A sound signal processing apparatus and method for enhancing a sound signal TECHNICAL FIELD

Generally, the present invention relates to the field of audio signal processing and reproduction. More specifically, the present invention relates to a sound signal processing apparatus and method for enhancing a sound signal.

BACKGROUND

Automatic gain control (AGC) aims at equalizing level variations in recorded speech to ensure a constant level of the speech signal. Analysis of speech signals recorded by microphones reveals that the cause of level variations in recorded speech can be separated into two independent causes, namely intentional speech level variations and unintentional variations, due, for example, to the changes of distance between a speaker and a microphone. Several automatic gain control (AGC) systems have been developed in order to equalize level variations. Currently developed AGC solutions are described in US 8,121 ,835 and "Automatic Spatial Gain Control for an Informed Spatial Filter in Acoustics, Speech and Signal Processing (ICASSP)" Braun, S. et al., E. A. P. (2014)., 2014 IEEE International Conference on (pp. 830-834). They act, however, on both intentional and unintentional signal level fluctuations of the emitted speech energy.

Therefore in order to be able to convey a realistic sound field impression, e.g. for immersive teleconferencing systems, it is of paramount importance to detect the cause of level variations. Doing so would allow for fully equalizing of unintentional variations due to distance fluctuations while preserving intentional (natural) dynamic changes of the speech signals.

The idea of distinguishing between intentional and unintentional signal level variations, and equalizing only the unintentional ones, has recently been investigated from numerous perspectives. One of the solutions that have been proposed is to estimate the talker- microphone distances by acoustic source localization (ASL). Several ASL methods have been developed to equalize level variations arising from distance fluctuations between a source and a microphone for systems with synchronized microphones at known positions and without simultaneously active talkers. Such systems are described in, for example, US 7,924,655 and "Energy-based sound source localization and gain normalization for ad hoc microphone arrays. Acoustics, Speech and Signal Processing", Liu, Z. et al, ICASSP 2007, IEEE International Conference on. Vol. 2. IEEE, 2007.

However, conventional ASL methods exhibit at least one of the following deficiencies. In some conventional ASL methods the microphones have to be synchronized and/or their positions have to be known. Some conventional ASL methods cannot handle

simultaneously active talkers. In some conventional ASL methods the large estimation errors do not allow to equalize close-talk level variations. Some conventional ASL methods are computationally complex. Thus, there is a need for an improved sound signal processing apparatus and method allowing, in particular, for automatic gain control.

SUMMARY It is an object of the invention to provide an improved sound signal processing apparatus and method allowing, in particular, for automatic gain control.

The foregoing and other objects are achieved by the subject matter of the independent claims. Further implementation forms are apparent from the dependent claims, the description and the figures.

According to a first aspect, the invention relates to a sound signal processing apparatus for enhancing a sound signal from a target source. The sound signal processing apparatus comprises a plurality of microphones, wherein each microphone is configured to receive the sound signal from the target source, an estimator configured to estimate a first power measure on the basis of the sound signal from the target source received by a first microphone of the plurality of microphones, a second power measure on the basis of the sound signal from the target source received by at least a second microphone of the plurality of microphones, which is located more distant from the target source than the first microphone, wherein the estimator is further configured to determine a gain factor on the basis of a ratio between the second power measure and the first power measure, and an amplifier configured to apply the gain factor to the sound signal from the target source received by the first microphone.

Thus, an improved sound signal processing apparatus is provided allowing for an automatic gain control by equalizing sound signal level fluctuations due to the distance variations between a target source and a microphone in a computationally efficient manner.

In a first possible implementation form of the sound signal processing apparatus according to the first aspect of the invention, the estimator is further configured to estimate the first power measure on the basis of a plurality of band-limited power measures in a plurality of frequency bands of the sound signal from the target source received by the first microphone and a plurality of probabilities defining the likelihood of the target source being active in the plurality of frequency bands.

Thus, a more robust equalization of sound signal level fluctuations due to the distance variations between the target source and the first microphone is provided.

In a second possible implementation form of the sound signal processing apparatus according to the first aspect of the invention as such or the first implementation form thereof, the estimator is further configured to estimate the first power measure on the basis of the following equation:

% = pj(n) ^■ diag{w] Xi(n), wherein s denotes the first power measure, w denotes a weighting vector defining a plurality of weights for the plurality of frequency bands, x _t denotes a power measure vector defining the plurality of band-limited power measures in the plurality of frequency bands of the sound signal from the target source received by the first microphone for the time instant n and Pi n) denotes a probability vector defining the plurality of probabilities for a time instant n on the basis of the following equation:

[Piin Piin, !) Pi(n, K)] ^T wherein Pi n, k) denotes the probability of the target source being active at the first microphone in a frequency band k at time instant n and wherein K denotes the number of frequency bands. Thus, a a more robust equalization of sound signal level fluctuations due to the distance variations between the target source and the first microphone is provided taking into account likelihoods for the different microphones being active. Depending on the importance of specific frequency bands the plurality of weights allows weighting different frequency bands differently.

