EP1468416B1 - Method for qualitative evaluation of a digital audio signal - Google Patents

Description

The present invention relates to a method for evaluating a digital audio signal, in particular a digitally transmitted signal and / or a digital signal to which digital coding has been applied, in particular with rate reduction and / or decoding. A digitally transmitted signal may be an autonomous audio signal (broadcast) or an audio signal that accompanies a program such as an audiovisual program.

The field of digital radiocommunications and broadcasting is expanding, particularly with the advent of digital television and radio telephones. New instruments must be developed to measure the quality of all the systems needed to implement this technology, and thus ensure quality of service.

It is for this purpose that subjective tests are used. These tests make it possible to judge the quality of sound signals by making them listen by auditors, experts or novices. This method is long and expensive because the conditions to be met during these tests are numerous and strict (choice of panelists, listening conditions, sequences, chronology of tests, etc ...). Nevertheless, it makes it possible to build databases of reference signals with the notes that have been assigned to them. It is these tests that make it possible to obtain Mean Opinion Score (MOS) ratings, which are recognized as the benchmark for quality estimation.

To try to minimize the number of these subjective tests, many studies have been done on the human hearing system. From then on, ear modeling and psychoacoustic phenomena were developed, which allowed the analysis and estimation of sound signal quality by objective methods. The measured quality being that perceived by the human ear, it is named objective perceptual quality.

It is possible to differentiate three classes of objective qualification methods: The first ("complete reference") directly compares the original signal to the degraded signal (after coding, broadcasting, multiplexing, ...), the second only compares parameters extracted from two signals (called reduced reference). In the third, the defects generated by the diffusion chain are detected using their main known characteristics. This last class makes it possible to overcome the constraints related to the use of the reference signal. In all other cases, the reference must be transmitted instead of comparison then perfectly synchronized with the degraded signal. This makes the system complex and more expensive.

Degradations due to transmission errors significantly reduce the quality of the signal. They appear during the broadcast, a MPEG digital stream for example or during broadcasting, including radio, on the Internet.

In such a context, it is desirable to have a method that makes it possible objectively to measure the quality of an audio signal after broadcasting, without using a reference signal and / or by using a reduced reference. Indeed, only these techniques are suitable for monitoring a broadcast network for example where several measuring points distant from each other may be necessary. It is also interesting to take advantage of the relative simplicity of such a method for measuring the quality of a digital audio signal transmitted or not, which has been subjected to digital coding, in particular rate reduction, and / or decoding. .

The number of audio quality methods developed is very variable depending on the class considered. Indeed, a large number of fully referenced methods have been developed. Only a few methods have been developed without reference or with reduced reference.

The fully referenced methods for which the signal to be evaluated is compared to the reference signal correspond to conventional techniques used to estimate the quality of audio coders for example. Their general principle is based on the calculation, through a perceptual model of hearing, of an internal representation of the original signal and of the degraded signal, then on a comparison of these two internal representations. Such a method is described in the article of John G. BEERENDS and JAN A. STEMERDINK entitled "A Perceptual Audio Quality Measure Based on a Psychoacoustic Sound Representation", published in "Journal of Audio Engineering Society," Vol. 12, December 1992, pages 963 to 978 .

These models of hearing are established from masking experiments, in order to obtain a representation that is as faithful as possible, and must make it possible to predict if the deteriorations will be audible or not. All degradation on a signal is not audible or annoying. These perceptual models with reference are based on the schema of the Figure 1 . Many methods, more or less complete and elaborate, are based on this principle. Recently, the PEAQ algorithm has been standardized by the ITU-R (ITU-R BS.1387 standard). This algorithm is based on classical principles by associating a model of quality prediction using a network of neurons.

The major interest of these techniques is to be able to detect very small degradations but, it must be borne in mind that they are intended to study the influence of a coding. The measurements obtained are relative: only the difference is taken into account in this type of measurements. In the case of an encoder of very good quality, a signal with significant degradation will be coded and then decoded almost transparently, and therefore, the score will be very high. In addition, for a signal that has been modified (equalized, colored, ...) between the calculation of the reference and the comparison, the note may be low even if the two signals are of very good perceptual quality.

Regarding methods without reference, these remain very few. Output-Based Objective Speech Quality (OBQ) is the most accomplished non-reference technique. This method of estimating the quality of a speech signal only, without a reference signal, is based on the calculation of perceptual parameters representing the content of the signal, gathered into a vector. These vectors, calculated on non-degraded signals, will constitute a reference base. The quality will be estimated by comparing the same parameters extracted from the degraded signals with the vectors of the reference base. The main method using neural networks is OSSQAR (Objective Scaling of Sound Quality and Reproduction). The general principle of this method is to use an auditory model in conjunction with a neural network. The network is trained to predict the subjective quality of a signal from its perceptual representation calculated by the hearing model, to simulate the phenomena of psychoacoustics. It should be noted that the results obtained by these methods are much better when the signals are part of the learning base or at least when they have similar characteristics.

Such methods are therefore not suitable for evaluating the quality of any signals, for example the audio signals of a radio or TV program.

As indicated above, most of the objective perceptual measurement algorithms with complete reference operate according to an identical principle: it is a question of comparing the degraded sound signal with the original signal (signal before transmission and / or coding and / or decoding, called reference signal). These algorithms therefore need to have a reference signal, which is also synchronized very precisely with the signal to be tested. These conditions can only be fulfilled in simulation or when testing encoders and other "compact" or geographically non-distributed systems; on the other hand, it is very different when receiving a signal broadcast from A ₁ and A ₂ reception antennas ( Figure 2 ).

The reference signal must be available at the different points of comparison. Also, to be able to use a method with complete reference, the only possibility is to transmit the reference, without error, to the points of comparison, then to synchronize it perfectly. For reasons of spectrum congestion and therefore of cost, these techniques with complete reference are not applicable in practice, because they would require the use of a second transparent transmission path.

The non-reference methods proposed allow to obtain good results but only in the case of signals with characteristics known and modeled during the learning phase. Non-reference methods therefore work poorly on any signal.

It has been suggested to use a so-called "reduced" reference in which the reference audio signal is characterized by one or more numbers. Such a process has been described in the French patent application. FR 2,769,777 filed October 13, 1997 . However, this method does not make it possible to process all the samples, particularly because the rate of the proposed reference signal is too great (at least 36 kbits / s for windows of 1024 signal samples) to satisfy the practical conditions. implementation and implementation in a television broadcasting network.

