EP3381204A1 - Method and device for estimating sound reverberation - Google Patents

Abstract

The invention relates to a method for estimating sound reverberation in a medium (7), which comprises the following steps: (a) a measurement step in which an acoustic signal emitted in the medium is collected; (b) a step of determining the distribution of the acoustic energy decrease rate, during which a distribution of the acoustic energy decrease rate is determined from the acoustic signal collected in step (a); and (c) an estimation step during which a reverberation time and a level of reverberation of the sound in the medium is estimated by regression from a characteristic function of the distribution of the acoustic energy decrease rate determined in step (b).

Description

 Method and apparatus for estimating reverberation

 acoustics DOMAIN OF THE INVENTION

 The present invention relates to methods and devices for estimating acoustic reverberation.

 BACKGROUND OF THE INVENTION

 Estimating the acoustic reverberation of a medium is essential for capturing acoustic signals such as speech in a reverberant medium such as, for example, a room in a building.

 When a sound is emitted and picked up by a microphone in a reverberant environment, the microphone picks up not only the received signal directly, but also reverberant signals in the medium.

 This reverberation is reflected by the impulse response of the medium, from which various known parameters, including the reverberation time, flow. The impulse response can be measured directly by emitting an acoustic pulse in the medium, but this method is cumbersome and difficult to envisage for performing repeated measurements while one or more speakers are speaking in the room.

The reverberation time can be estimated blindly, for example while one or more speakers are speaking. The most common parameter used to represent the reverberation time is the 60 dB RT ₆₀ reverberation time.

 By way of example, the document US2014169575 describes a method for blind estimation of the reverberation time in a room.

However, the reverb time is not representative of the distance between the transmitter and the microphone, which however has a significant impact on the reverb level. The aforementioned known methods of the type do not therefore make it possible to process the acoustic signals sensed satisfactorily.

 OBJECTS AND SUMMARY OF THE INVENTION

 The present invention therefore aims to provide a method for estimating acoustic reverberation, which avoids this disadvantage.

 For this purpose, the invention proposes a method for estimating acoustic reverberation in a medium, comprising the following steps:

 (a) a measurement step in which at least one acoustic signal is sensed in the medium,

(b) an observation step during which an acoustic energy decay rate distribution is determined from the acoustic signal picked up in step (a) and the characteristic function of the decay rate distribution is determined acoustic energy,

 (c) an estimation step in which a characteristic reverberation time and a characteristic reverberation level of the sound in the medium are estimated from data representative of the acoustic energy decay rate distribution determined at step (b), the regression being made with reference to:

 reference characteristic functions representative respectively of several distributions of decay rates of acoustic energy,

 reference characteristic reverberation times corresponding to said reference characteristic functions,

- and reverb levels characteristic of reference corresponding to said reference characteristic functions.

 Thanks to these provisions, and notably because the method of estimation is applied to the distribution of decay rates of acoustic energy, it is possible to determine both a characteristic reverberation time and a reverberation level characteristic of the sound in the middle, reliably. These two parameters make it possible to satisfactorily process the sound signals picked up.

 In various embodiments of the method according to the invention, one or more of the following provisions may also be used:

 during the estimation step (c), a kernel function estimator is used and the characteristic reverberation time and the characteristic reverberation level are simultaneously determined;

 during the estimation step (c), a Nadaraya-Watson estimator is used;

during the estimation step (c), the reverberation level characteristic of the sound in the medium (7) is chosen from the clarity index C _T and the definition index D _T ;

 during the observation step (b), the energy decay rates are determined by calculating the energy Em of the acoustic signal on successive signal frames m, and then calculating a logarithmic ratio between the energy of the signal two successive frames: p (m) = log (- ^ -) (5);

 m-l

the method further comprises a preliminary calibration phase comprising the following steps: (a ') at least one initial step of determination of reference signals in which one determines a plurality of reference acoustic signals corresponding to said reference characteristic reverberation times and said reference characteristic reverberation levels,

 (b ') at least one initial observation step during which, for each reference acoustic signal, an acoustic energy decay rate distribution and the reference characteristic function are determined;

 during said step of determining reference signals, at least a portion of the reference acoustic signals and the characteristic reverberation times and reference characteristic reverberation levels corresponding to said reference acoustic signals are determined by calculation from a predetermined set of impulse responses;

 during said step of determining reference signals, at least a portion of the reference acoustic signals, the characteristic reverberation times and the reference characteristic reverberation levels corresponding to said reference acoustic signals are determined by measurement.