The power measure vector defining the plurality of band-limited power measures in the plurality of frequency bands of the sound signal from the target source received by the first microphone during the time instant n can be defined by the following equation: wherein Xi n, k) denotes the intensity of the sound signal from the target source received by the first microphone in a frequency band k at time instant n and wherein K denotes the number of frequency bands.

In a third possible implementation form of the sound signal processing apparatus according to the first or the second implementation form of the first aspect, the estimator is configured to determine the plurality of probabilities defining the likelihood of the target source being active in the plurality of frequency bands for each frequency band on the basis of a difference between the first power measure of the sound signal from the target source received by the first microphone and the second power measure of the sound signal from the target source received by the second microphone.

Thus, a more reliable gain control is provided taking into account the likelihood of the target source being active to better separate the signal from the target source from competing sound sources.

In a fourth possible implementation form of the sound signal processing apparatus according to the second or third implementation form of the first aspect, the estimator is further configured to estimate the first power measure using the plurality of probabilities defined by the following equation: wherein Pi n, k) denotes the probability that the target source is active at the first microphone in a frequency band k at time instant n, Xi n, k) denotes the intensity of the sound signal from the target source received by the first microphone in a frequency band k at time instant n and Xj (n, k) denotes the intensity of the sound signal from the target source received by the y ^' -th microphone in a frequency band k at time instant n.

Thus, a more reliable gain enhancement is provided taking into account the likelihood of the target source being active to better separate the signal from the target source from competing sound sources.

In a fifth possible implementation form of the sound signal processing apparatus according to the first aspect of the invention as such or any one of the first to fourth implementation form thereof, the estimator is further configured to estimate the second power measure on the basis of the sound signal received by the first microphone, the sound signal received by the second microphone and the sound signal received by at least one further microphone of the plurality of microphones. Taking further microphones into account allows obtaining a more reliable estimate for the second power measure and, therefore, an improved value for the gain factor.

In a sixth possible implementation form of the sound signal processing apparatus according to the first aspect of the invention as such or the fifth implementation form thereof, the estimator is further configured to estimate the second power measure further on the basis of a plurality of microphone reliability measures, wherein the microphone reliability measure of each microphone of the plurality of microphones depends on the distance of the microphone to the target source and on the ratio of the sound signal from the target source received by the microphone to noise and/or interference signals.

Weighting each microphone by a reliability measure leads to a further improved estimate of the second power measure and, consequently, the gain factor.

In a seventh possible implementation form of the sound signal processing apparatus according to the first aspect of the invention as such or any one of the first to sixth implementation form thereof, the ratio between the second power measure and the first power measure is weighted by a normalized measure for the probability that the target source is active at the first microphone during a time instant n. Thus, an increased robustness and smoothness of the gain control is provided.

In an eighth possible implementation form of the sound signal processing apparatus according to the first aspect of the invention as such or the seventh implementation form thereof, the probability that the target source is active at the first microphone during a time instant n is based on a plurality of probabilities that the target source is active at the first microphone during a time instant n for a plurality of frequency bands.

Thus, an increased robustness of the estimate of the probability that the target source is active is obtained.

In a ninth possible implementation form of the sound signal processing apparatus according to the first aspect of the invention as such or the eighth implementation form thereof, the estimator is configured to determine the normalized measure for the probability that the target source is active at the first microphone during a time instant n on the basis of the following equation: wherein μ;(η) denotes the normalized measure for the probability that the target source is active at the first microphone during a time instant n, w denotes a weighting vector defining a plurality of weights for the plurality of frequency bands and Pi n) denotes a probability vector defining the plurality of probabilities that the target source is active at the first microphone during a time instant n for a plurality of frequency bands. Thus, an advantageous gain factor smoothing is provided.

In a tenth possible implementation form of the sound signal processing apparatus according to the first aspect of the invention as such or the ninth implementation form thereof, the probability vector Pi n) is defined by the following equation:

[PiinA Piin, Pi(n, K)] ^T wherein Pi n, k) denotes the probability of the target source being active at the first microphone in a frequency band k at time instant n, K denotes the number of frequency bands and the probability Pi n, k) is determined on the basis of the following equation: wherein Xi n, k) denotes the intensity of the sound signal from the target source received by the first microphone in a frequency band k at time instant n and Xj (n, k) denotes the intensity of the sound signal from the target source received by the y ^' -th microphone in a frequency band k at time instant n.

Thus, the probability of the target source being active is efficiently and robustly estimated by considering the intensity of the sound signal. In an eleventh possible implementation form of the sound signal processing apparatus according to the first aspect of the invention as such or any one of the first to tenth implementation form thereof, the amplifier is configured to apply a unity gain factor to the sound signal from the target source received by the first microphone, in case a normalized measure for the probability that the target source is active at the first microphone during a time instant n is smaller than a predefined threshold value.

Thus, a unity gain factor is provided, in case the likelihood of the target source being active is too low. In a twelfth possible implementation form of the sound signal processing apparatus according to the first aspect of the invention as such or any one of the first to eleventh implementation form thereof, the amplifier is configured to apply a unity gain factor to the sound signal from the target source received by the first microphone, in case a normalized measure for the probability that the target source is active at the first microphone during a time instant n is smaller than a predefined probability threshold value, by applying a gating function, a low pass filter and/or a Schmitt trigger to the normalized measure for the probability that the target source is active at the first microphone during a time instant n.