In an article entitled "An Efficient Method for Objective Audio and Video Quality Assessment in Digital Television Applications" (N. Montard et al., Signal Processing & Theories and Applications Proceedings of EUSIPCO 2000, 10th European Signal Processing Conference "held in Tampere, Finland) from 4 to 8 September 2000, pages 2189-2192, ISBN: 952-15-0443-9 ), of which one of the co-authors is also the inventor of the present invention, there is described a method in which reference parameters specific to the degraded audio / video signal and to the audio / video reference signal are calculated and then compared. and analyzed in order to reduce certain indicators of the deteriorations suffered in the chain of transmission of the signal considered. These indicators are then themselves the subject of a second analysis which will make it possible to obtain an overall quality score.

The present invention provides a method in which the indicators are simpler and can be calculated in real time and in continuous time, and require a significantly lower bit rate. Degradations that can only modify a few samples, while degrading the quality significantly, the proposed method allows the entire audio stream to be analyzed.

The method according to the invention makes it possible to reliably estimate the quality of an audio signal having passed through a digital type transmission or coding. Indeed, the disturbances that the transmission channels undergo can induce the appearance of errors on the transmitted data; these errors result in impairments in the final audio signal.

The proposed technological approach is to perform a measurement on the audio signal at the input and another at the output channel or any other system to be studied. A comparison between these measures makes it possible to ensure the "transparency" of the transmission channel and to evaluate the importance of the introduced impairments.

Used together or not with non-reference methods, detecting degradations based on the signature of the characteristics of the most important faults to be searched for, the proposed approach allows a reliable estimation of the introduced degradations. It also makes it possible to compensate for a lack of a reference signal. This method makes it possible to reduce the reference flow required for quality estimation in the case of measurements with reduced reference, and the number of parameters to be used in the case of measurements without reference.

The invention thus relates to a method for evaluating a digital audio signal according to claim 1,

The digital audio signal to be evaluated may be a signal which has been digitally transmitted and / or which has been subjected to digital coding, especially to rate reduction, from a digital reference signal.

The quality indicator vector may be constituted by said minimum value, or else by a binary value resulting from the comparison of said minimum value with a given threshold. Also, the method can be characterized in that it implements the calculation of a quality score by determining a cumulative time interval during which said minimum value is lower than a given threshold and / or by determining the number of times per second where said minimum value is below a given threshold or in that said minimum values are generated for both the reference audio signal and the audio signal to be evaluated and that a quality vector is generated comparing the corresponding minimum values of the reference audio signal and the audio signal to be evaluated, for example by calculating the difference or the ratio between said minimum values.

• la figure 1 est un organigramme illustrant une évaluation de qualité à référence complète.
• la figure 2 illustre une transmission audio avec perte de qualité,
• les figures 3 à 6 et 8 à 10 illustrent des procédés d'évaluation alternatifs qui ne correspondent pas à la définition de l'invention selon la revendication 1. Le procédé d'évaluation selon l'invention est décrit en relation avec la figure 7,
• et les figures 11 et 12 illustrent un système de qualitométrie audio mettant en oeuvre la présente invention.

Other features and advantages of the invention will appear better on reading the following description in connection with the drawings in which:

• the figure 1 is a flowchart illustrating a full reference quality assessment.
• the figure 2 illustrates an audio transmission with loss of quality,
• the Figures 3 to 6 and 8 to 10 illustrate alternative evaluation methods which do not correspond to the definition of the invention according to claim 1. The evaluation method according to the invention is described in connection with the figure 7 ,
• and the figures 11 and 12 illustrate an audio quality system embodying the present invention.

Management and recovery of decoding errors is not standardized. The influence of these errors on the perceived quality therefore depends on the decoder used.

The audibility of these defects is also related to the type of element assigned in the frame, for example MPEG, and to its audio content.

In the case of significant errors due to transmission, the signal quality drops sharply. These impairments appear during the broadcast, an MPEG digital stream for example, and are, most of the time, impulse type. They may also appear when streaming an audio stream over the Internet, or when coding or decoding.

For this type of fault, the quality can be estimated in a binary way: either the signal has not been degraded and the quality will depend on the initial coding used, or errors have been introduced and significant degradations appear.

The estimation of the quality can then be done by methods without reference, by accounting for the degradations detected on regular time intervals of the order for example of the second. Subjective tests have indeed made it possible to obtain a reliable estimate of perceived quality, based on the number and length of interruptions linked to impulse-type impairments in a signal.

For the measurements obtained with reduced reference, the proposed method makes it possible to reduce the flow necessary for the transport of the reference. This allows the use of reserved lanes at relatively low flow. These measurements make it possible to detect degradations other than those due to transmission errors.

Thus, the present invention allows a reduction of the flow rate in the case of measurements with reduced reference and, by the addition of simple measurements without reference, to conserve measurements on the important degradations in the case of a loss of the reference, for example, by locally generating a vector that simply characterizes the degradations, and that could therefore easily be processed and transmitted to a control installation, including centralized.

Measurements carried out along the chain and at various points in the network, inform the system of monitoring and management of the broadcast in digital television, on its overall performance. The measurements of the signal degradations inform the broadcast operator of the quality of service delivered.

• Avec référence réduite. L'approche technologique proposée consiste à effectuer une mesure sur le signal audio, à l'entrée, et une autre à la sortie de la chaîne de transmission ou tout autre système à étudier (codeur, décodeur, etc...). Une comparaison entre ces mesures permet de s'assurer de la "transparence" de la chaîne ou du système et d'évaluer l'importance des dégradations introduites. A la différence de la technique antérieure :
  • le procédé effectue une évaluation en temps réel et en temps continu.
  • les mesures de référence à l'entrée de la chaîne représentent une quantité de données très faible par rapport aux données du signal audio, d'où sa classification en « référence réduite ».
  • les données ou mesures de référence utilisées sont aussi bien une représentation réduite du contenu du signal, qu'une mesure de l'importance d'un type de dégradation.

The process is characterized by two complementary modes of operation:

• With reduced reference. The technological approach proposed is to measure the audio signal at the input and another at the output of the transmission chain or any other system to study (encoder, decoder, etc ...). A comparison between these measures makes it possible to ensure the "transparency" of the chain or the system and to evaluate the importance of the introduced impairments. Unlike the prior art:
  • the method performs an evaluation in real time and in continuous time.
  • the reference measurements at the input of the string represent a very small amount of data compared to the data of the audio signal, hence its classification as "reduced reference".
  • the reference data or measurements used are both a reduced representation of the signal content and a measure of the importance of a type of degradation.

The invention makes it possible to overcome a lack of a reference signal. For this, the method defines measurements for the characteristic digital defects to be sought. Unlike the prior art, the proposed approach allows an estimation of the impairments introduced on any signal, and reliably and this approach can be implemented both at the scale of a transmission network and locally on an equipment. In addition, the calculation complexity according to the method is low, and the indicator obtained represents a small amount of data compared to the digital audio stream.

