Acoustic signal classification system

• US 8,428,945 B2
• Filed: 05/11/2011
• Issued: 04/23/2013
• Est. Priority Date: 08/30/1999
• Status: Expired due to Fees

• Current Assignee: 2236008 Ontario Inc. (Blackberry Limited)
• Original Assignee: QNX Software Systems Limited (Canada) (Blackberry Limited)
• Inventors: Zakarauskas, Pierre
• Primary Examiner(s): Armstrong, Angela A
• Application Number:
  US13/105,621
• Publication Number:
  US 20110213612A1
• Time in Patent Office: 713 Days
• Field of Search: 704/201, 704/205, 704/233, 704/226
• US Class Current: 704/233
• CPC Class Codes: G10L 15/20 Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech G10L21/02 takes precedence
• CPC Class Codes: G10L 17/26 Recognition of special voice characteristics, e.g. for use in lie detectors; Recognition of animal voices

Abstract

A system classifies the source of an input signal. The system determines whether a sound source belongs to classes that may include human speech, musical instruments, machine noise, or other classes of sound sources. The system is robust, performing classification despite variation in sound level and noise masking. Additionally, the system consumes relatively few computational resources and adapts over time to provide consistently accurate classification.

Claims

1. A method for classifying a signal, comprising:

obtaining an input signal;

determining a time-frequency representation of the input signal;

determining whether a signal of interest is present in the time-frequency representation;

identifying an interval of the time-frequency representation that contains the signal of interest;

measuring a signal-to-noise ratio of the time-frequency representation;

rescaling frequency bands of the time-frequency representation by a weighting proportional to the measured signal-to-noise ratio of the time-frequency representation;

comparing the rescaled time-frequency representation of the interval to a plurality of time-frequency spectrogram templates in response to identifying the interval of the time-frequency representation that contains the signal of interest; and

selecting a matching time-frequency spectrogram template from the plurality of time-frequency spectrogram templates, by a processor, based on an amount of distortion between the rescaled time-frequency representation and at least one of the plurality of time-frequency spectrogram templates.

2. The method of claim 1, where identifying the interval comprises identifying that a harmonic is present in the interval of the time-frequency representation;

and

where comparing the rescaled time-frequency representation comprises comparing the rescaled time-frequency representation of the interval to the plurality of time-frequency spectrogram templates in response to identifying that the harmonic is present in the interval.

3. The method of claim 1, where identifying the interval comprises:

determining a power level as a function of time and frequency for the time-frequency representation;

estimating a background noise level for the time-frequency representation; and

identifying that the power level exceeds the background noise level by a pre-selected threshold in the interval;

where comparing the rescaled time-frequency representation comprises comparing the rescaled time-frequency representation of the interval to the plurality of time-frequency spectrogram templates in response to identifying that the power level exceeds the background noise level by the pre-selected threshold in the interval.

4. The method of claim 3, where the pre-selected threshold is a pre-selected number of standard deviations beyond a mean background noise level.

5. The method of claim 1, further comprising:

identifying an interval of the time-frequency representation that does not contain signal-of-interest content; and

forgoing spectrogram template matching for the interval of the time-frequency representation that does not contain the signal-of-interest content.

6. The method of claim 1, further comprising:

identifying an interval of the time-frequency representation that does not contain a harmonic; and

forgoing spectrogram template matching for the interval of the time-frequency representation that does not contain the harmonic.

7. The method of claim 1, where selecting the matching time-frequency spectrogram template comprises selecting the matching time-frequency spectrogram template, by the processor, from the plurality of time-frequency spectrogram templates based on an average distance measurement between the rescaled time-frequency representation and at least one of the plurality of spectrogram templates.

8. A method for classifying a signal, comprising:

obtaining an input signal;

determining a time-frequency representation of the input signal;

determining whether a signal of interest is present in the time-frequency representation;

identifying an interval of the time-frequency representation that contains the signal of interest;

comparing the time-frequency representation of the interval to a plurality of time-frequency spectrogram templates in response to identifying the interval of the time-frequency representation that contains the signal of interest;

selecting a matching time-frequency spectrogram template from the plurality of time-frequency spectrogram templates, by a processor, based on an average distance measurement between the time-frequency representation and at least one of the plurality of spectrogram templates; and

weighting the time-frequency representation or the at least one of the plurality of spectrogram templates based on a measured signal-to-noise ratio so that a first frequency band of the input signal, with a lower signal-to-noise ratio than a second frequency band of the input signal, contributes less to the average distance measurement than the second frequency band.