According to a second aspect, the invention relates to a sound signal processing method for enhancing a sound signal form a target source. The sound signal processing method comprises the steps of: receiving the sound signal from the target source by a plurality of microphones; estimating a first power measure on the basis of the sound signal from the target source received by a first microphone of the plurality of microphones and a second power measure on the basis of the sound signal from the target source received by at least a second microphone of the plurality of microphones, which is located more distant from the target source than the first microphone; determining a gain factor on the basis of a ratio between the second power measure and the first power measure; and applying a gain factor to the sound signal from the target source received by the first microphone. In a first possible implementation form of the sound signal processing method according to the second aspect of the invention, the estimation step comprises estimating the first power measure on the basis of a plurality of band-limited power measures in a plurality of frequency bands of the sound signal from the target source received by the first microphone and a plurality of probabilities defining the likelihood of the target source being active in the plurality of frequency bands.

In a second possible implementation form of the sound signal processing method according to the second aspect of the invention as such or the first implementation form thereof, the estimation step comprises estimating the first power measure on the basis of the following equation:

% = pj(n) ^■ diag{w] Xi(n), wherein s denotes the first power measure, w denotes a weighting vector defining a plurality of weights for the plurality of frequency bands, x _t denotes a power measure vector defining the plurality of band-limited power measures in the plurality of frequency bands of the sound signal from the target source received by the first microphone for the time instant n and Pi n) denotes a probability vector defining the plurality of probabilities for a time instant n on the basis of the following equation:

_Pi (n) = [Ρ , 1), Ρ , 2) Ρ , Κ)] ^Τ , wherein Pi n, k) denotes the probability of the target source being active at the first microphone in a frequency band k at time instant n and wherein K denotes the number of frequency bands. The power measure vector defining the plurality of band-limited power measures in the plurality of frequency bands of the sound signal from the target source received by the first microphone during the time instant n can be defined by the following equation: wherein Xi n, k) denotes the intensity of the sound signal from the target source received by the first microphone in a frequency band k at time instant n and wherein K denotes the number of frequency bands.

In a third possible implementation form of the sound signal processing method according to the first or the second implementation form of the second aspect, the plurality of probabilities defining the likelihood of the target source being active in the plurality of frequency bands are determined for each frequency band on the basis of a difference between the first power measure of the sound signal from the target source received by the first microphone and the second power measure of the sound signal from the target source received by the second microphone.

In a fourth possible implementation form of the sound signal processing method according to the second or third implementation form of the second aspect, the estimation step comprises estimating the first power measure using the plurality of probabilities defined by the following equation: wherein Pi n, k) denotes the probability that the target source is active at the first microphone in a frequency band k at time instant n, Xi n, k) denotes the intensity of the sound signal from the target source received by the first microphone in a frequency band k at time instant n and Xj (n, k) denotes the intensity of the sound signal from the target source received by the y ^' -th microphone in a frequency band k at time instant n.

In a fifth possible implementation form of the sound signal processing method according to the second aspect of the invention as such or any one of the first to fourth

implementation form thereof, the estimation step comprises estimating the second power measure on the basis of the sound signal received by the first microphone, the sound signal received by the second microphone and the sound signal received by at least one further microphone of the plurality of microphones.

In a sixth possible implementation form of the sound signal processing method according to the second aspect of the invention as such or the fifth implementation form thereof, the estimation step comprises estimating the second power measure further on the basis of a plurality of microphone reliability measures, wherein the microphone reliability measure of each microphone of the plurality of microphones depends on the distance of the microphone to the target source and on the ratio of the sound signal from the target source received by the microphone to noise and/or interference signals.

In a seventh possible implementation form of the sound signal processing method according to the second aspect of the invention as such or any one of the first to sixth implementation form thereof, the step of determining the gain factor on the basis of the ratio between the second power measure and the first power measure comprises weighting the ratio by a normalized measure for the probability that the target source is active at the first microphone during a time instant n.

In an eighth possible implementation form of the sound signal processing method according to the second aspect of the invention as such or the seventh implementation form thereof, the probability that the target source is active at the first microphone during a time instant n is based on a plurality of probabilities that the target source is active at the first microphone during a time instant n for a plurality of frequency bands. In a ninth possible implementation form of the sound signal processing method according to the second aspect of the invention as such or the eighth implementation form thereof, the normalized measure for the probability that the target source is active at the first microphone during a time instant n is determined on the basis of the following equation: μ = ^ψ^, wherein μ; (η) denotes the normalized measure for the probability that the target source is active at the first microphone during a time instant n, w denotes a weighting vector defining a plurality of weights for the plurality of frequency bands and Pi n) denotes a probability vector defining the plurality of probabilities that the target source is active at the first microphone during a time instant n for a plurality of frequency bands. In a tenth possible implementation form of the sound signal processing method according to the second aspect of the invention as such or the ninth implementation form thereof, the probability vector Pi n) is defined by the following equation:

Pi ri) = [PiinA Piin, Pi(n, K)] ^T , wherein ^(n, k) denotes the probability of the target source being active at the first microphone in a frequency band k at time instant n, K denotes the number of frequency bands and the probability Pi n, k) is determined on the basis of the following equation: wherein Xi n, k) denotes the intensity of the sound signal from the target source received by the first microphone in a frequency band k at time instant n and X _j (n, k) denotes the intensity of the sound signal from the target source received by the y ^' -th microphone in a frequency band k at time instant n.