 Moreover, the invention also relates to a device for estimating acoustic reverberation in a medium, comprising:

 (a) measuring means for sensing at least one acoustic signal emitted in the medium,

 (b) means for determining an acoustic energy decay rate distribution from the acoustic signal picked up by the measuring means, and for determining the characteristic function of the acoustic energy decay rate distribution,

(c) means for estimating a reverberation time characteristic and a reverberation level characteristic of sound in the medium from data representative of the acoustic energy decay rate distribution, the regression being made with reference to:

 reference characteristic functions representative respectively of several distributions of decay rates of acoustic energy,

 reference characteristic reverberation times corresponding to said reference characteristic functions,

 and reference characteristic reverberation levels corresponding to said reference characteristic functions.

 BRIEF DESCRIPTION OF THE DRAWINGS

 Other features and advantages of the invention will become apparent from the following description of one of its embodiments, given by way of non-limiting example, with reference to the accompanying drawings.

 On the drawings:

 FIG. 1 is a diagrammatic view illustrating the reverberation of the sound in a room when a subject speaks so that his words are picked up by a device according to one embodiment of the invention, FIG. 2 is a schematic diagram of the device of Figure 1.

 DESCRIPTION DETAILED

 In the different figures, the same references designate identical or similar elements.

The aim of the invention is to estimate the acoustic reverberation of a medium 7, for example a room in a building as shown diagrammatically in FIG. 1, so as to process the acoustic signals picked up by a device An electronic device 1 may be, for example, a telephone in the example shown, or a computer or the like.

 When a sound is emitted in the medium 7, for example by a person 3, this sound propagates to the microphone 2 along various paths 4, either direct or after reflection on one or more walls 5, 6 of the medium 7.

 As represented in FIG. 2, the electronic device 1 may comprise, for example, an electronic central unit 8 such as a processor or the like, connected to the microphone 2 and to various other elements, including for example a loudspeaker 9, a keyboard 10 , a screen 11. The electronic central unit 8 can communicate with an external network 12, for example a telephone network.

 The invention enables the electronic device 1 to measure blindly two characteristic parameters of the reverberation of the medium 7:

- a reverberation time characteristic, such as reverberation time 60 dB R _go?

 and a characteristic reverberation level (for example clarity or definition index, or direct signal index on reverberated signal).

 These parameters can be used to suppress the echo effects or more generally to optimize the sound signals picked up by the microphone 2. The parameters in question are estimated in a repetitive manner, so that the device 1 adapts for example to changes of speakers 3, to displacements of speakers 3, to movements of the device 1 or other objects in the medium 7.

The reverberation time at 60dB RT ₆₀ can be defined by the inverse integration method of Manfred R. Schroeder (New Method of Measuring Reverberation Time, The Journal of the Acoustical Society of Ameri.ca, 37 (3): 409, 1965) by the EDC Energy Decay Curve:

Or :

- h is the impulse response of the medium of length N _h , n is a time index, for example a number of samples obtained by a sampling of constant time steps, n being between 1 and N _h .

 R 60 is the time at the time index n required for EDC (n) to decrease by 60 dB.

Although the reverberation time R ₆ o is the most commonly used, we could estimate another reverberation time characteristic of the medium 7.

The reverberation level is most commonly represented by the clarity index: ^c ' ^{= 1G1} * (¾) ^dB - ^<2>

or by the definition index: D _{t =} 10log ₁₀ (|| f) dB, (3)

or :

- Ν _τ is the number of samples with constant time steps corresponding to the time τ, generally between 0.1 ms and 1 s,

n is a time index between 1 and Ν _τ , representative of a number of samples of constant time step,

 h (n) is the impulse response of medium 7.

These indices have been described in particular by PA Naylor and ND Gaubitch (Speech Dereverberation, Springer, eds edition, 2010). The two most commonly used values of τ are 50 ms and 80 ms, in particular 50 ms (indices C ₅ o and D ₅₀ ), but other durations are possible and more generally other indices reflecting the ratio of the direct sound to the its reverberation could be estimated in the method according to the invention, implemented for example by the aforementioned electronic central unit 8.

 This process comprises the following steps:

(a) an acoustic signal measurement step,

(b) an observation step during which an acoustic energy decay rate distribution is determined from the acoustic signals measured in step (a),

 (c) an estimation step in which it is estimated, by regression from the sound energy decay rate distribution determined in step (b), a characteristic reverberation time and a characteristic reverberation level sound in the middle 7.

 (a) Measurement step:

 During this step, the microphone 2 captures "blind" (that is to say, without prior knowledge of the signals transmitted), an acoustic signal broadcast in the medium 7, for example when the speaker 3 speaks. This signal is sampled and stored in processor 8 or an ancillary memory (not shown).

 (b) Observation stage:

 During this step, an acoustic energy decay rate distribution is determined from the acoustic signal measured in step (a).