9. A method for classifying a signal, comprising:

obtaining an input signal;

determining a time-frequency representation of the input signal;

determining whether a signal of interest is present in the time-frequency representation;

identifying an interval of the time-frequency representation that contains the signal of interest;

comparing the time-frequency representation of the interval to a plurality of time-frequency spectrogram templates in response to identifying the interval of the time-frequency representation that contains the signal of interest; and

selecting a matching time-frequency spectrogram template from the plurality of time-frequency spectrogram templates, by a processor, based on an amount of distortion between the time-frequency representation and at least one of the plurality of time-frequency spectrogram templates;

where selecting the matching time-frequency spectrogram template comprises;

calculating a first root mean square distance difference between the time-frequency representation and a first template of the plurality of time-frequency spectrogram templates;

calculating a second root mean square distance difference between the time-frequency representation and a second template of the plurality of time-frequency spectrogram templates;

identifying the first template as the matching time-frequency spectrogram in response to a determination that the first root mean square distance difference is lower than the second root mean square distance difference; and

identifying the second template as the matching time-frequency spectrogram in response to a determination that the second root mean square distance difference is lower than the first root mean square distance difference.

10. A method for classifying a signal, comprising:

obtaining an input signal;

determining a time-frequency representation of the input signal;

measuring a signal-to-noise ratio of the time-frequency representation or a template of a plurality of time-frequency spectrogram templates; and

rescaling the template according to the measured signal-to-noise ratio;

determining whether a signal of interest is present in the time-frequency representation;

identifying an interval of the time-frequency representation that contains the signal of interest;

comparing the time-frequency representation of the interval to the plurality of time-frequency spectrogram templates in response to identifying the interval of the time-frequency representation that contains the signal of interest; and

selecting a matching time-frequency spectrogram template from the plurality of time-frequency spectrogram templates, by a processor, based on an amount of distortion between the time-frequency representation and at least one of the plurality of time-frequency spectrogram templates, where selecting the matching time-frequency spectrogram template comprises comparing the time-frequency representation with the rescaled template.

11. An acoustic signal classification system, comprising:

a processor;

memory coupled to the processor, the memory comprising instructions operable to cause the processor to;

obtain an input signal;

determine a time-frequency representation of the input signal;

measure a signal-to-noise ratio of the time-frequency representation or a template of the plurality of time-frequency spectrogram templates; and

rescale the template according to the measured signal-to-noise ratio;

determine whether a signal of interest is present in the time-frequency representation;

identify an interval of the time-frequency representation that contains the signal of interest;

compare the time-frequency representation of the interval to a plurality of time-frequency spectrogram templates in response to identifying the interval of the time-frequency representation that contains the signal of interest; and

select a matching time-frequency spectrogram template from the plurality of time-frequency spectrogram templates based on an amount of distortion between the time-frequency representation and at least one of the plurality of time-frequency spectrogram templates, where the instructions operable to cause the processor to select the matching time-frequency spectrogram template comprise instructions operable to cause the processor to compare the time-frequency representation with the rescaled template.

12. An acoustic signal classification system, comprising:

a processor;

memory coupled to the processor, the memory comprising instructions operable to cause the processor to;

obtain an input signal;

determine a time-frequency representation of the input signal;

determine whether a signal of interest is present in the time-frequency representation;

identify an interval of the time-frequency representation that contains the signal of interest;

measure a signal-to-noise ratio of the time-frequency representation;

rescale frequency bands of the time-frequency representation by a weighting proportional to the measured signal-to-noise ratio of the time-frequency representation;

compare the rescaled time-frequency representation of the interval to a plurality of time-frequency spectrogram templates in response to identifying the interval of the time-frequency representation that contains the signal of interest; and

select a matching time-frequency spectrogram template from the plurality of time-frequency spectrogram templates based on an amount of distortion between the rescaled time-frequency representation and at least one of the plurality of time-frequency spectrogram templates.

