Signal processing systems

• US 9,459,733 B2
• Filed: 02/22/2013
• Issued: 10/04/2016
• Est. Priority Date: 08/27/2010
• Status: Active Grant

• Current Assignee: InputDynamics Limited
• Original Assignee: InputDynamics Limited
• Inventors: Godsill, Simon, Christensen, Jens, Bisutti, Giovanni
• Primary Examiner(s): Awad, Amr
• Assistant Examiner(s): FLORES, ROBERTO W
• Application Number:
  US13/774,434
• Publication Number:
  US 20140071095A1
• Time in Patent Office: 1,320 Days
• Field of Search: 345/173, 345/177, 345/178, 178/18.04
• US Class Current: 1/1
• CPC Class Codes: G06F 3/043 using propagating acoustic waves
• CPC Class Codes: G06F 3/0433 in which the acoustic waves are either generated by a movable member and propagated within a surface layer or propagated within a surface layer and captured by a movable member

Abstract

This invention relates to methods, apparatus, and computer program code for processing acoustic signal data to determine where an object has been tapped with a stylus, finger nail or the like. The method involved storing a set of labelled training data comprising digitized waveforms from a sensor for taps at a plurality of different locations. The labelled training data is then processed to determine mean value and covariance data for the waveforms, which is afterwards used in conjunction with a digitized waveform of a tap at an unknown location to identify the location of the tap. Preferably the covariance is decomposed into a plurality of basis functions for each region each with a respective weighting, which are used to represent captured data for an unknown tap and parameters of the representation are classified to locate the tap.

Claims

1. A method of detecting a tap on an object having at least one acoustic/vibration sensor, including a method of determining a threshold for declaring a detected tap, the method comprising:

capturing tap data comprising a digitised waveform of a tap on said object captured by said at least one acoustic/vibration sensor;

processing said captured tap data by applying a probabilistic tap detection procedure to generate time series tap metric data comprising data providing a detection metric of a potential said tap at a succession of times, wherein said time series tap metric data comprises a time series of tap detection probabilities;

determining a threshold for a classifier wherein said threshold comprises a probability threshold to apply to said tap detection probabilities from said probabilistic tap detection procedure to declare a detected tap; and

applying said classifier to said time series of tap detection probabilities data to declare a detected said tap on said object; and

wherein said determining of said probability threshold for said classifier comprises;

identifying a region of background noise without taps;

adding to this background noise one or more examples of stored clean taps at one or more time locations in said background noise to generate synthetic training data;

applying said probabilistic tap detection procedure to said generated synthetic training data comprising said one or more examples of stored clean taps in said background noise without taps, to generate training probability data; and

using said training probability data to determine said threshold for said classifier.

2. A method as claimed in claim 1 wherein said identifying said region of background noise comprises:

identifying potential said taps for which said detection metric is greater than a threshold value; and

defining said region of background noise to exclude time windows around each said identified potential said tap.

3. A method as claimed in claim 1 for detecting the location of said tap on said object, wherein said stored clean tap data comprises data for taps at a plurality of different locations on said object, wherein said tap detection procedure is configured for determining a tapped region of said object from said tap data when said object is tapped in one of a plurality of tap-sensing regions of the object;

and wherein the method comprises determining a plurality of said classifier thresholds, one for each said tap-sensing region.

4. A method as claimed in claim 1 wherein said determining of said threshold for said classifier further comprises determining background probability data from said probabilistic tap detection procedure applied to said background noise;

the method further comprising determining said threshold for said classifier from a combination of said background probability data and said training probability data.

5. A method as claimed in claim 1 wherein determining said threshold for said classifier from a combination of said background probability data and said training probability data comprises determining a threshold between a first value of said background probability data and a second value of said training probability data.

6. An electronic device for detecting a tap on an object having at least one acoustic/vibration sensor, configured to determine a threshold for declaring a detected tap, wherein the electronic device comprises a processor, working memory, program memory, and a communications interface, the program memory storing processor control code to:

capture tap data comprising a digitised waveform of a tap on said object captured by said at least one acoustic/vibration sensor;

process said captured tap data by applying a probabilistic tap detection procedure to generate time series tap metric data comprising data providing a detection metric of a potential said tap at a succession of times, wherein said time series tap metric data comprises a time series of tap detection probabilities;

provide said time series of tap detection probabilities to a classifier for classifying said detection probabilities of potential said taps to declare one or more detected taps;

determine a threshold for said classifier, wherein said threshold comprises a probability threshold to apply to said tap detection probabilities from said probabilistic tap detection procedure to declare a detected tap; and

apply said classifier to said time series of tap detection probabilities data to declare a detected said tap on said object; and

wherein to determine of said probability threshold for said classifier the electronic device is further configured to;

identify a region of background noise in said captured tap data without taps;

add to this background noise one or more examples of stored clean taps at one or more time locations in said background noise to generate synthetic training data;

apply said probabilistic tap detection procedure to said generated synthetic training data comprising said one or more examples of stored clean taps in said background noise without taps, to generate training probability data; and

use said training probability data to determine said threshold for said classifier.