In an eleventh possible implementation form of the sound signal processing method according to the second aspect of the invention as such or any one of the first to tenth implementation form thereof, the step of applying a gain factor to the sound signal comprises applying a unity gain factor to the sound signal from the target source received by the first microphone, in case a normalized measure for the probability that the target source is active at the first microphone during a time instant n is smaller than a predefined threshold value.

In a twelfth possible implementation form of the sound signal processing method according to the second aspect of the invention as such or any one of the first to eleventh implementation form thereof, the step of applying a gain factor to the sound signal comprises applying a unity gain factor to the sound signal from the target source received by the first microphone, in case a normalized measure for the probability that the target source is active at the first microphone during a time instant n is smaller than a predefined probability threshold value, by applying a gating function, a low pass filter and/or a Schmitt trigger to the normalized measure for the probability that the target source is active at the first microphone during a time instant n. The sound signal processing method according to the second aspect of the invention can be performed by the sound signal processing apparatus according to the first aspect of the invention. Further features of the sound signal processing method according to the second aspect of the invention result directly from the functionality of the sound signal processing apparatus according to the first aspect of the invention and its different implementation forms and vice versa.

According to a third aspect the invention relates to a computer program comprising program code for performing the sound signal processing method according to the second aspect of the invention or any of its implementation forms when executed on a computer.

The invention can be implemented in hardware and/or software.

BRIEF DESCRIPTION OF THE DRAWINGS

Further embodiments of the invention will be described with respect to the following figures, wherein:

Fig. 1 shows a schematic diagram of a sound signal processing apparatus for enhancing a sound signal from a target source according to an embodiment;

Fig. 2 shows a schematic diagram illustrating an exemplary scenario including a sound signal processing apparatus according to an embodiment with four microphones and one target source with varying target source-microphone distances;

Fig. 3 shows a schematic diagram illustrating a decomposition of the signal energy variations as received by a microphone of a sound signal processing apparatus according to an embodiment into intentional level variations and unintentional level variations; Fig. 4 shows a schematic diagram illustrating short time fluctuations of the signal energy as received by microphones of a sound signal processing apparatus according to an embodiment located at different distance from the target source;

Fig. 5 shows a schematic diagram illustrating power measure ratio between a microphone close to the target source and a distant microphone of a sound signal processing apparatus according to an embodiment; Fig. 6 shows a schematic diagram illustrating an exemplary scenario including a sound signal processing apparatus according to an embodiment with three microphones and three different target sources; Fig. 7 shows a schematic diagram illustrating gain factor estimation by an estimator of a sound signal processing apparatus according to an embodiment;

Fig. 8 shows a schematic diagram illustrating a microphone reliability distribution in an exemplary scenario including a sound signal processing apparatus according to an embodiment with six microphones and two active target sources;

Fig. 9 shows a schematic diagram illustrating a gain factor estimation stage with a subsequent gain factor update stage of an estimator of a sound signal processing apparatus according to an embodiment;

Fig. 10 shows a schematic diagram illustrating an estimator and an additional close-talker detection stage of a sound signal processing apparatus according to an embodiment;

Fig. 1 1 shows a schematic diagram illustrating a sound signal processing apparatus according to an embodiment; and

Fig. 12 shows a schematic diagram of a signal processing method for enhancing a sound signal from a target source according to an embodiment. In the various figures, identical reference signs will be used for identical or at least functionally equivalent features.

DETAILED DESCRIPTION OF EMBODIMENTS In the following description, reference is made to the accompanying drawings, which form part of the disclosure, and in which are shown, by way of illustration, specific aspects in which the present invention may be placed. It is understood that other aspects may be utilized and structural or logical changes may be made without departing from the scope of the present invention. The following detailed description, therefore, is not to be taken in a limiting sense, as the scope of the present invention is defined be the appended claims. For instance, it is understood that a disclosure in connection with a described method may also hold true for a corresponding device or system configured to perform the method and vice versa. For example, if a specific method step is described, a corresponding device may include a unit to perform the described method step, even if such unit is not explicitly described or illustrated in the figures. Further, it is understood that the features of the various exemplary aspects described herein may be combined with each other, unless specifically noted otherwise.

Figure 1 shows a schematic diagram of a sound signal processing apparatus 100 for enhancing or equalizing a sound signal from a target source, for instance, a speaker. The sound signal processing apparatus 100 comprises a first microphone 101 a configured to receive the sound signal from the target source and a second microphone 101 b configured to receive the sound signal from the target source. As illustrated in figure 1 , the second microphone 101 b is located more distant from the target source than the first microphone 101 a.

