Systems and methods for hands-free voice control and voice search

• US 9,484,028 B2
• Filed: 02/19/2014
• Issued: 11/01/2016
• Est. Priority Date: 07/06/2009
• Status: Active Grant

• Current Assignee: Sensory Incorporated
• Original Assignee: Sensory Incorporated
• Inventors: Vermeulen, Pieter J., Shaw, Jonathan, Mozer, Todd F.
• Primary Examiner(s): WOZNIAK, JAMES S
• Application Number:
  US14/184,306
• Publication Number:
  US 20140180691A1
• Time in Patent Office: 986 Days
• Field of Search: 704/231, 704/236, 704/242, 704/248, 704/251, 704/254
• US Class Current: 1/1
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 15/32 Multiple recognisers used in sequence or in parallel; Score combination systems therefor, e.g. voting systems
• CPC Class Codes: G10L 2015/081 Search algorithms, e.g. Baum-Welch or Viterbi
• CPC Class Codes: G10L 2015/088 Word spotting

Abstract

In one embodiment the present invention includes a method comprising receiving an acoustic input signal and processing the acoustic input signal with a plurality of acoustic recognition processes configured to recognize the same target sound. Different acoustic recognition processes start processing different segments of the acoustic input signal at different time points in the acoustic input signal. In one embodiment, initial states in the recognition processes may be configured on each time step.

Claims

1. A method comprising:

receiving an acoustic input signal that includes a plurality of spoken words or phrases; and

processing the acoustic input signal with a plurality of acoustic recognition processes to recognize a predetermined target sound within the acoustic input signal,

wherein the acoustic input signal is temporally divided into a plurality of frames,

wherein the plurality of acoustic recognition processes are implemented using a single Viterbi search of a plurality of states corresponding to acoustic units of an acoustic model of the predetermined target sound, the plurality of states including initial states and final states of the acoustic model, each initial state in the plurality of states being associated with successive frames in the plurality of frames,

wherein, at each frame, one or more of the plurality of acoustic recognition processes determine whether the predetermined target sound has been recognized, and

wherein a determination that the target sound has been recognized ends the processing of the acoustic input signal.

2. The method of claim 1 wherein each state in the plurality of states is associated with a score.

3. The method of claim 2 wherein the score comprises a first component indicating a probability of a match between a particular state and a received acoustic unit of the acoustic input signal, a second component comprising a highest score of a predecessor state, and a third component comprising an offset.

4. The method of claim 2 wherein calculating the score for each state comprises increasing the score by a per frame offset so that scores of different durations are comparable.

5. The method of claim 4 wherein the offset is a constant greater than one.

6. The method of claim 4 wherein different states have different associated offset values.

7. The method of claim 1 wherein the plurality of states include a state corresponding to an acoustic model for garbage.

8. The method of claim 1 wherein a score of each initial state is reset on each frame if the score is less than a threshold.

9. The method of claim 8 wherein the score of each initial state is reset to a predetermined value on each frame if the score is less than the threshold, before calculating a current score for the initial state based on a received acoustic unit for a current frame.

10. The method of claim 9 wherein said predetermined value is a constant.

11. The method of claim 9 wherein different initial states are reset to different predetermined values.

12. The method of claim 1 wherein the plurality of acoustic recognition processes operate on multiple words in parallel.

13. A non-transitory computer readable storage medium having stored thereon program code executable by a processor, said program code comprising:

code that causes the processor to receive an acoustic input signal that includes a plurality of spoken words or phrases; and

code that causes the processor to process the acoustic input signal with a plurality of acoustic recognition processes to recognize a predetermined target sound within the acoustic input signal,

wherein the plurality of acoustic recognition processes are implemented using a single Viterbi search of a plurality of states corresponding to acoustic units of an acoustic model of the predetermined target sound, the plurality of states including initial states and final states of the acoustic model, each initial state in the plurality of states being associated with successive frames in the plurality of frames,

wherein, at each frame, one or more of the plurality of acoustic recognition processes determine whether the predetermined target sound has been recognized, and

wherein a determination that the target sound has been recognized ends the processing of the acoustic input signal.

14. The non-transitory computer readable storage medium of claim 13 wherein each state in the plurality of states is associated with a score.

15. The non-transitory computer readable storage medium of claim 14 wherein calculating the score for each state comprises increasing the score by a per frame offset so that scores of different durations are comparable.

16. The non-transitory computer readable storage medium of claim 13 wherein a score of each initial state is reset on each frame if the score is less than a threshold.

17. A computer system comprising:

one or more processors, each processor configured to;

receive an acoustic input signal that includes a plurality of spoken words or phrases; and

process the acoustic input signal with a plurality of acoustic recognition processes to recognize a predetermined target sound within the acoustic input signal,

wherein the plurality of acoustic recognition processes are implemented using a single Viterbi search of a plurality of states corresponding to acoustic units of an acoustic model of the predetermined target sound, the plurality of states including initial states and final states of the acoustic model, each initial state in the plurality of states being associated with successive frames in the plurality of frames,

wherein, at each frame, one or more of the plurality of acoustic recognition processes determine whether the predetermined target sound has been recognized, and

wherein a determination that the target sound has been recognized ends the processing of the acoustic input signal.