Multibeam keyword detection system and method
A system and method provides for multibeam keyword detection. A composite audio signal may include sound components. The system and method groups the sound components into subsets based on the angles of arrival of sound components. Keyword detectors evaluate each subset and determine whether a keyword is present.
1. A system for keyword detection comprising:
an audio input configured to receive a composite audio signal from a microphone array;
a beamforming engine comprising a plurality of beamformers coupled to receive the composite audio signal and configured to form a plurality of beam signals from the composite audio signal, each of the plurality of beam signals having an associated beam angle and comprising audio data corresponding to sounds captured by the microphone array within a range of angles of arrival with respect to the beam angle;
a keyword detection module having a plurality of keyword detectors, each keyword detector configured to receive a respective one of the plurality of beam signals and to detect a keyword; and
a multibeam detection engine configured to receive, from each of the plurality of keyword detectors, keyword presence data indicating whether or not the keyword detector has detected a keyword and configured to produce a keyword detection output indicating whether or not a keyword has been detected by one or more of the keyword detectors based on the received keyword presence data.
2. The system for keyword detection of claim 1 , further comprising a preprocessor coupled between the audio input and the beamforming engine and configured to filter the composite audio signal.
3. The system for keyword detection of claim 2 , wherein the preprocessor comprises an acoustic echo canceler to at least partially cancel echoes in the composite audio signal.
4. The system for keyword detection of claim 1 , wherein when the multibeam detection engine produces the keyword detection output indicating that the keyword has been detected in response to a first keyword detector detecting the keyword, the multibeam detection engine is further configured to inhibit a duplicate production of the keyword detection output in response to a second keyword detector also detecting the keyword within a temporally proximate period of time following the first keyword detector.
5. The system for multibeam keyword detection of claim 1 , further comprising a memory that stores configuration information of the microphone array, wherein the beamforming engine is configured to access the memory to produce the plurality of beam signals.
6. The system for multibeam keyword detection of claim 1 , wherein the multibeam detection engine is configured to cause the keyword detection output to indicate that the keyword has been detected if the keyword presence data of any of the keyword detectors indicates that the keyword has been detected.
7. The system for multibeam keyword detection of claim 1 , wherein the keyword presence data includes false alarm data comprising a probability of a false alarm, and wherein the multibeam detection engine is configured to produce the keyword detection output using the false alarm data.
8. The system for multibeam keyword detection of claim 7 , wherein the multibeam detection engine is configured to cause the keyword detection output to indicate that the keyword has not been detected if the false alarm data of any of the keyword detectors indicates the probability of false alarm.
9. The system for multibeam keyword detection of claim 1 , wherein the plurality of beamformers comprises four beamformers and wherein the range of angles of arrival associated with each beamformer of the plurality of beamformers is approximately 90 degrees in width.
10. The system for multibeam keyword detection of claim 1 , wherein the plurality of beamformers comprises four beamformers and wherein the range of angles of arrival associated with at least one beamformer of the plurality of beamformers is variable.
11. The system for multibeam keyword detection of claim 1 , wherein the plurality of beamformers comprises four beamformers and wherein the range of angles of arrival associated with at least one beamformer of the plurality of beamformers is variable and at least one further beamformer of the plurality of beamformers is non-variable.
12. The system for multibeam keyword detection of claim 1 , wherein the multibeam detection engine is further configured to provide an angle of arrival associated with a detected keyword.
13. A method of keyword detection comprising:
receiving from a microphone array a composite audio signal;
forming a plurality of beam signals from the composite audio signal, each of the plurality of beam signals having an associated beam angle and comprising audio data corresponding to sounds captured by the microphone array within a range of angles of arrival with respect to the beam angle;
processing, by a plurality of keyword detectors, the plurality of beam signals to detect a keyword in the plurality of beam signals, each keyword detector configured to receive a respective one of the plurality of beam signals and to detect a keyword; and
processing keyword presence data from all of the plurality of keyword detectors to produce a keyword detection output indicating whether or not a keyword has been detected based on the keyword presence data.
14. The method of keyword detection of claim 13 , further comprising preprocessing the composite audio signal before forming the plurality of beam signals.
15. The method of keyword detection of claim 14 , wherein preprocessing comprises performing acoustic echo cancelation to at least partially cancel echoes in the composite audio signal.
16. The method of multibeam keyword detection of claim 13 , wherein forming the plurality of beam signals includes accessing configuration information of the microphone array.
17. The method of multibeam keyword detection of claim 13 , wherein at least one beam signal of the plurality of beam signals has at least one of (i) a variable width and (ii) a variable direction.
18. The method of multibeam keyword detection of claim 13 , wherein processing the keyword presence data includes causing the keyword detection output to indicate that the keyword has been detected if the keyword presence data of any of the keyword detectors indicates that the keyword has been detected.
19. The method of multibeam keyword detection of claim 13 , wherein the keyword presence data includes false alarm data, and wherein processing the keyword presence data includes causing the keyword detection output to indicate that the keyword has not been detected if the keyword presence data of any of the keyword detectors indicates a probability of a false alarm.
20. The method of multibeam keyword detection of claim 13 , wherein processing the keyword presence data further includes providing an angle of arrival associated with a detected keyword.