Finally, the method can be applied indifferently to purely digital signals or to signals having undergone, after transmission, a digital to analog and then an analogue to digital conversion.

The first three methods described below are of the type called "with reduced reference". These methods, however, do not correspond to the definition of the invention according to claim 1.

To obtain greater precision in the estimation of quality, some of the parameters developed use perceptual models: The principle of objective perceptual measurements is based on the transformation of the physical representation (sound pressure, level, time and frequency) into the psychoacoustic representation (sound strength, masking level, time and critical bands or barks) of two signals (the reference signal and the signal to be evaluated) in order to compare them. This transformation takes place through a modeling of the human auditory apparatus (generally, this modeling consists of a spectral analysis in the field of Barks followed by the phenomena of spreading). A distance can then be calculated between the psychoacoustic representations of the two signals, a distance that can be related to the quality of the signal to be evaluated (the lower the distance, the closer the signal to be evaluated is to the original signal and the better is its quality ).

The first method implements a parameter called "Perceptual Account Gap".

• Fenêtrage du signal temporel en blocs puis, pour chacun des blocs, calcul de l'excitation induite par le signal en utilisant un modèle d'audition. Cette représentation des signaux tient compte des phénomènes de la psychoacoustique, et fournit un histogramme dont les comptes sont les valeurs des composantes basilaires. Cela permet de ne prendre en considération que les composantes audibles du signal et donc de se limiter à l'information utile. Pour obtenir cette excitation, les modélisations classiques peuvent être utilisées : atténuation de l'oreille externe et moyenne, intégration selon les bandes critiques et masquages fréquentiels. Les fenêtres temporelles choisies sont d'environ 42 ms (2048 points à 48 kHz) avec un recouvrement de 50%. Cela permet d'obtenir une résolution temporelle de l'ordre de 21 ms.

The calculation of this parameter is broken down into several stages, necessary to take psychoacoustics into account. These are applied to the reference signal and the degraded signal. These steps are as follows:

• Windowing of the temporal signal in blocks then, for each of the blocks, calculation of the excitation induced by the signal by using a model of hearing. This representation of the signals takes into account the phenomena of psychoacoustics, and provides a histogram whose accounts are the values of the basilar components. This allows to consider only the audible components of the signal and therefore to limit itself to the useful information. To achieve this excitement, classical modeling can be used: attenuation of the outer and middle ear, integration according to critical bands and frequency masking. The time windows chosen are approximately 42 ms (2048 points at 48 kHz) with a coverage of 50%. This makes it possible to obtain a temporal resolution of the order of 21 ms.

Several steps are necessary for this modeling. For the first step, the outer and middle ear attenuation filter is applied to the power spectral density, obtained from the signal spectrum. This filter also takes into account the absolute threshold of hearing. The notion of critical bands is modeled by a transformation of the frequency scale into a basilar scale. The next step is the calculation of the individual excitations to account for masking phenomena, thanks to the frequency spreading function in the basal scale and a nonlinear addition. The last step allows to obtain the loudness compressed, by a function power, to model the nonlinear sensitivity in frequency of the ear, by a histogram comprising the 109 basilar components.

The counts of the obtained histogram are then grouped into three classes. This vectorization makes it possible to obtain a visual representation of the evolution of the structure of the signals. This also makes it possible to obtain a simple and concise characterization of the signal and thus to have a particularly interesting reference parameter.

$${\mathrm{S}}_2=\mathrm{73}\mathrm{soit\; z}=\frac{\mathrm{24}}{\mathrm{109}}*\mathrm{73}=\mathrm{16},\mathrm{073}\mathrm{Barks}$$

Several strategies exist to set the bounds of these three accounts: The simplest is to separate the histogram into three zones of equal sizes. Thus, the 109 basilar components, (or the 24 components that constitute the excitation and constitute a simplified representation) represent 24 Barks and can be separated at the following indices: $${\mathrm{<figure-callout\; id="1"\; label="S"\; filenames="imgf0002.png"\; state="\{\{state\}\}">S</figure-callout>}}_{\mathrm{1}}=\mathrm{36}\mathrm{either\; z}=\frac{\mathrm{24}}{\mathrm{109}}*\mathrm{36}=\mathrm{7},\mathrm{927}\mathrm{<figure-callout\; id="2"\; label="Barks\; S"\; filenames="imgf0007.png"\; state="\{\{state\}\}">Barks\; </figure-callout>}$$

$${\mathrm{S}}_{}$$$${\mathrm{}}_2=\mathrm{73}\mathrm{either\; z}=\frac{\mathrm{24}}{\mathrm{109}}*\mathrm{73}=\mathrm{16},\mathrm{073}\mathrm{Barks}$$

$${\mathrm{S}}_2=100\mathrm{soit\; z}=\frac{\mathrm{24}}{\mathrm{109}}*100=22,018\mathrm{Barks}$$

The second strategy takes into account the scaling areas of Beerends. Indeed a gain compensation between the excitation of the reference signal and that of the signal to be tested is performed by the ear, the fixed limits are then as follows: $${\mathrm{<figure-callout\; id="1"\; label="S"\; filenames="imgf0002.png"\; state="\{\{state\}\}">S</figure-callout>}}_{\mathrm{1}}=\mathrm{9}\mathrm{either\; z}=\frac{\mathrm{24}}{\mathrm{109}}*\mathrm{9}=1,\mathrm{982}\mathrm{<figure-callout\; id="2"\; label="Barks\; S"\; filenames="imgf0007.png"\; state="\{\{state\}\}">Barks\; </figure-callout>}$$

$${\mathrm{S}}_{}$$$${\mathrm{}}_2=100\mathrm{either\; z}=\frac{\mathrm{24}}{\mathrm{109}}*100=22,018\mathrm{Barks}$$

$$\mathrm{Y}={\mathrm{C}}_{\mathrm{2}}/\mathrm{N}*\sin \left(\mathrm{\pi }/\mathrm{3}\right)$$

• avec C₁ : somme des excitations basilaires pour les hautes fréquences (au-dessus de S₂)
• C₂ : compte associé aux fréquences moyennes (composantes entre S₁ et S₂)
• et N = C₁ + C₂ + C₃ : Somme totale des valeurs des composantes.