13. The system of claim 12, where the instructions operable to cause the processor to identify the interval comprise instructions operable to cause the processor to identify that a harmonic is present in the interval of the time-frequency representation;

and

where the instructions operable to cause the processor to compare the rescaled time-frequency representation comprise instructions operable to cause the processor to compare the rescaled time-frequency representation of the interval to the plurality of time-frequency spectrogram templates in response to identifying that the harmonic is present in the interval.

14. The system of claim 12, where the instructions operable to cause the processor to identify the interval comprises instructions operable to cause the processor to:

determine a power level as a function of time and frequency for the time-frequency representation;

estimate a background noise level for the time-frequency representation; and

identify that the power level exceeds the background noise level by a pre-selected threshold in the interval;

where the instructions operable to cause the processor to compare the rescaled time-frequency representation comprise instructions operable to cause the processor to compare the rescaled time-frequency representation of the interval to the plurality of time-frequency spectrogram templates in response to identifying that the power level exceeds the background noise level by the pre-selected threshold in the interval.

15. The system of claim 12, the memory further comprising instructions operable to cause the processor to:

identify an interval of the time-frequency representation that does not contain a harmonic; and

forgo spectrogram template matching for the interval of the time-frequency representation that does not contain the harmonic.

16. A product, comprising:

a non-transitory computer readable medium; and

instructions stored on the non-transitory computer readable medium that cause a processor in a signal processing system to;

obtain an input signal;

determine a time-frequency representation of the input signal;

determine whether a signal of interest is present in the time-frequency representation;

identify an interval of the time-frequency representation that contains the signal of interest;

measure a signal-to-noise ratio of the time-frequency representation;

rescale the time-frequency representation according to the measured signal-to-noise ratio of the time-frequency representation;

compare the time-frequency representation of the interval to a plurality of time-frequency spectrogram templates in response to identifying the interval of the time-frequency representation that contains the signal of interest; and

select a matching time-frequency spectrogram template from the plurality of time-frequency spectrogram templates based on an amount of distortion between the time-frequency representation and at least one of the plurality of time-frequency spectrogram templates.

17. An acoustic signal classification system, comprising:

a processor;

memory coupled to the processor, the memory comprising instructions operable to cause the processor to;

obtain an input signal;

determine a time-frequency representation of the input signal;

determine whether a signal of interest is present in the time-frequency representation;

identify an interval of the time-frequency representation that contains the signal of interest;

compare the time-frequency representation of the interval to a plurality of time-frequency spectrogram templates in response to identifying the interval of the time-frequency representation that contains the signal of interest;

select a matching time-frequency spectrogram template from the plurality of time-frequency spectrogram templates based on an average distance measurement between the time-frequency representation and at least one of the plurality of spectrogram templates; and

weight the time-frequency representation or the at least one of the plurality of spectrogram templates based on a measured signal-to-noise ratio so that a first frequency band of the input signal, with a lower signal-to-noise ratio than a second frequency band of the input signal, contributes less to the average distance measurement than the second frequency band.

18. An acoustic signal classification system, comprising:

a processor;

memory coupled to the processor, the memory comprising instructions operable to cause the processor to;

obtain an input signal;

determine a time-frequency representation of the input signal;

determine whether a signal of interest is present in the time-frequency representation;

identify an interval of the time-frequency representation that contains the signal of interest;

compare the time-frequency representation of the interval to a plurality of time-frequency spectrogram templates in response to identifying the interval of the time-frequency representation that contains the signal of interest; and

select a matching time-frequency spectrogram template from the plurality of time-frequency spectrogram templates based on an amount of distortion between the time-frequency representation and at least one of the plurality of time-frequency spectrogram templates;

where the instructions operable to cause the processor to select the matching time-frequency spectrogram template comprise instructions operable to cause the processor to;

calculate a first root mean square distance difference between the time-frequency representation and a first template of the plurality of time-frequency spectrogram templates;

calculate a second root mean square distance difference between the time-frequency representation and a second template of the plurality of time-frequency spectrogram templates;

identify the first template as the matching time-frequency spectrogram in response to a determination that the first root mean square distance difference is lower than the second root mean square distance difference; and

identify the second template as the matching time-frequency spectrogram in response to a determination that the second root mean square distance difference is lower than the first root mean square distance difference.