The sound signal processing apparatus 100 further comprises an estimator 103 configured to estimate a first power measure on the basis of the sound signal from the target source received by the first microphone 101 a and a second power measure on the basis of the sound signal from the target source received by the second microphone 101 b. The term "power measure" used herein is to be understood as any measure that allows quantifying the strength of the sound signal received from the target source, such as a measure of the intensity, energy and/or power of the sound signal. A power measure may be determined, for instance, on the basis of the squared magnitude, magnitude or root- mean-square of the sound signal. Moreover, determining the power measure may involve a temporal averaging operation.

The estimator 103 is further configured to determine a gain factor on the basis of a ratio between the second power measure and the first power measure.

The sound signal processing apparatus 100 further comprises an amplifier 105 configured to apply the gain factor to the sound signal from the target source received by the first microphone 101 a. Figure 2 shows a schematic diagram illustrating an exemplary scenario including a sound signal processing apparatus 100 according to an embodiment with four microphones 101 a-d and a target source with varying source-microphone distance. By way of example, the first microphone 101 a is the closest microphone to the target source and the second microphone 101 d has the largest distance from the target source. It has been found that the short time variations of the signals, recorded by each microphone 101 a-d can be decomposed into intentional signal variations, for example the changes of intonation or tone of the voice, and unintentional signal variations, for example due to the distance fluctuations between the target source and the microphones 101 a-d.

Figure 3 shows a schematic diagram illustrating how the short-time signal energy fluctuations of the sound signal received by the first microphone 101 a can be

decomposed into short-time energy fluctuations of the sound signal and a time-varying gain due to distance variations between the target source and the first microphone 101 a. By way of example, in figure 3 and some of the following figures the first power measure and the second power measure are based on the energy of the sound signal.

Figure 4 shows a schematic diagram illustrating the short-time signal fluctuations of the signal energy as observed by all four microphones 101 a-d of the sound signal processing apparatus shown in figure 2, which are located at different distances from the target source. The respective solid lines represent the recorded speech energy and the respective dashed lines represent the time-varying gain due to distance variations between the target source and the respective microphone 101 a-d. As can be taken from figure 4, the energy (or alternatively a different power measure) of the sound signal received by the different microphones 101 a-d decreases with increasing distance from the target source. The signal energy received by the first microphone 101 a is the highest and the signal energy captured at the microphone 101 d is the lowest. However, the intentional short term fluctuations (shown in solid lines) of the signals are the same.

On the other hand, the unintentional short term fluctuations, e.g. due to distance variations, of the sound signals received by the different microphones 101 a-d decrease with increasing distance from the target source, because the relative change or fluctuation of the distance, i.e. the ratio of the change (or difference) of the distance between the target source and a respective microphone 101 a-d compared to the overall distance between the target source and the respective microphone 101 a-d, decreases with increasing distance between the target source and the respective microphone 101 a-d. Accordingly, as can be taken from figure 4, the unintentional short term fluctuations, e.g. due to distance variations (the target source first moves farther away from the microphones 101 a-d and afterwards moves closer again), of the speech signals received by the different microphones 101 a-d are strongest at the first microphone 101 a, which is closest to the target source, and smaller at the other microphones 101 b-d, which are located at a greater distance from the target source.

Since the signal at the first microphone 101 a close to the target source and the signals at the other microphones 101 b-d more distant from the target source have more or less identical intentional fluctuations and differ only in unintentional (distance) fluctuations, embodiments of the invention equalize unintentional level variations by applying a gain factor based on a ratio between a first power measure of the signal at the first microphone 101 a and a second power measure of the signal at one of the more distant microphones 101 b-d.

Figure 5 shows a schematic diagram illustrating the power measure ratio between the first microphone 101 a close to the target source and the more distant microphone 101 d.

Although in theory a temporal synchronization is beneficial to compensate for the time difference between the arrival of the sound signal at the first microphone 101 a close to the target source and at the more distant microphone 101 d, it has been found out that in practice this does not affect the result because the typical delay is very small (e.g., -100 samples), whereas the short-time signal power measure is estimated over a longer period (e.g., -2000 samples).

Embodiments of the present invention allow for robust and efficient estimation of the first power measure, e.g. close short-time speech energies, and the second power measure, e.g. distant short-time speech energies, even for challenging multi-talker scenarios. Figure 6 shows an exemplary scenario including a sound signal processing apparatus 100 with three microphones 101 a-c and three target sources (referred to as t1 , t2 and t3 in figure 6). In this exemplary scenario there can be an active target source close to every microphone 101 a-c. It can be assumed that the short-time signal energy at each microphone 101 a-c can be decomposed into sound energies corresponding to the different target sources and a noise plus interference term, e.g. m ₁ = s _n + s ₂₁ + s ₃₁ + n , where m ₁ is the total short-time signal energy observed by the first microphone 101 a, s _i; is the short-time speech energy (i.e. power measure) corresponding to target source i received by microphone j, and n is the total short-time energy of microphone self-noise, diffuse noise and interfering sources. The first power measure Sjj (n) (wherein n denotes the time instant) is estimated by the estimator 103 of the sound signal processing apparatus 100 on the basis of the sound signal from the i-th target source received by the i-th microphone 101 a-c. The second power measure s (n) is estimated on the basis of the sound signal from the i-th target source received by another microphone 101 a-c, which is located more distant from the i- th target source than the i-th microphone 101 a-c.