The trajectory is then represented in a triangle, called a triangle of frequencies. For each block three counts C ₁ , C ₂ and C ₃ are obtained, hence two Cartesian coordinates according to the following formulas: $$\mathrm{X}={\mathrm{<figure-callout\; id="1"\; label="VS"\; filenames="imgf0002.png"\; state="\{\{state\}\}">VS</figure-callout>}}_{\mathrm{1}}/\mathrm{NOT}+\frac{{\mathrm{<figure-callout\; id="2"\; label="VS"\; filenames="imgf0007.png"\; state="\{\{state\}\}">VS</figure-callout>}}_{\mathrm{2}}/\mathrm{NOT}}{\mathrm{2}}$$

$$\mathrm{Y}={\mathrm{<figure-callout\; id="2"\; label="VS"\; filenames="imgf0007.png"\; state="\{\{state\}\}">VS</figure-callout>}}_{\mathrm{2}}/\mathrm{NOT}*\sin \left(\mathrm{\pi }/\mathrm{3}\right)$$

• with C ₁ : sum of basilar excitations for high frequencies (above S ₂ )
• C ₂ : account associated with the average frequencies (components between S ₁ and S ₂ )
• and N = C ₁ + C ₂ + C ₃ : Total sum of component values.

A point (X, Y) constituting a vector is thus obtained for each time window of the signal, which corresponds to the transmission of two values per window of for example 1024 bits, ie a bit rate of 3 kbits / s for a sampled audio signal. at 48 kHz. For a complete sequence, the associated representation is thus a trajectory parameterized by time, as shown in FIG. Figure 3 .

The (Euclidean) distance between the reference signal and the degraded signal is then calculated. In the case of continuous quality estimation, the distance between the points makes it possible to estimate the importance of the impairments introduced between the reference signal and the degraded signal. This distance can be considered as a perceptual distance because of the use of psychoacoustic models.

To estimate a quality score for a signal of several seconds, it is possible to calculate an overall measurement of the difference between the two signals. For this, several metrics can be used. These can be of diffuse type (average of the distances between the vertices, intercepted area, ...), local (maximum, minimum distances between vertices, ...) and depend on the position in the triangle.

It is also possible to take into account the barely perceptible differences ("Just Noticeable Difference"). These thresholds make it possible to determine the audibility of the differences that have appeared. They can be modeled by tolerance zones depending on the position in the triangle to take into account the variability of masking phenomena.

In all cases, the two trajectories must be synchronized in advance.

The principle of calculating this comparative parameter can thus be summarized by the diagram Figure 4 .

The main advantage of the parameter comes from the fact that psycho-acoustic phenomena are taken into account without increasing the flow rate. necessary to transfer the reference. This reduces the reference to 2 values for 1024 signal samples (3 kbit / s).

The second method implements autoregressive modeling of the signal.

The general principle of linear prediction is to model the signal as a combination of its past values. The idea is to calculate the N coefficients of a prediction filter by an autoregressive modeling (any pole). With this adaptive filter, it is possible to obtain a predicted signal from the actual signal. The prediction errors or residuals are calculated by difference between these two signals. The presence and amount of noise in a signal can be determined by analyzing these residues.

The comparison of the residues obtained on the reference signal and those calculated from the degraded signal, and therefore of the noise levels, makes it possible to estimate the importance of the modifications and defects inserted.

The reference to be transmitted corresponds to the maximum of the residues over a time window of given size. It is indeed not interesting to transmit all the residues if the rate of the reference wants to be reduced.

• L'algorithme de LEVINSON-DURBIN qui est décrit par exemple dans l'ouvrage de M. BELLANGER - Traitement numérique du signal - Théorie et pratique (MASSON éd. 1987) p. 393 à 395 . Pour l'utiliser, il faut disposer d'une estimation de l'autocorrélation du signal sur un ensemble de N₀ échantillons. Cette autocorrélation est utilisée pour résoudre le système d'équations de Yule-Walker et ainsi obtenir les coefficients du filtre prédicteur. Seules les N premières valeurs de la fonction d'autocorrélation peuvent être utilisées, où N désigne l'ordre de l'algorithme, c'est-à-dire le nombre de coefficients du filtre. Sur une fenêtre de 1024 échantillons, on garde le maximum de l'erreur de prédiction.
• L'algorithme du gradient qui est décrit par exemple dans l'ouvrage précité de M. BELLANGER p. 371 et suivantes. Le principal inconvénient du paramètre précédent est la nécessité, dans le cas d'une implantation sur DSP, de stocker les N₀ échantillons pour estimer l'autocorrélation, avoir les coefficients du filtre puis calculer les résidus. Ce second paramètre permet d'éviter cela en utilisant un autre algorithme permettant de calculer les coefficients du filtre : l'algorithme du gradient. Celui-ci utilise l'erreur commise pour mettre à jour les coefficients. Les coefficients du filtre sont modifiés dans la direction du gradient de l'erreur quadratique instantanée, avec le signal opposé.

To adapt the coefficients of the prediction filter, two methods are given below by way of example:

• The LEVINSON-DURBIN algorithm, which is described, for example, in M. BELLANGER - Digital Signal Processing - Theory and Practice (MASSON, 1987) p. 393 to 395 . To use it, it is necessary to have an estimate of the autocorrelation of the signal on a set of N ₀ samples. This autocorrelation is used to solve the system of Yule-Walker equations and thus obtain the coefficients of the predictor filter. Only the first N values of the autocorrelation function can be used, where N denotes the order of the algorithm, that is to say the number of coefficients of the filter. On a window of 1024 samples, the maximum of the prediction error is kept.
• The gradient algorithm which is described for example in the aforementioned work by M. BELLANGER p. 371 and following. The main disadvantage of the previous parameter is the need, in the case of an implementation on DSP, to store the N ₀ samples to estimate the autocorrelation, to have the coefficients of the filter then to calculate the residues. This second parameter makes it possible to avoid this by using another algorithm making it possible to calculate the coefficients of the filter: the algorithm of the gradient. This uses the error made to update the coefficients. The Filter coefficients are changed in the direction of the gradient of the instantaneous squared error, with the opposite signal.

Once the residues obtained by difference between the predicted signal and the real signal, only the maximum of their absolute values, over a time window of given size T, is retained. The reference vector to be transmitted can thus be reduced to a single number.

After transmission then synchronization, the comparison consists of a simple calculation of the distance between the maxima of the reference and the degraded signal, for example by difference.

• Le principal avantage des deux paramètres est le débit nécessaire au transfert de la référence. Celui-ci permet de réduire la référence à 1 nombre réel pour 1024 échantillons de signal.

The Figure 5 summarizes the parameter calculation principle:

• The main advantage of the two parameters is the bit rate needed to transfer the reference. This reduces the reference to 1 real number for 1024 signal samples.

On the other hand, no model of psychoacoustics is taken into account.

The third method implements an autoregressive modeling of the basilar excitation.

Compared to classical linear prediction, this method makes it possible to take into account the phenomena of psychoacoustics, in order to obtain an evaluation of perceived quality. For that, the calculation of the parameter goes through a modeling of various principles of the hearing. A linear prediction models the signal as a combination of its past values. Residue analysis (or prediction errors) is used to determine and estimate the presence of noise in a signal. The major drawback when using these techniques is the fact that there is no consideration of the principles of psychoacoustics. Thus, it is not possible to estimate the amount of noise actually perceived.