For this purpose, the energy envelope of the reverberated signal d _x (n) as first described by Wen et al. (JYC Wen, EAP Habits, and PA Naylor, Blind estimation of time-based reverb on the distribution of signal decay rates, Acoustics, Speech. and Signal Processing, 2008, ICASSP 2008, IEEE International Conference pages 329-332, March 2008).

By performing a calculation on de _ω signal samples frames separated by jumps of R signal samples, a total energy of the frame m can be calculated by the formula:

and then estimate the rate of energy decay by calculating the logarithmic ratio between two successive frames:

λ _χ "p (m) = log (^ -). (5)

Indeed, the envelope power of _x (n) can be expressed by the formula: d _x (n) = ^{y ₂ ί (6)

do ^hn s if A _h = A _s

where A _s and À _h are respectively the rate of energy decay of the emitted anechoic signal and medium 7 (the signal picked up is a convolution of the anechoic signal emitted

(speech) with the impulse response of the medium between the speaker 3 and the microphone 2, n being the temporal index defined above).

Since the sum is dominated by the exponential term corresponding to the largest value of λ, the energy decay rate of the reverberated signal λ _χ can be approximated by:

λ _χ ~ max [A _h , A _s ] (7),

which justifies formula (5) above.

The calculation of p (m) can typically be done on a number M of frames of at least 2000, corresponding to at least 1 min of signal according to the selected analysis parameters. The frames can have an individual duration of 10 to 100 ms, in particular of the order of 32 ms. The frames may overlap each other, for example with a recovery rate of the order of 50% between successive frames. Different values of the energy decay rate p (m) are thus obtained, which have a certain statistical distribution (number of embodiments or probability of realization as a function of the energy decay rate p (m), as explained, for example, in the article by Wen et al., above).

The characteristic function of the energy decay rate distribution is then determined by the following formula (see Audrey Feuerverger and Roman A. Mureika [The empirical characterist ic function and its applications, Ann Statist., 5 (1)): 88-97, 01 1977]):

Φχ (Π = / e ^lfx dF _x {x = E [e ^] (8)

where X is here the aforementioned energy decay rate p (m) estimated for various values of m (formula (5)), F _x the cumulative distribution of X and f is a dimensionless variable generally referred to as angular frequency.

 The characteristic function can be calculated for angular frequencies f ranging, for example, from 0 to 0.4, in increments of 0.001.

(c) Estimation step:

We start from the characteristic function Φρ (f), computed for p / 2 frequencies f (p being an even natural integer), the frequency range f and their sampling being provided so that I Φ _{ρ (m)} (f) I is preferably between 0.1 and 1.

Typically, p can be for example between 256 and 2048. Since the characteristic function is a complex number, it can be represented by a vector of X belonging to R ^p constituting the random input vector x of the estimator used. The random output vector y of the estimator, belonging to R ² , has as component the two estimated parameters, for example (RT ₆₀ , C ₅ o) or (RT ₆ o, D ₅ o) ·

 The estimator used may advantageously be a kernel function estimator, for example a Nadaraya-Watson estimator. Such an estimator has the advantage of simultaneously determining the characteristic reverberation time and the characteristic reverberation level.

 The estimator in question can be determined in advance in an initial calibration phase, where at least one initial step of determination of reference signals (a ') and at least one initial observation step (b') is implemented.

 During the initial step of determining reference signals, a plurality of reference acoustic signals and reference characteristic reverberation times and corresponding reference characteristic reverb levels are determined.

 During the initial observation step, the acoustic energy decay rate distribution and the reference characteristic function for each reference acoustic signal are determined identically or similarly to the observation step ( (b) above.

The acoustic reference signals are generally vocal signals of the number of N and correspond to N cases of different figures (different speakers, different positions, different media 7). N can be several hundred or even thousands. The initial step of determining reference signals can be performed:

 with real new measurements made for example with an electronic device 1 of a given model (in this case, the characteristic reverberation time and the characteristic reverberation level can also be measured);

 and / or with synthetic acoustic signals.

 In the case of real measurements, these will generally not be done in the particular medium where the electronic device 1 will be used, although this case may be considered.

 The aforementioned synthetic acoustic signals can be calculated by convolving pre-recorded impulse responses with pre-recorded speech anechoic signals from different speakers. The prerecorded impulse responses may for example come from impulse response databases, for example from open access databases such as Aachen Impulse Response databases.

