain in the vehicle, there is no guarantee that the same test conditions can be found to evaluate the improvement (or the deterioration) of the situation with relevance and objectivity.

Summary of the invention [00 The invention aims to eliminate, or at least mitigate, all or part of the aforementioned drawbacks of the prior art, 1.00099 To this end, a first aspect of the invention proposes a classification method of the noise level of an audio signal for testing the quality of radio reception of a vehicle audio system under test, said method comprising the following steps, carried out by a processing unit [001011 for the reception of a signal audio test; generating from the test audio signal a characteristic vector having a determined number of components corresponding to respective determined characteristics of the test audio signal; and, determining, based on the feature vector components of the test audio signal and further based on a prediction model associating given feature vector components with labels characterizing the noise level of a corresponding audio signal, a label characterizing the noise level of the test audio signal. ) 011] Grace. according to the invention, it is possible to test the level of performance of a radio reception system of a motor vehicle by classifying the noise level of audit signals received by the vehicle concerned.

In particular, the method according to the invention enables the recording of signals and the reproducibility of the tests, limits the resources necessary for carrying out the tests and does not involve complex training of personnel intended to carry out the tests.

[00121 Embodiments taken alone or in combination further provide that

[001] - the method further comprises, clans a prior learning phase

Receiving at least one pre-recorded audio signal, associated with a given label characterizing a determined noise level of said recorded audio signal; generating from each prerecorded audio signal a characteristic vector with a determined number of components corresponding to determined characteristics of the prerecorded audio signal; and, the use of each feature vector and its associated label as input data of a machine learning algorithm, and the creation, by the machine learning algorithm, of a prediction model associating given components of feature vectors with given labels characterizing a determined noise level.

- the test audio signal is obtained from a radiofrequency signal. effectively received by an antenna of the motor vehicle and effectively converted into an audio signal by the tuner of said motor vehicle, and supplied to the processing unit; [001.6 - the test audio signal which is obtained after conversion of the radiofrequency signal by the tuner of the motor vehicle, is stored in a memory and transmitted to the processing unit after the expiration of a determined period; [00171 - each audio signal has a determined duration equal to a few seconds, in particular between 3 and 8 seconds, for example equal to 5 seconds; [) 0181 conversion of an audio signal, received by the processing unit, into a characteristic vector with a determined number of components corresponding to determined characteristics of the audio signal, is carried out according to a signal processing method audio consisting in calculating a cepstrum of type “tLfelfrequency” from the audio signal, and in which the components of the characteristic vector correspond to the calculated cepstral coefficients; [00191 - the determined number of components of a vector of characteristics is a number between 150 and 200, for example equal to 193 andiou the given labels characterizing a determined noise level are included a list of predefined values in number included between 6 and 10, for example equal to 6 (for example in the no parasite ", u very slight parasite", "slight parasite", "parasite", "important parasite", "very important parasite");

- during the step of determining a label characterizing the noise level of a test audio signal, the processing unit further determines an index, called the confidence index, the value of which reflects the precision of the prediction model for said test audio signal.

In a second aspect, the invention also relates to a processing unit, comprising software means for executing all the steps of the method according to the first aspect above,) 022) A last aspect of the invention relates to a motor vehicle comprising a processing unit according to the second aspect.

Brief description of the drawings

[0023] Other characteristics and advantages of the invention will appear further on. reading the description which follows.

This is purely illustrative and should be read in conjunction with the accompanying drawings on which

[0074] [lai 1] is a block diagram illustrating the functional interactions between the various elements involved in the implementation of the method according to the invention;

[0025] ltig.2] is a font diagram; mnel illustra of the modes of implementation of the procedC, according to the invention; ) 026 is a diagram of steps illustrating the implementation of the method according to, in n; and,

If ig.4] is a graph giving an example of si al test tL, 'used for the implementation of the method according to the invention.

Detailed description of embodiments

In the description of embodiments which will follow and in the figures of the accompanying drawings, the same or similar elements bear the same numerical references to the drawings,

Referring to Figure 1, there may be described an example of a device in which the classification method according to the invention can be implemented.