The process follows the general principle of classical linear prediction. It also incorporates the phenomena of psychoacoustics to adapt to the non-linear sensitivity frequency (loudness) and intensity (pitch) of the human ear.

The spectrum of the signal is modified by means of an auditory model before calculating the coefficients of the linear prediction by an autoregressive modeling (any pole). The coefficients thus obtained make it possible to model the signal in a simple way while taking into account psychoacoustics. It is these prediction coefficients that will be transmitted and will serve as a reference when comparing with the degraded signal.

• Fenêtrage temporel du signal puis calcul d'une représentation interne du signal par modélisation des phénomènes de la psychoacoustique. Cette étape correspond au calcul de la sonie compressée, qui est en fait l'excitation induite par le signal au niveau de l'oreille interne. Cette représentation des signaux tient compte des phénomènes de la psychoacoustique, et est obtenue à partir du spectre du signal, en utilisant les modélisations classiques : atténuation de l'oreille externe et moyenne, intégration selon les bandes critiques et masquages fréquentiels. Cette étape du calcul est identique au paramètre décrit précédemment ;
• Modélisation autorégressive de cette sonie compressée afin d'obtenir les coefficients d'un filtre RIF de prédiction, tout comme dans une prédiction linéaire classique. La méthode utilisée est celle de l'autocorrélation, par résolution des équations de Yule-Walker. La première étape pour l'obtention des coefficients de prédiction est donc le calcul de l'autocorrélation du signal.

The first part of the calculation of this parameter corresponds to the modeling of the principles of psychoacoustics using classical hearing models. The second part is the calculation of the linear prediction coefficients. The last part corresponds to the comparison of the prediction coefficients calculated for the reference signal and those obtained for the degraded signal. The different stages of this method are as follows:

• Time window of the signal then calculation of an internal representation of the signal by modeling phenomena of psychoacoustics. This step corresponds to the calculation of the compressed loudness, which is actually the excitation induced by the signal at the level of the inner ear. This representation of the signals takes into account the phenomena of psychoacoustics, and is obtained from the spectrum of the signal, by using the classical modelizations: attenuation of the external and average ear, integration according to the critical bands and masking frequencies. This calculation step is identical to the parameter described previously;
• Autoregressive modeling of this compressed loudness in order to obtain the coefficients of a prediction RIF filter, just as in a classical linear prediction. The method used is that of autocorrelation, by solving the Yule-Walker equations. The first step in obtaining the prediction coefficients is therefore the calculation of the autocorrelation of the signal.

By considering the compressed loudness as a filtered spectral power, it is possible to calculate the autocorrelation of the perceived signal by inverse Fourier transformation.

One of the methods to solve this system of Yule-Walker equations and thus obtain the coefficients of a predictor filter is the use of the Levinson-Durbin algorithm.

• Estimation des dégradations par le calcul d'une distance entre les vecteurs issus de la référence et du signal dégradé. C'est une comparaison des vecteurs de coefficients obtenus pour la référence et pour le signal audio transmis, qui permet d'estimer les dégradations introduites lors de la transmission. Celle-ci doit se faire sur un nombre adapté de coefficients. Plus le nombre est important, plus les calculs peuvent être précis, mais plus le débit nécessaire à la transmission de la référence est élevé. Plusieurs distances peuvent être utilisées pour comparer les vecteurs de coefficients. L'importance relative des coefficients peut par exemple être prise en compte.

It is the prediction coefficients that constitute the reference vector to be transmitted to the point of comparison. The transformations used during the final calculation on the degraded signal are the same as for the initial phase on the reference signal.

• Estimation of the impairments by calculating a distance between the vectors derived from the reference and the degraded signal. It is a comparison of the coefficient vectors obtained for the reference and for the transmitted audio signal, which makes it possible to estimate the impairments introduced during the transmission. This must be done on an adapted number of coefficients. The larger the number, the more accurate the calculations can be, but the higher the rate needed to transmit the reference. Several distances can be used to compare vectors of coefficients. The relative importance of the coefficients can for example be taken into account.

The principle of the method is summarized according to the following scheme ( Figure 6 ).

The modeling of psychoacoustic phenomena makes it possible to obtain 24 basilar components. The order N of the prediction filter is 32. From these, 32 coefficients of the autocorrelation are estimated, which gives 32 prediction coefficients of which only 5 to 10 coefficients are conserved as the indicator vector of quality, for example the first 5 to 10 coefficients.

The main advantage of the parameter comes from taking into account the phenomena of psychoacoustics. To do this, it was necessary to increase the rate necessary to transfer the reference to 5 or 10 values for 1024 signal samples (21 ms for an audio signal sampled at 48 kHz), ie a flow rate of 7.5 to 15 kbit / s.

The following methods, of which only the first one actually corresponds to the definition of the invention according to claim 1, may be used with or without reference. This makes it possible to maintain the most important damage detection measures, even if no reference parameter is available at the checkpoint, at the time when the comparison should be made.

The first of these methods, according to the definition of the invention according to claim 1, implements a detection of dishes in the activity of the signal.

The concept of activity, which can be approximated by a derivation operation in the audio signal, is used to identify breaks and interruptions in the time signal.

These types of faults are characteristic of decoding errors after transmission of the digital audio stream or during the broadcasting of sound sequences on the Internet. This occurs when the network rate becomes insufficient to ensure the arrival of all the necessary frames at the time of decoding for example.

These degradations, which introduce zones of very weak activity, are translated on the auditory level by different sensations in the listener: cut of the sound, blurred sound, impulse noise ...

