Robust talker localization in reverberant environment
View Patent ↗A method of locating a talker in a reverberant environment comprises receiving multiple audio signals from a microphone array that include direct path audio signal and reverberation signal components. The direct path audio signal components of the multiple audio signals are detected and are used to weight the multiple audio signals. A position estimate based on the weighted audio signals is then calculated. Periods of speech activity are detected and a final position estimate is generated during the periods of speech activity.
1. A method of locating a talker in a reverberant environment comprising the steps of:
receiving multiple audio signals from a microphone array, said audio signals including direct path audio signal and reverberation signal components;
calculating position estimates of a source of said audio signals based on said audio signals;
rapidly detecting the direction of the direct path audio signal component of said multiple audio signals based on said calculated position estimates;
using the rapidly detected direction to weight the calculated position estimates;
detecting periods of speech activity; and
generating a final position estimate of said source during said periods of speech activity based on the weighted position estimates,
wherein the direction of the direct path audio signal component is detected based on the earliest calculated position estimates,
wherein the direction of the direct path audio signal is detected over a duration equal to approximately 10 to 15 msec.,
wherein the direction of the direct path audio signal component is detected by: (i) storing a succession of calculated position estimates; (ii) counting occurrences of the calculated position estimates during periods of speech activity; and (iii) determining the direction of the direct path audio signal component when a current calculated position estimate occurs more than a threshold number of times, and
wherein calculated position estimates are not weighted when the direction of the direct path audio signal component is not detected within said duration.
2. The method of claim 1 wherein said calculated position estimates are weighted according to:
P
E
=
{
E
S
T
,
if
Energy
[
EST
]
>
k
*
max
{
Energy
[
ED
_
EST
]
}
,
ED
_
EST
,
otherwise
where:
Energy[EST] is the energy of beamformer instances positioned in the direction of the calculated position estimates;
max{Energy[ED_EST]} is the maximum energy of the beamformer instances positioned in the direction of the calculated position estimates over the duration; and
k is the weighting coefficient having a value less than 1.
3. The method of claim 1 further comprising the step of verifying the weighted calculated position estimates thereby to generate said final position estimate.
4. The method of claim 3 wherein said verifying further includes the step of storing a succession of weighted position estimates, counting occurrences of said weighted position estimates during periods of speech activity and within a window and selecting the weighted position estimate that is repeated more than a threshold number of times within said window.
5. The method of claim 4 further comprising the step of discarding weighted position estimates calculated during periods where there is no speech activity.
6. The method of claim 1 wherein the calculated position estimates are based on output energy values of beamformers processing the audio signals received from said microphone array and wherein said weightings are based on accumulated values over a time interval T.
7. A method of locating a talker in a reverberant environment comprising the steps of:
receiving multiple audio signals from a microphone array, said audio signals including direct path audio signal and reverberation signal components;
calculating position estimates of a source of said audio signals based on said audio signals;
rapidly detecting the direction of the direct path audio signal component of said multiple audio signals based on said calculated position estimates;
using the rapidly detected direction to weight the calculated position estimates;
detecting periods of speech activity; and
generating a final position estimate of said source during said periods of speech activity based on the weighted position estimates,
wherein the calculated position estimates are based on output energy values of beamformers processing the audio signals received from said microphone array and wherein said weightings are based on accumulated values over a time interval T, and
wherein said calculated position estimates are weighted according to:
P
E
=
{
E
S
T
,
if
Energy
[
EST
]
>
k
*
max
{
Energy
[
ED
_
EST
]
}
,
ED
_
EST
,
otherwise
where:
Energy[EST] is the energy of beamformer instances positioned in the direction of the calculated position estimates;
max{Energy[ED_EST]} is the maximum energy of the beamformer instances positioned in the direction of the calculated position estimates over the duration; and
k is the weighting coefficient having a value less than 1.
8. The method of claim 7 wherein said time interval T is sufficient so that energy output values are accumulated for a period of time expected to be longer than the path of said reverberation signal components.
9. The method of claim 7 further comprising the step of inhibiting the final position estimate from being changed if no interval of silence separates the weighted position estimates.