The block diagram of Figure 1 illustrates. the functional interactions between the different elements involved in this implementation.

As has already been said above, the method makes it possible to test the performance of the radio reception system of a motor vehicle.

More particularly, the method makes it possible to establish the classification of the noise level for various audio signals themselves obtained from radiofrequency signals picked up by the vehicle radio reception system. Thus, in the example shown, the motor vehicle 103 captures , via an on-board receiving antenna, a radiofrequency signal 102.

Typically, it is an FM radio type signal (for “F ruant? Nu y Modulation”) which carries the audio signal of interest.

Such a radiofrequency signal may have been transmitted, for example, by a relay antenna or microwave relay 101, or by any other device for transmitting radiofrequency signals.

In various embodiments of the method, the receiving antenna of the motor vehicle may or may not be integrated into a tuner 104 of the vehicle.

In all cases, such a tuner 104 recovers the radiofrequency signal and converts it into an audio signal.

In conventional use, it is this audio signal which is transmitted to the vehicle's speakers which ultimately emit the sound perceptible by the driver and the passengers of the vehicle.

In order to implement the various steps of the method, the audio signal thus supplied by the tuner 104 can be either transmitted directly to the processing unit 106 which executes the classification method according to the invention, evenly recorded by a device 105, that is to say, for example, stored in a memory, to be subsequently processed by the processing unit 106.

Thus, the classification of the various audio signals received can be carried out in real time, as new signals are received, when the vehicle is traveling in different configurations (environment, arrangement of vehicle equipment, etc.).

But, as a variant it. can also be performed after the expiration of a fixed period, left to the discretion of the user.

In other words, it is possible, in the second case, to apply a signal processing desired by the user to an audio signal which has been received and recorded before transmitting it to the processing unit 106.

In addition, the processing unit 106 can both be on board the motor vehicle and be located in a laboratory (for example if it is a computer) in which various tests can be carried out on the devices. recorded audio signals.

The advantage of a laboratory implementation is that the tests can be performed in an environment isolated from ambient noise, and in a reproducible manner.

With reference to FIG. 2 and to FIG. 3, there will now be described embodiments of the classification method according to the invention.

FIG. 2 shows a schematic representation of the course of the process and FIG. Shows a diagram of steps illustrating the various steps carried out during the execution of the method. 10033j The method takes place in two distinct phases: a learning phase or training phase 201, on the one hand, and a test phase 202, on the other hand.

The learning phase leads to the establishment of a prediction model which makes it possible to classify the audio signals weighted thereafter. in the test phase The ballast phase consists of using this model to classify the audio signals tested.

Let us begin by describing the learning phase 201, which comprises steps 301, 302 and 303.

[003511] Step 301 consists in the reception, by the processing unit 106, of at least one pre-recorded audio signal 203, and which is associated with a label characterizing the determined noise level of this signal.

By pre-recorded audio signal is meant, for example, an audio signal stored in a memory after having been generated according to a process in accordance with that described above with reference to FIG. 1.

It can also be a synthesized signal (that is to say generated by a modeling) in order to generate sufficiently different audio signals to best train the machine learning algorithm which will be described in more detail. far.

In both cases, a sufficient number of these pre-recorded Andean signals (and at least one such signal) is used during the learning phase.

The labels which are associated with the noise signals designate noise level classes.

These classes make it possible to simply identify the noise level of a given signal, with a qualitative notion.

For example, in a wagering mode. in particular, the labels are included in the following list “no interference”, “very slight interference”, “slight interference”, “interference”, “significant interference: >>,“ very significant interference ”.

Any other list, preferably with a number of predefined values between 6 and 10, can be envisaged.Each audio signal can therefore be assigned a label characterizing its noise level, according to these six different levels, on a qualitative scale ranging from l absence of interference at the most important interference.

This number is considered sufficient to account for the variation in audio quality, and remains reasonable enough to allow comparative analysis by a human being.