The first step of calculating the parameter corresponds to the estimation of the temporal activity of the signal. To do this, the second derivative operator is used. It makes it possible to have a sufficiently precise estimate of the activity and requires only very few calculations.

ou $$\mathrm{fʺ}\left({\mathrm{x}}_{\mathrm{0}}\right)=\mathrm{f}\left({\mathrm{x}}_{\mathrm{0}}+\mathrm{1}\right)-\mathrm{2.}\mathrm{f}\left({\mathrm{x}}_{\mathrm{0}}\right)+\mathrm{f}\left({\mathrm{x}}_{\mathrm{0}}-\mathrm{1}\right)$$

où f(t) correspond à la valeur de l'échantillon à l'instant t.To simulate this second derivative operation in a simple way, the following formula is used: $$\mathrm{f"}\left({\mathrm{x}}_{\mathrm{0}}\right)=\mathrm{f}\left({\mathrm{x}}_{\mathrm{0}}+\mathrm{2}\right)-\mathrm{2.}\mathrm{f}\left({\mathrm{x}}_{\mathrm{0}}\right)+\mathrm{f}\left({\mathrm{x}}_{\mathrm{0}}-\mathrm{2}\right)$$

or $$\mathrm{f"}\left({\mathrm{x}}_{\mathrm{0}}\right)=\mathrm{f}\left({\mathrm{x}}_{\mathrm{0}}+\mathrm{1}\right)-\mathrm{2.}\mathrm{f}\left({\mathrm{x}}_{\mathrm{0}}\right)+\mathrm{f}\left({\mathrm{x}}_{\mathrm{0}}-\mathrm{1}\right)$$

where f (t) corresponds to the value of the sample at time t.

où y(t) correspond à l'activité.A sliding average, on N values (for example N = 21, which corresponds to 0.5 ms for a sampling frequency of 48 KHz), then makes it possible to smooth the variations of the curve obtained and thus avoid false detections. Only one result will be retained per block of M results (M corresponds for example to 2048 audio samples). It is the minimum of average M which is conserved and transmitted. The parameter is thus obtained at time t by the following formula: $$\mathrm{dishes}\left(\mathrm{t}\right)=\underset{\mathrm{k}\in \mathrm{M}}{\min }\left(\frac{\mathrm{1}}{\mathrm{NOT}}{\displaystyle \underset{\mathrm{i}\in \mathrm{NOT}}{\Sigma }}\left|\mathrm{there}\left(\mathrm{t}-\mathrm{k}-\mathrm{i}\right)\right|\right)$$

where y (t) is the activity.

où Plats_r(t) et Plats_d(t) sont respectivement le paramètre calculé sur la référence et sur le signal dégradé.If the parameter is used with reference, then, after synchronization of the data, the comparison step consists of a simple difference which makes it possible to locate the areas where the signal has been replaced by decoding dishes. Only the moments, where the activity is strongly diminished on the degraded signal, are interesting. So the formula for comparison is as follows: $$\mathrm{d}\left(\mathrm{t}\right)=\max \left(\mathrm{0},{\mathrm{dishes}}_{\mathrm{r}}\left(\mathrm{t}\right)-{\mathrm{dishes}}_{\mathrm{d}}\left(\mathrm{t}\right)\right)$$

where Plats _r (t) and Plats _d (t) are respectively the parameter calculated on the reference and on the degraded signal.

To further reduce the rate required to transport the reference, it is also possible to compare the parameter Flat (t) , calculated on the signal, with a threshold S and thus obtain a binary parameter. During the onset of degradations, the drop in activity is indeed important enough to be detected in this way.

In this case, the comparison serves only to confirm the presence of the degradations. No more confusion is possible between the zones of silence and the zones of weak activity of the signal. Using the parameter without reference nevertheless makes it possible to identify the degradations.

In order to pass from a deterioration detection parameter to the estimation of a perceptual quality score, the psychoacoustic importance of the degradations detected must be analyzed. Depending on their length and number, the perceived degradation will be very different.

The next step is therefore to use correspondence curves from the binary parameter. These curves make it possible to obtain a grade of quality from the accumulated length and the number of impulse degradations detected per second. These curves are based on subjective tests. Different curves can be established depending on the type of audio signals (mainly speech or music). Once the estimate is obtained, it is also possible to use a filter simulating the response of a panelist. This makes it possible to take into account the dynamic effect of the votes and the reaction time to the degradations.

The parameter can be summarized according to the following schema Figure 7 .

The main advantage of the parameter is the possibility of making measurements without reference. Another interesting point is the bit rate needed to transfer the reference. This allows to reduce the reference to 1 real number is a rate of 1.5 kbits / s (or even 1 bit in case of thresholding or a bit rate of 47 bits / s) for 1024 samples of signal. It should also be noted that the algorithms are very simple and of reduced complexity, which allows its implementation in parallel with other parameters.

The second of these methods, which does not correspond to the definition of the invention according to claim 1, implements peak detection of the activity.

This parameter, just like the previous one, is based on signal activity. It detects stalls, breaks, cuts of a portion of the audio signal and aberrant samples by looking for peaks in the signal activity.

Thus, this time, only the maxima for blocks of M samples are preserved. It is not interesting to transmit then compare the totality of values of the activity, mainly if the objective is to obtain a method requiring only a reduced reference.

où y(t) est l'activité du signal calculée par le filtre.The parameter is thus obtained at time t by the following formula: $$\mathit{ActTemp}\left(t\right)=\underset{\mathit{k}\in \mathit{M}}{\max }\left(\mathrm{there}\left(t-k\right)\right)$$

where y (t) is the activity of the signal calculated by the filter.

In the case of use with reference, this same calculation is performed on the reference signal and on the degraded signal.

After synchronization of the two streams, the comparison of these maxima of the activity makes it possible to detect the zones where the signal has been disturbed.

To perform this comparison, the ratio between the value measured on the reference and that obtained on the degraded signal allows the detection of degradations. It is possible to detect areas where activity has been greatly reduced by choosing the maximum ratio and its inverse.

où ActTemp_r(t) et ActTemp_d(t) sont respectivement le paramètre calculé sur la référence et sur le signal dégradé.The following formula is used: $$\mathrm{d}\left(\mathrm{t}\right)=\max \left(\frac{{\mathrm{ActTemp}}_{\mathrm{d}}\left(\mathrm{t}\right)}{{\mathrm{ActTemp}}_{\mathrm{r}}\left(\mathrm{t}\right)},\frac{{\mathrm{ActTemp}}_{\mathrm{r}}\left(\mathrm{t}\right)}{{\mathrm{ActTemp}}_{\mathrm{d}}\left(\mathrm{t}\right)}\right)$$

where ActTemp _r (t) and ActTemp _d (t) are respectively the parameter calculated on the reference and on the degraded signal.

In the case where the reference is not available, it is possible to use thresholding to detect whether the parameter is greater than a threshold S ', which indicates the presence of impairments. To avoid false detections due to impulse signals (attacks, percussions, ...), the threshold must have a fairly large value, which may lead to non-detections.

As in the previous case, the use of the correspondence curves is possible to estimate a perceptual quality. The method consists in integrating the degradations detected by this parameter, to the others found by the preceding parameter for example, and thus obtaining an overall perceptual estimate.

The principle of the parameter is presented in the following diagram Figure 8 .

As with the previous parameter, the advantage of the parameter lies in the possibility of making detections without reference.

The reduced complexity and the low bit rate required for the transport of the reference, limited to 1 value, ie a bit rate of 1.5 kbit / s (or even 1 bit in the case of thresholding, ie a bit rate of 47 bits / s) for 1024 sampled signal samples at 48 kHz, are also interesting points.