Figure 7 shows a schematic diagram illustrating the estimation of the gain factor ^ (n) for the i-th microphone 101 a-c by the estimator 103 of the sound signal processing apparatus 100 according to an embodiment. The gain factor ^ (n) calculation is based on the power measures Sjj (n) and s (n) . The estimator 103 of the sound signal processing apparatus 100 according to an embodiment is configured to determine the gain factor ^ (n) for the i- th microphone 101 a-c as a ratio between the second power measure s (n) and the first power measure Sjj (n).

As the gain factor ^ ( ) determined by the estimator 103 of the sound signal processing apparatus depends on the second power measure s (n), different embodiments of the invention provide different ways for selecting the "distant microphone" and estimating the second power measure s (n) . The selected distant microphone should be far away from the target source to reduce the impact of unintentional distance variations, while still allowing for accurate estimation of the second power measure s (n) in order to be able to obtain a meaningful reference signal showing the intentional fluctuations.

Thus, in the case, where a dedicated distant microphone is available, this microphone should be used by the estimator 103 of a sound signal processing apparatus 100 for estimating the second power measure s (n) . However, in other embodiments, where, for instance, such a dedicated distant microphone is not available, the estimator 103 of the sound signal processing apparatus 100 is configured to use (possibly weighted) statistical estimators based on the estimated power measures of all microphones to estimate the second power measure s (n), as will be described in more detail further below.

Determining the second power measure s (n) on the basis of such statistical estimators allows, for instance, to account for abrupt gain changes when the positions of target sources and/or microphones change and to make the estimation of the second power measure more robust, as the estimation of the second power measure based on a single distant microphone is prone to errors under noisy or reverberant conditions. In an embodiment, the statistical estimators for the second power measure sf are the mean or median value over all estimated power measures corresponding to one target source. However, the mean value can be prone to errors due to outliers, whereas using the median value can lead to abrupt changes of the second power measure sf. In order to obtain a smooth estimate for the second power measure sf embodiments of the invention use the mean value between two percentile measures, as will be described in more detail further below. When a large number of microphones with reasonable spatial distribution is available, these statistical estimators are not corrupted by some microphones that are too close to the desired target source or too far away. However, if only a few microphones are available, embodiments of the invention can improve the robustness of the estimation of the second power measure by using weighted statistical estimators. In an embodiment, for each sf estimation a reliability coefficient for each microphone (also referred to as spot reliability) is estimated and the power measures from microphones with low spot reliability contribute less to the value of the second power measure estimated by the estimator 103 than the power measures from microphones with a larger spot reliability.

Even if, as in the example shown in figure 6, there is a plurality of target sources, not all of them are necessarily active at the same time. Therefore, in an embodiment, the estimator 103 is further configured to estimate the first power measure on the basis of a plurality of band-limited power measures in a plurality of frequency bands of the sound signal from the target source received by the first microphone 101 a and a plurality of probabilities defining the likelihood of the target source being active in the plurality of frequency bands.

In an embodiment, the estimator 103 is further configured to estimate the first power measure su on the basis of the following equation:

% = ΡΪ i ) ^■ diag{w] ^■ Xi(ri), (1 ) wherein s _u denotes the first power measure, w denotes a weighting vector defining a plurality of weights for the plurality of frequency bands, x _t denotes a power measure vector defining the plurality of band-limited power measures in the plurality of frequency bands of the sound signal from the target source received by the first microphone 101 a for the time instant n and Pi n) denotes a probability vector defining the plurality of probabilities for a time instant n on the basis of the following equation:

[PiinAX Piin, Pi(n, K)] ^T (2) wherein Pi n, k) denotes the probability of the target source being active at the first microphone 101 a-c in a frequency band k at time instant n and wherein K denotes the number of frequency bands. The probabilities Pi n, k) are herein also referred to as "close-talker probabilities".

In an embodiment, the power measure vector _; (n) defining the plurality of band-limited power measures in the plurality of frequency bands of the sound signal from the target source received by the first microphone 101 a-c during the time instant n is defined by the following equation: wherein Xi n, k) denotes the intensity of the sound signal from the target source received by the first microphone 101 a in a frequency band k at time instant n and wherein K denotes the number of frequency bands. In an embodiment, Xj (n, k) can be the generally complex valued output of a filterbank at time instant n and in frequency subband k (e.g. obtained by a short-time Fourier transform).

In an embodiment, the estimator 103 is configured to determine the plurality of probabilities defining the likelihood of the target source being active in the plurality of frequency bands for each frequency band on the basis of a difference between the first power measure of the sound signal from the target source received by the first microphone 101 a and the second power measure of the sound signal from the target source received by the second microphone 101 b,c.