[00371 In addition to this example, those skilled in the art will appreciate that the number of classes (therefore of labels) used to characterize the noise level as well as the criteria on which these classes are based can vary according to the modes of implementation. of the process.

For example, the noise level can be characterized by the inter and / at level by the level of granularity of an audio signal, instead of a characterization according to the magnitude of the so-called signal interference.

Typically, and regardless of the mode of noise characterization, labels may have been previously associated with prerecorded audio signals, being determined by human testers qualitatively identifying the noise level of each signal.

The processing unit step 302, from each prerecorded audio signal, a vector of characteristics 204 having a determined number of components corresponding to respective characteristics of the recorded audio signal, to generate such a vector of characteristics, the processing unit may, for example, use an audio signal processing method based on the use of the power spectral density of the noise.

More precisely, the method consists in calculating, in a manner known per se, from the audio signal, its cepstrum of. “Afel-frequency” type. The cepstral coefficients thus determined are then assigned to the components of the vector of characteristics.

Typically, the number of components of a feature vector is a number between 150 and 200.

Tl can be equal, for example, to 193. [) 040] FIG. 4 shows an example of the evolution of an audio signal as a function of time t.

In the figure, the gray colored graph illustrates the evolution of the amplitude A (normalized between -1 and 1) of the audio signal, The solid black line illustrates, for different portions of this audio signal, the evolution of the value corresponding to the noise level class.

In the example shown, the number 0 corresponds to the “no parasite” class and the number 6 corresponds to the “very significant parasite” class as they were introduced above.

Each portion of this black curve therefore corresponds to a recorded audio signal with which is associated. a given label.

Typically, each recorded audio signal has a duration of about a few seconds, in particular between 3 and 8 seconds, and for example equal to 5 seconds. [0041; Returning to the flowchart of FIG. 3, during step 303, each characteristic vector and its associated label are used as input data of a machine learning algorithm 205.

This type of algorithm is known per se to Phot-moulting in the art.

Such an algorithm systematically evolves by itself, on the basis of input data, in order to be able to best perform a given task. More particularly, in the modes of implementation of the machine learning algorithm created, on the basis of aforementioned input data, a prediction model 206 associating data components of feature vectors with data labels characterizing a determined noise level, [) 042] As has already been said above, the machine learning algorithm 205 preferably uses several feature vectors as input data to establish the most accurate prediction model possible.

Where appropriate, these feature vectors and the label associated with each of them can be grouped together in a matrix used by the algorithm to generate the model.

The learning phase 2.01 ends with the implementation of step 303 which allows the processing unit 106 to establish a relevant prediction model for labeling tested audio signals, which by hypothesis does not are not yet labeled.

The steps which will now be described are implemented in the test phase of the method.

This test phase comprises steps 304, 305 and 306 which will now be described. [t} 045! Step 304 consists of the reception, by the processing unit 106, of an audio test signal 207.

As has already been said with reference to FIG. 1, this signal has typically been generated by the tuner of a motor vehicle, from a radiofrequency signal actually received by the receiving antenna of the vehicle.

During step 305, and similarly to step 302, the processing unit generates, from the test audio signal, a vector of characteristics 208 before a determined number of components corresponding to characteristics respective values of the test audio signal. () 04] Finally, during step 306, the processing unit determines, on the basis of the components of the characteristic vector 208 of the test audio signal 207 and also on the basis of the prediction model 206 associating components feature vector data with labels characterizing the noise level of a corresponding audio signal, a label 209 characterizing the noise level of the test audio signal.

In a particular embodiment of the method, the processing unit determines, at each occurrence of the test phase of an audio signal, a confidence index 210, the value of which reflects the precision of the model. prediction for the signal in question.

The determination of this type of index, known per se to those skilled in the art, makes it possible to evaluate the relevance of the classification in order to take into account or not the audio signal and its label, depending on the value of the index, in postprocessings which are not described here.

[0049] Thus, advantageously, the method makes it possible to reduce the resources required (fuel, vehicle, time, etc.) to test the performance of a radio reception system compared to current systems.