The method which follows, which does not correspond to the definition of the invention according to claim 1, implements the study of the minimum of the signal spectrum to locate the degradations.

It is mainly useful for the detection of so-called "impulse" degradations. It is indeed important to note that the majority of impairments introduced during the transmission of an audio signal are of this type. These are very localized in time and very spread in frequency. Thus, by assimilating them to a broadband white noise of very short duration in the signal, it is possible to detect them by analyzing the characteristics of the spectrum.

The first step in calculating these parameters corresponds to the estimation of the spectrum of the signal. For this, the signal is windowed in blocks of N samples (N = 1024 or 2048 for example), with an overlap of N / 2 samples. This makes it possible to have a sufficient temporal resolution and to analyze all the signal, taking into account that the use of the windows strongly attenuates the influence of the edges of these time windows.

This also makes it possible not to penalize too much the calculation time during the implementation. A fast Fourier transformation then makes it possible to go into the frequency domain.

avec x_i les N composantes du spectre X en dB (par calcul de distance).The occurrence of degradation increases the spectrum minimum due to the introduction of broadband white noise in all frequency components of the spectrum. It is this principle which made it possible to develop this parameter, calculated simply according to the formula: $$\mathit{MinSpe}=\min \left({\mathit{x}}_{\mathit{i}}\right)\mathrm{for}\mathrm{1}\le \mathrm{i}\le \mathrm{NOT}$$

with x _i the N components of the X spectrum in dB (by distance calculation).

In the case of use with reference, a simple comparison, after synchronization of the values obtained on the reference and the degraded signal, is generally not sufficient for the detection of degradations. Indeed, the variability of minima obtained with a non-degraded signal is important.

It is thus necessary to make comparisons by blocks of M values according to the following principle: For each block, it is conserved only the maximum of the M minima obtained on the reference. This provides a reference value of the initial noise level for the block. This value is compared to the M minima obtained on the degraded signal.

By keeping only the moments when the minima are increased, it is possible to detect the moments when noise has been added to the signal.

où x_r,i est la i^ème des N composantes du spectre obtenu sur la référence,
x_d,i est la i^ème des N composantes du spectre obtenu sur le signal dégradé,
et min_k le k^ième des M minima du bloc considéré.The distance obtained is thus, for each instant t: $$\mathrm{d}\left(\mathrm{t}\right)=\max \left\{\underset{\mathrm{i}\in \mathrm{NOT}}{\min }\left({\mathrm{x}}_{\mathrm{d},\mathrm{i}}\left(\mathrm{t}\right)\right)-\underset{\mathrm{k}\in \mathrm{M}}{\max }\left[\underset{\mathrm{i}\in \mathrm{NOT}}{{\min }_{\mathrm{k}}}\left({\mathrm{x}}_{\mathrm{r},\mathrm{i}}\left(\mathrm{t}\right)\right)\right],\mathrm{0}\right\}$$

where x _{r, i} is the ^ith of the N components of the spectrum obtained on the reference,
x _{d, i} is the ^ith of the N components of the spectrum obtained on the degraded signal,
and min _k the k ^th of the M minima of the block considered.

If the reference is not available, it is possible to use an average of the minima of the spectrum obtained previously by the algorithm. The rest of the comparison is then done in the same way.

As in the previous cases, the use of correspondence curves is possible by integrating the degradations detected by this parameter to the others and thus obtain a perceptual measurement.

The method can be summarized as follows by the two following diagrams Figure 9 .

Again, the main advantage of these parameters is the ability to make measurements without reference. Another interesting point is the bit rate needed to transfer the reference. This allows to reduce the reference to 1 real number and even 1 integer, ie a rate of at most 1.5 kbits / s for N (for example 1024) signal samples. The reduced complexity of the algorithm is also an asset.

In the following method, which does not correspond to the definition of the invention according to claim 1, in which the Spectral Flattening is analyzed, two parameters, SF ₁ and SF ₂ , make it possible to estimate the "flattening" of the spectrum. hence the term sometimes used of "statistical flattening". They correspond to the study of the shape of the spectrum and its evolution along the sequence studied. When a broadband noise occurs in the signal, a white noise DC component will cause a flattening of the spectrum.

SF ₁ parameter

avec X, le spectre du signal et x_i les composantes du spectre.At the onset of a degradation, the components that had values close to zero, will pass to significant values. The product of the spectrum components will thus greatly increase, while their sum will vary only very little. To exploit this, the estimation parameter of the flattening of the spectrum SF ₁ is calculated according to the following formula: $${\mathrm{<figure-callout\; id="1"\; label="SF"\; filenames="imgf0002.png"\; state="\{\{state\}\}">SF</figure-callout>}}_{\mathrm{1}}=\mathrm{10.}\log \mathrm{10}\left(\frac{\mathrm{Arithmetic\; average}\left(\mathrm{X}\right)}{\mathrm{MoyenneGéométrique}\left(\mathrm{X}\right)}\right)=\mathrm{10.}\log \mathrm{10}\left(\frac{\frac{\mathrm{1}}{\mathrm{NOT}}{\displaystyle \underset{\mathrm{i}=\mathrm{1}}{\overset{\mathrm{NOT}}{\Sigma }}}{\mathrm{x}}_{\mathrm{i}}}{\sqrt[\mathrm{NOT}]{{\displaystyle \underset{\mathrm{i}=\mathrm{1}}{\overset{\mathrm{NOT}}{\Pi }}}{\mathrm{x}}_{\mathrm{i}}}}\right)$$

with X, the spectrum of the signal and x _i the components of the spectrum.

This parameter is calculated in the same way on the reference and on the degraded signal. By comparison it is then possible to estimate the level of white noise inserted, and consequently the damage.

SF ₂ parameter

To calculate this parameter, the statistical flattening coefficient, called "kurtosis" or "concentration" was used. The estimate is made from the central moments of order 2 and 4. They make it possible to estimate the shape of the spectrum with respect to a normal distribution in the statistical sense of the term.