In an embodiment, the estimator 103 is further configured to estimate the first power measure su using the plurality of probabilities defined by the following equation: wherein P^n, k denotes the probability that the target source is active at the first microphone 101 a in a frequency band k at time instant n, Xi n, k) denotes the intensity of the sound signal from the target source received by the first microphone 101 a in a frequency band k at time instant n and Xj (n, k) denotes the intensity of the sound signal from the target source received by the y ^' -th microphone 101 b,c in a frequency band k at time instant n.

Assuming approximate spectral disjointness of the sound signals from the target sources, i.e. each time frequency slot is mostly dominated by a single target source, the probability Pi n, k) as defined before can be used to separate the power measure of the desired i-th target source from the power measures of competing target sources in above equation 4 defining the first power measure s^ . The weighting vector w in above equation 1 can be used to emphasize certain frequency ranges.

As mentioned before, the second power measure s (n) can be calculated by using the estimated power measure of one distant microphone 101 b,c, which is located more distant from the i-th target source than the i-th microphone 101 a. Therefore, as already mentioned before, if a dedicated single microphone or a microphone array with sufficient distance to all target sources is available, the second power measure s (n) can be estimated using these microphones.

In an embodiment, the estimator 103 is further configured to estimate the second power measure on the basis of the sound signal received by the first microphone 101 a, the sound signal received by the second microphone 101 b,c and the sound signal received by at least one further microphone 101 b,c of the plurality of microphones 101 a-c.

In an embodiment, the estimator 103 is configured to estimate the second power measure s (n) further on the basis of a plurality of microphone reliability measures, wherein the microphone reliability measure of each microphone 101 a-c of the plurality of microphones 101 a-c depends on the distance of the microphone 101 a-c to the target source and on the ratio of the strength of the sound signal from the target source to the strength of noise and/or interference signals. In an embodiment, the second power measure s (n) can be determined by the estimator 103 on the basis of the following equation: s (n) = G{s _i: (n), _i: (n)}, (5) wherein s _i: (n) denotes the vector of the power measures corresponding to the i-th target source, defined by the following equation: Si- n) = [s _il (n),s _i2 (n),...,s _iM (n)] ^T , (6) and λί.(η) denotes the vector of microphone reliability measures, defined by the following equation:

A-i-in) = _i:L (n), i _i2 (n), ...,λ _ίΜ (η)] ^τ , (7) wherein Λ. _ί; (η) denotes the microphone reliability measure of the y ^' -th microphone 101a-c with respect to the i-th target source and M denotes the total number of microphones 101a-c.

In an embodiment, the function G{-} is given by the following equation:

G{s _i: (n), _i: (n)}= ⁶ » _t (n) _t (n ^)} +e _t (n) _t (n)} _{| (8)} wherein 8 _R {v \ q] denotes the R-t weighted percentile of the entries in vector v given the weights in vector q.

Using the microphone reliability measures Λ. _ί; (η) of the y ^' -th microphone 101a-c with respect to the i-th target source allows to give more weight to more reliable microphones 101a-c. For example, the most reliable estimates of the power measure s (n) can be expected from microphones 101a-c that are far away from the i-th target source and still exhibit reasonable signal to noise and interference ratios. That is, the microphone reliability measure Α _ί; (η) is low if

Sijf i) /sjj n) < δ ₁ or ¾(n)/s _i; -(n) < δ ₂ , (9) wherein δ and δ ₂ are empirical constants. Values close to 1 may be chosen for δ , values between 4 and 10 for δ ₂ .

Figure 8 shows an exemplary distribution of microphone reliability measures Λ. _ί; (η) for a scenario with six microphones 101 a-f and six target sources, two of which are active. In figure 8 dark areas correspond to low spot reliabilities, whereas bright areas correspond to high spot reliabilities. As described before, in an embodiment, the one (or more) gain factors g;(n) are estimated under the assumption that each microphone 101 a-c corresponds to an active signal target source, located close to the respective microphone. However, when the signal of the target source at the i-th microphone 101 a-f is low, the corresponding gain factor should adapt smoothly. Therefore, in an embodiment, the estimator 103 of the sound signal processing apparatus 100 is further configured to determine the gain factor on the basis of the ratio between the second power measure and the first power measure by weighting the ratio with a normalized measure for the probability that the target source is active at the first microphone 101 a-f during a time instant n, which is herein referred to as the frame reliability μ _ί (η).

In an embodiment, the probability that the target source is active at the first microphone 101 a during a time instant n is based on a plurality of probabilities that the target source is active at the first microphone 101 a during a time instant n for a plurality of frequency bands.

In an embodiment, the estimator 103 is further configured to determine the frame reliability μ _ί (η), i.e. the normalized measure for the probability that the target source is active at the first microphone 101 a-c during a time instant n, on the basis of the following equation: w ^T pi(n)

i 00 = (10) wherein w denotes a weighting vector defining a plurality of weights for the plurality of frequency bands and Pi n) denotes a probability vector defining the plurality of probabilities that the target source is active at the first microphone 101 a during a time instant n for a plurality of frequency bands. In an embodiment, the probability vector Pi n) is defined by equations (3) and (4) described above.

