Localization algorithm for conferencing
View Patent ↗A conferencing device and method of talker localization for steering a plurality of sensors arranged in sectors around a housing to a current active direction beam. A beamformer is implemented for determining a loudest sector based on signals from the sensors. In the event the loudest sector persists for a first duration then forming signal beams from sensor signals in the loudest sector and a plurality of adjacent sectors over a further duration to create respective signal beam envelopes. Then selecting a highest energy of the beam envelopes to be a loudest non-active beam and comparing the loudest non-active beam to the current active direction beam. If the highest energy one of the beams is greater than the energy of the current active direction beam by a threshold, then selecting the highest energy one of the beams to be the current active direction beam and directing the sensors thereto.
1. A method of talker localization in a conferencing device for steering a plurality of sensors arranged in sectors around a housing of the device, the method comprising:
initially having sensors steered to a current active direction;
determining a loudest sector by averaging signals from the sensors;
in the event that the loudest sector persists for a first duration:
forming at least three signal beams from sensor signals over a further duration, such that the first signal beam points at a sector including the current active direction, the second signal beam points at the loudest sector and the third and any subsequent signal beams point at one or more sectors adjacent to the loudest sector, but wherein not all sectors are pointed at by a corresponding signal beam, calculating respective beam envelopes of the at least three signal beams to estimate their energies; and
selecting, from the signal beams pointing at the loudest sector and the one or more adjacent sectors, a highest energy one of the signal beams to be a loudest non-active beam;
comparing an energy of the loudest non-active beam to an energy of the signal beam pointing at the sector including the current active direction; and
if the energy of the loudest non-active beam is greater than the energy of the signal beam pointing at the sector including the current active direction:
selecting the direction associated with the loudest non-active beam to be the current active direction, and steering the sensors to the current active direction.
2. The method of claim 1 , further comprising filtering the signals from the sensors prior to determining the loudest sector.
3. The method of claim 2 , comprising filtering using a flattening filter.
4. The method of claim 2 , comprising filtering using a bandpass filter.
5. The method of claim 1 , wherein averaging of signals from the sensors comprises summing together signal energies from adjacent sensors with one additional sensor to compensate for non-symmetry effects of the conferencing device.
6. The method of claim 5 , further comprising performing voice activity detection prior to determining the loudest sector.
7. The method of claim 1 , wherein the steps of selecting the direction associated with the loudest non-active beam to be the current active direction and steering the sensors to the current active direction, are performed only if the energy of the loudest non-active beam is greater than the energy of the signal beam pointing at the sector including the current active direction by a threshold amount.
8. A conferencing device comprising:
a housing;
a plurality of sensors arranged in sectors around the housing; and a beamformer for:
initially having sensors steered to a current active direction;
determining a loudest sector by averaging signals from the sensors; in the event that the loudest sector persists for a first duration:
forming at least three signal beams from sensor signals over a further duration, such that the first signal beam points at a sector including the current active direction, the second signal beam points at the loudest sector and the third and any subsequent signal beams point at one or more sectors adjacent to the loudest sector, but wherein not all sectors are pointed at by a corresponding signal beam, calculating respective beam envelopes of the at least three signal beams to estimate their energies; and
selecting, from the signal beams pointing at the loudest sector and the one or more adjacent sectors, a highest energy one of the signal beams to be a loudest non-active beam;
comparing an energy of the loudest non-active beam to an energy of the signal beam pointing at the sector including the current active direction; and
if the energy of the loudest non-active beam is greater than the energy of the signal beam pointing at the sector including the current active direction:
selecting the direction associated with the loudest non-active beam to be the current active direction; and
steering the sensors to the current active direction.
9. The conferencing device of claim 8 , further comprising a plurality of filters for filtering the signals from the sensors prior to determining the loudest sector.
10. The conferencing device of claim 9 , wherein at least one of the filters is a flattening filter.
11. The conferencing device of claim 9 , wherein at least one of the filters is a bandpass filter.
12. The conferencing device of claim 8 , wherein the beamformer includes a summer for summing together signal energies from adjacent sensors with one additional sensor to compensate for non-symmetry effects of the conferencing device.
13. The conferencing device of claim 12 further comprising a voice activity detector for detecting voice activity prior to the beamformer determining the loudest sector.
14. The conferencing device of claim 8 , wherein the first duration is sufficient to permit stabilized signal beam forming.
15. The conferencing device of claim 8 , wherein said first duration and the further duration are equivalent.
16. The conferencing device of claim 8 , further including a band stop filter for filtering the signal beams during the further duration.
17. The method of claim 7 , wherein the threshold is at least 1.3.
18. The method of claim 1 , wherein averaging signals from the sensors comprises a microphone envelope and averaging technique.
19. The conferencing device of claim 8 , wherein the steps of selecting the direction associated with the loudest non-active beam to be the current active direction and steering the sensors to the current active direction, are performed only if the energy of the loudest non-active beam is greater than the energy of the signal beam pointing at the sector including the current active direction by a threshold amount.
20. The conferencing device of claim 19 , wherein the threshold is at least 1.3.