avec moments centrés m_k définis par : $${\mathrm{m}}_{\mathrm{k}}=\frac{{\displaystyle \sum _{\mathrm{i}=\mathrm{1}}^{\mathrm{N}}}{\left({\mathrm{x}}_{\mathrm{i}}-\bar{\mathrm{X}}\right)}^{\mathrm{k}}}{\mathrm{N}}$$

où X est la moyenne arithmétique des N composantes x_i du spectre X en dB.The calculation corresponds to the ratio between the centered moment of order 4 and the centered moment of order 2 (variance) squared of the coefficients of the spectrum. The formula used is as follows: $${\mathrm{<figure-callout\; id="2"\; label="SF"\; filenames="imgf0007.png"\; state="\{\{state\}\}">SF</figure-callout>}}_{\mathrm{2}}=\frac{{\mathrm{m}}_{\mathrm{4}}\left(\mathrm{X}\right)}{{{\mathrm{m}}_{\mathrm{4}}}^{\mathrm{2}}\left(\mathrm{X}\right)}=\frac{{\mathrm{m}}_{\mathrm{4}}\left(\mathrm{<figure-callout\; id="2"\; label="X\; \delta "\; filenames="imgf0007.png"\; state="\{\{state\}\}">X\; </figure-callout>},\right)}{{\mathrm{\delta }}^{}}=\mathrm{NOT}\mathrm{.}\frac{{\displaystyle \underset{\mathrm{i}=\mathrm{1}}{\overset{\mathrm{NOT}}{\Sigma }}}{\left({\mathrm{x}}_{\mathrm{i}}-\tilde{\mathrm{X}}\right)}^{\mathrm{4}}}{{\left({\displaystyle \underset{\mathrm{i}=\mathrm{1}}{\overset{\mathrm{NOT}}{\Sigma }}}{\left({\mathrm{x}}_{\mathrm{i}}-\tilde{\mathrm{X}}\right)}^{\mathrm{2}}\right)}^{\mathrm{2}}}$$

with centered moments m _k defined by: $${\mathrm{m}}_{\mathrm{k}}=\frac{{\displaystyle \underset{\mathrm{i}=\mathrm{1}}{\overset{\mathrm{NOT}}{\Sigma }}}{\left({\mathrm{x}}_{\mathrm{i}}-\tilde{\mathrm{X}}\right)}^{\mathrm{k}}}{\mathrm{NOT}}$$

or X is the arithmetic mean of the N components x _i of the X spectrum in dB.

As for the parameter SF ₁ , the higher the value obtained, the more the signal is concentrated and the less noise there is in the signal. This is calculated on the reference and on the degraded signal. By comparison, the level of white noise inserted is estimated.

• Dans le cas d'une comparaison avec la référence, une simple distance du type différence ou autre est suffisante pour détecter les dégradations. Si aucune référence n'est disponible, il est nécessaire d'effectuer une détection des pics dans la variation des paramètres pour rechercher les dégradations. Cela peut être fait en utilisant la technique, classique en traitement de l'image, de la morphologie mathématique à niveau de gris (érosions et dilatations).

The scheme of the Figure 10 presents the principle (valid for the two parameters above):

• In the case of a comparison with the reference, a simple distance of the difference or other type is sufficient to detect the degradations. If no reference is available, it is necessary to perform peak detection in the variation of parameters to search for degradations. This can be done using the classic technique of image processing, grayscale mathematical morphology (erosions and dilations).

The advantages and limitations of these parameters are identical to those of the preceding parameters: limited necessary flow, without possible reference and use of the correspondence curves to estimate the perceptual importance of the degradations.

In the context of monitoring a digital television broadcast network, the reference audio signal corresponds to the signal at the input of the broadcast network. The reference parameters are calculated on this signal and then transmitted via a specific data channel to the desired measurement point. It is at this point is calculated the same parameters necessary for the comparison for the establishment of measurements with reduced reference. Measurements without reference are also calculated. In the case where the reference parameters are not available (not present, erroneous, ...) these measurements are sufficient to detect the most important errors. The dotted subsystems of the Figure 11 are no longer used.

The measurements obtained without reference and those obtained with reduced reference (in the case where they could be calculated) are used by a model to estimate the importance of the degradation introduced during the diffusion.

• Plusieurs points de mesure peuvent ainsi être établis. Une fois ces estimations de dégradations obtenues, il est aisé de les transmettre vers un centre de surveillance du réseau, ce qui permet d'avoir une vue d'ensemble des performances du réseau.

The scheme of the Figure 11 summarizes this example of realization:

• Several measurement points can thus be established. Once these degradation estimates are obtained, it is easy to transmit them to a monitoring center of the network, which gives an overview of network performance.

The same scheme as above can be used to visualize (with or without reference) the performance of radio broadcasting on the Internet. In this case, the data channel used to carry the reference parameters may be the network itself, as well as returning the estimated ratings to the monitoring center. The reference signal corresponds to the signal sent by the server, and the degraded signal is that decoded at the chosen measurement point. This can for example be used to select the most appropriate server based on the connection location by accessing the data of a monitoring center. The diagram ( Figure 12 ) illustrates this embodiment in the case where the reference parameters are sent by the network and where the notes obtained use a specific transmission channel.

A method according to the invention is applicable whenever it is necessary to identify defects on an audio signal that has been transmitted by any broadcast network (cable, satellite, terrestrial, Internet, DVB, DAB, .. .).

The proposed method exploits two classes of methods: techniques with reduced reference and those without reference. It is particularly interesting when the rate available for the transmission of the reference is limited.

Thus, this invention is applicable for operational purposes for metrology equipment and for supervisory systems of audio signal distribution networks. One of its advantageous characteristics lies in the combination of measurements made with and without reference. Finally, this invention corresponds to the requirements imposed in the quality of service management systems.

Claims (8)

1. A method for qualitatively evaluating a digital audio signal, which calculates in real time and in continuous time in successive time windows, a quality indicator obtained from said digital audio signal, characterized in that said quality indicator consists of a vector associated with each time window, said vector having a dimension at least one hundred times less than the number of audio samples in a time window, said dimension ranging from 1 to 10, preferably from 1 to 5, and more particularly from 2 to 5, and in that the generation of said quality indicator vector implements, at least for the audio signal to evaluate, the following steps :

   a) calculating a temporal activity of the signal in each time window,

   b) calculating a sliding average over N₁ successive values of the temporal activity;

   c) retaining the minimum value from M₁ successive values of the sliding average.
2. The method according to claim 1, characterized in that said quality indicator vector consists of said minimum value.
3. The method according to claim 1, characterized in that said quality indicator vector consists of a binary value resulting from comparing said minimum value with a given threshold.
4. The method according to any one of claims 1 to 3, characterized in that it calculates a quality mark by determining a cumulated time interval, during which said minimum value is less than a given threshold Si and/or by determining the number of times per second when said minimum value is less than a given threshold S'₁.
5. The method according to one of claims 1 to 4, characterized in that said minimum values are generated at the same time for a reference audio signal and for the audio signal to evaluate and in that a quality vector is generated by comparing the corresponding minimum values of the reference audio signal and audio signal to be evaluated.
6. The method according to one of preceding claims, characterized in that said audio signal to be evaluated is an audio signal transmitted digitally.
7. The method according to one of preceding claims, characterized in that said audio signal to be evaluated is a digital audio signal to which digital coding has been applied.
8. The method according to claim 7, characterized in that said digital coding is a bit rate reduction coding.

Images