(http://www.openairlib.net/auralization .db), MARDY (Wen et al., Evaluation of speech dereverberation algorithms using the Mardy database, Sept. IWAENC 2006, Paris), QueenMary (R. Stewart and M. Sandler In this paper, I would describe the results of the IEEE International Conference on, pages 165-168, March 2010), for example with RT ₆₀ reverberation times ranging from 0.3. s to 8 s and clarity indices C ₅ o from -10 dB to 25 dB. Anechoic speech signals may be signals recorded from various speakers, for example example of different ages and sexes, with for example recording times for example of a few minutes, for example of the order of 5 minutes.

 The energy decay rate distributions can for example be calculated over frames of 10 to 100 ms, in particular of the order of 32 ms. The frames may overlap each other, for example with a recovery rate of the order of 50% between successive frames. The characteristic functions can be calculated for angular frequencies ranging for example from 0 to 0.4, in increments of 0.001.

N yields of the aforementioned x and y vectors are thus obtained, and the Nadaraya-Watson estimator can then be terminated by the formula:

or :

 - xi, yi, i = 1 to N, are the N realizations of the x, y vectors used for the calibration step,

- Κ _λ (χ, Xi) is a kernel function of window λ (λ is a constant also called smoothing parameter),

- x is the unknown input vector (measurement made in the measurement step (a) in order to estimate the vector y by the formula = / (*)).

The kernel function K _\ (x, x ±) is a function of x and Xi as defined in particular by Scholkopf et al. (B. Scholkpf eand AJ Smola, Learning with Kernels, MIT Press, Cambridge, MA, 2001).

In particular, it is possible to use a Gaussian kernel, for example with a window of λ = 5.10 ^-4 (non-limiting example):

The tests carried out show that the process of the invention is more accurate than the methods of the prior art for determining the reverberation time, and it also allows to determine the reverberation level at the same time as the reverberation time, which is a significant gain.

Claims

A method for estimating acoustic reverberation in a medium (7), comprising the steps of:  (a) a measurement step in which at least one acoustic signal emitted in the medium (7) is picked up,  (b) an observation step in which an acoustic energy decay rate distribution is determined from the acoustic signal measured in step (a) and the characteristic function of the decay rate distribution is determined acoustic energy,  (c) an estimation step in which a characteristic reverberation time and a characteristic reverberation level of the sound in the medium (7) are estimated by regression from said characteristic function determined in step (b) , the regression being made with reference to:  reference characteristic functions representative respectively of several distributions of decay rates of acoustic energy,  reference characteristic reverberation times corresponding to said reference characteristic functions,  and reference characteristic reverberation levels corresponding to said reference characteristic functions. The method of claim 1, wherein during the estimation step (c), a kernel function estimator is used and the characteristic reverberation time and the characteristic reverberation level are determined simultaneously. The method of claim 2, wherein in the estimating step (c), a Nadaraya-Watson estimator is used. 4. Method according to any one of the preceding claims, wherein, during the estimation step (c), the reverberation level characteristic of the sound in the medium (7) is chosen from the clarity index C _T and the definition index D _T.  5. Method according to any one of the preceding claims, wherein during the observation step (b), the energy decay rates are determined by calculating the energy Em of the acoustic signal on successive frames. signal and then calculating a logarithmic ratio between the energy of two successive frames: p (m) = log (^ 2-) (5).  The method as claimed in any one of the preceding claims, further comprising a preliminary calibration phase comprising the following steps: (a ') at least one initial step of determining reference signals in which a plurality of acoustic signals of reference corresponding to said reference characteristic reverberation times and said reference characteristic reverberation levels,  (b ') at least one initial observation step during which, for each reference acoustic signal, an acoustic energy decay rate distribution and the reference characteristic function are determined. The method according to claim 6, wherein during said step of determining reference signals, at least a portion of the acoustic reference signals and the characteristic reverberation and reference characteristic reverberation levels corresponding to said reference acoustic signals, from a predetermined set of impulse responses.  The method according to claim 6, wherein during said step of determining reference signals, at least a portion of the reference acoustic signals are determined by measurement, the characteristic reverberation times and the corresponding reference characteristic reverb levels. to said reference acoustic signals.  9. Apparatus for estimating acoustic reverberation in a medium (7), comprising:  (a) measuring means (2) for sensing at least one acoustic signal emitted in the medium (7), (b) means (8) for determining an acoustic energy decay rate distribution from the acoustic signal picked up by the measuring means, and for determining the characteristic function of the acoustic energy decay rate distribution,  (c) means (8) for estimating a characteristic reverberation time and a reverberation level characteristic of sound in the medium from data representative of the acoustic energy decay rate distribution, the regression being made with reference to :  reference characteristic functions representative respectively of several distributions of decay rates of acoustic energy, reference characteristic reverberation times corresponding to said reference characteristic functions, and reference characteristic reverberation levels corresponding to said reference characteristic functions.