The weighting vector w can be chosen to emphasize important frequency regions and can be written as follows: w = [w(l), w(2) w(K) ] ^T (1 1 ) wherein the different weights w(/c) lie between zero and one, i.e. 0 < w(/c) < 1. In an embodiment, the frame reliability μ;(η) can serve as the adaptation step size used in a gain factor update stage 103b of the estimator 103 of the sound signal processing apparatus 100, as shown in figure 9. In addition to the gain factor update stage 103b the estimator shown in figure 9 comprises a gain factor estimation stage 103a. Figure 9 serves as a summary of the components of the gain factor estimation stage 103a and the gain factor update stage 103b of the estimator 103 already described above.

In case a target source is inactive, an estimation of a gain factor on the basis of such a target source can be erroneous. Therefore, in an embodiment, the amplifier 105 is configured to apply a unity gain factor to the sound signal from the target source received by the first microphone 101 a, in case the frame reliability μ^η), i.e. the normalized measure for the probability that the target source is active at the first microphone 101 a during a time instant n, is smaller than a predefined threshold value. A corresponding embodiment is shown in figure 10.

Further to the components shown in figure 9 the sound processing apparatus 100 shown in figure 10 comprises a close-talker detection stage 107. The close-talker detection stage 107, which can be implemented as part of the amplifier 105, is configured to check whether the frame reliability μ^η), i.e. the normalized measure for the probability that the target source is active at the first microphone 101 a during a time instant n, is smaller than a predefined threshold value. If this is the case, the amplifier 105 can be switched to a unity or inactive gain 108. In order to perform its function, the close-talker detection stage 107 can comprise a gating function 107a, a low pass filter 107b and/or a Schmitt trigger 107c.

Figure 1 1 shows a schematic diagram illustrating a sound signal processing apparatus 100 according to an embodiment. The sound signal processing apparatus 100 comprises a plurality of microphones 101 a-f, a multi-channel filterbank 102 for spectrally dividing the target source sound signals, an estimator 103 and an amplifier 105. The estimator 103 and the amplifier 105 can be configured in an identical or similar way as in the previously described embodiments.

Figure 12 shows a schematic diagram of a sound signal processing method 1200 for enhancing a sound signal from a target source. The sound signal processing method 1200 comprises the steps of receiving 1201 the sound signal from the target source by a plurality of microphones101 a-c, estimating 1203 a first power measure on the basis of the sound signal from the target source received by a first microphone 101 a of the plurality of microphones 101 a-c and a second power measure on the basis of the sound signal from the target source received by at least a second microphone 101 b,c of the plurality of microphones 101 a-c, which is located more distant from the target source than the first microphone 101 a; determining 1205 a gain factor on the basis of a ratio between the second power measure and the first power measure; and applying 1205 the gain factor to the sound signal from the target source received by the first microphone 101 a.

Embodiments of the invention realize different advantages. The sound signal processing apparatus 100 and the sound signal processing method 1200 provide means to equalize unintentional signal variations, e.g. due to the variations of distance between a target source and a microphone, while preserving the intentional signal variations.

Embodiments of the invention can be applied in different scenarios, for example with different number of target sources and microphones, unknown number of target sources, unknown target source positions, unknown microphone positions or non-synchronized microphones.

Embodiments of the invention allow for efficient gain factor estimation and update according to the activity and/or position of target sources and microphones. The described invention can be implemented, for example, in a multi-microphone conference scenario, teleconferencing with multiple talkers in the same room, etc. The invention allows for equalizing signal level variations due to the variation of talker positions while preserving the natural (dynamic) changes of the speech signal. While a particular feature or aspect of the disclosure may have been disclosed with respect to only one of several implementations or embodiments, such feature or aspect may be combined with one or more other features or aspects of the other implementations or embodiments as may be desired and advantageous for any given or particular application. Furthermore, to the extent that the terms "include", "have", "with", or other variants thereof are used in either the detailed description or the claims, such terms are intended to be inclusive in a manner similar to the term "comprise". Also, the terms "exemplary", "for example" and "e.g." are merely meant as an example, rather than the best or optimal. The terms "coupled" and "connected", along with derivatives may have been used. It should be understood that these terms may have been used to indicate that two elements cooperate or interact with each other regardless whether they are in direct physical or electrical contact, or they are not in direct contact with each other.

Although specific aspects have been illustrated and described herein, it will be

appreciated by those of ordinary skill in the art that a variety of alternate and/or equivalent implementations may be substituted for the specific aspects shown and described without departing from the scope of the present disclosure. This application is intended to cover any adaptations or variations of the specific aspects discussed herein.

Although the elements in the following claims are recited in a particular sequence with corresponding labeling, unless the claim recitations otherwise imply a particular sequence for implementing some or all of those elements, those elements are not necessarily intended to be limited to being implemented in that particular sequence.

Many alternatives, modifications, and variations will be apparent to those skilled in the art in light of the above teachings. Of course, those skilled in the art readily recognize that there are numerous applications of the invention beyond those described herein. While the present invention has been described with reference to one or more particular embodiments, those skilled in the art recognize that many changes may be made thereto without departing from the scope of the present invention. It is therefore to be understood that within the scope of the appended claims and their equivalents, the invention may be practiced otherwise than as specifically described herein.