In addition, the process allows the traceability of the tests since the results are reproducible and can be analyzed in the laboratory.

The test can be industrialized and does not require heavy specific training for the operators responsible for implementing it.

It thus avoids the need to call upon personnel with a very high qualification, allowing its easy execution by any person with a minimum of training.

The present invention has been described and illustrated in the present detailed description and in the figures of the accompanying drawings, in possible embodiments.

The present invention is not, however, limited to the embodiments presented.

Other variants and embodiments can be deduced and implemented by a person skilled in the art on reading the present description and the accompanying drawings.

In the claims, the term "comprising" or "comprising" does not exclude other elements or other steps.

A single processor or several other units can be used to implement the invention.

The different characteristics presented and / or claimed can be advantageously combined.

Their presence in the description or in different dependent claims does not exclude this possibility.

The reference signs cannot be understood as limiting the scope of the invention.

10 [Claim 1] [Claim 2] vendication 3] [Claim 4]

Claims (1)

Claims [Claim 1] Method for classifying the noise level of an audio signal for testing the radio reception quality of a motor vehicle audio system under test, said method comprising the following steps, carried out by a processing unit (106): - receiving (304) a test audio signal (207); - generating (305) from the test audio signal a vector of characteristics (208) having a determined number of components corresponding to respective determined characteristics of the test audio signal; and,  determining (306), based on the components of the characteristic vector of the test audio signal and further based on a prediction model (206) associating given components of characteristic vectors with labels characterizing the level of noise of a corresponding audio signal, of a label (209) characterizing the noise level of the test audio signal. [Claim 2] Classification method according to claim 1, further comprising, in a prior learning phase:  receiving (301) at least one pre-recorded audio signal, associated with a given label characterizing a determined noise level of said recorded audio signal; - generating (302) from each pre-recorded audio signal a vector of characteristics (204) with a determined number of components corresponding to determined characteristics of the pre-recorded audio signal; and, - the use (303) of each vector of characteristics and its associated label as input data of a machine learning algorithm (205), and the. creation, by the machine learning algorithm, of a prediction model associating given components of characteristic vectors with given labels characterizing a determined noise level. [Claim 3] Classification method according to claim 1 or claim 2, in which the test audio signal is obtained from a radiofrequency signal effectively received by an antenna of the motor vehicle and effectively converted into an audio signal by a tuner (104) of said vehicle automotive, and supplied to the processing unit. [Claim 4] The classification method according to claim 3, wherein the signal test audio which is obtained after conversion of the radiofrequency signal by the tuner of the motor vehicle, is stored in a memory (105) and transmitted to the processing unit (106) after the expiration of a duration [Claim 5] tick icriïiincc. Classification method according to any one of claims 1 to 4, in which each audio signal has a determined duration equal to a few seconds, in particular between 3 and 8 seconds, for example equal to 5 seconds. [Claim 6] Method according to any one of Claims 1 to 5, in which the conversion of an audio signal, received by the processing unit, into a vector of characteristics with a determined number of components corresponding to determined characteristics of the audio signal, is performed according to an audio signal processing method consisting in calculating a cepstrum of the “Mel-frequency” type from the audio signal, and in which the components of the characteristic vector correspond to the calculated cepstral coefficients, [Claim 7] Method according to any one of Claims 1 to 6, in which the determined number of components of a characteristic vector is a number between 150 and 200, for example equal to 193 and / or in which the labels characterizing a level of determined noise are included in a list of predefined values in number between 6 and 10, for example equal to 6. [Claim 8] Method according to any one of Claims 1 to 7, in which, during the step of determining a label characterizing the noise level of a test audio signal, the processing unit also determines an index, said confidence index, the value of which reflects the accuracy of the prediction model for said test audio signal. [Claim 9] Processing unit, comprising software means for executing all the steps of the method according to any one of claims 1 ?! R [Claim 10] îjL O » Motor vehicle comprising a processing unit according to claim 9. 1/3