Automated sound segment selection method and system

• US 8,494,844 B2
• Filed: 11/19/2009
• Issued: 07/23/2013
• Est. Priority Date: 11/19/2008
• Status: Active Grant

• Current Assignee: David N. Fernandes
• Original Assignee: Human Centered Technologies Incorporated
• Inventors: Fernandes, David N.
• Primary Examiner(s): Desir, Pierre-Louis
• Assistant Examiner(s): Serrou, Abdelali
• Application Number:
  US12/592,125
• Publication Number:
  US 20100153101A1
• Time in Patent Office: 1,342 Days
• Field of Search: 704207-209, 704/220, 704/231, 704/251
• US Class Current: 704/220
• CPC Class Codes: G10L 17/26 Recognition of special voice characteristics, e.g. for use in lie detectors; Recognition of animal voices
• CPC Class Codes: G10L 25/69 for evaluating synthetic or decoded voice signals

Abstract

A computerized method and system is provided for automatically selecting from a digitized sound sample a segment of the sample that is optimal for the purpose of measuring clinical metrics for voice and speech assessment. A quality measure based on quality parameters of segments of the sound sample is applied to candidate segments to identify the highest quality segment within the sound sample. The invention can optionally provide feedback to the speaker to help the speaker increase the quality of the sound sample provided. The invention also can optionally perform sound pressure level calibration and noise calibration. The invention may optionally compute clinical metrics on the selected segment and may further include a normative database method or system for storing and analyzing clinical measurements.

Claims

1. A method for selecting a segment of a sound sample of a subject'"'"'s voice that has been digitized by an analog to digital converter, comprising the steps of:

(a) receiving the sound sample comprising a time series of digital samples;

(b) determining a plurality of segments, each segment comprising a pre-defined number of consecutive digital samples;

(c) for each segment, calculating the value of a quality measure, the quality measure being a real-valued function of at least one parameter of the segment, the parameter reflecting the suitability of the segment for use in computing clinical metrics;

(d) selecting the segment that has the greatest value of the quality measure, the quality measure being greater for segments that are more suitable for use in computing clinical metrics; and

(e) storing the selected segment, or a pointer to the selected segment, in a memory;

wherein the quality measure is the product of functions w₁, w₂, w₃, w₄, and w₅, where;

(a) w₁ is a monotonically increasing, non-negative real-valued function computed on input value r_N, where r_N is the ratio of the sound pressure level of the segment to the estimated background noise, measured in decibels;

(b) w₂ is a monotonically decreasing, non-negative real-valued function computed on input value CV_SPL, where CV_SPL is the ratio of the standard deviation of the sound pressure level for the segment to the mean of the sound pressure level for the segment;

(c) w₃ is a monotonically decreasing, non-negative real-valued function computed on input value r_c, where r_c is the fraction of digital samples in the segment that are clipped by the analog-to-digital converter;

(d) w₄ is a monotonically increasing, non-negative real-valued function computed on input value r_v where r_v is the fraction of the segment that is voiced; and

(e) w₅ is a monotonically decreasing, non-negative real-valued function computed on input value CV_F0, where CV_F0 is the ratio of the standard deviation of the fundamental frequency for the segment to the mean of the fundamental frequency for the segment.

2. The method of claim 1 wherein the function w₁ is defined to be

3. The method of claim 2 wherein k₂ is 5, k₃ is 30, k₄ is 5, and k₅ is 20.

4. The method of claim 3 wherein the sound pressure level is calibrated to produce an absolute sound pressure level in decibels and ρ

, x₁ and x₂ are chosen so that 0.3<

ρ

<

0.6, 20 dB<

x₁<

34 dB and 32 dB<

x₂<

46 dB.

5. The method of claim 4 wherein ρ

is 0.4, x₁ is 30 dB and x₂ is 42 dB.

6. The method of claim 1 wherein the sound pressure level is calibrated to produce an absolute sound pressure level in decibels.

7. The method of claim 6 wherein the value of r_N is calculated in real-time as the subject presents a sound to the analog-to-digital converter, based on the last N digital samples, where N is a pre-defined number, and an indication of the value of r_N is presented to the subject.

8. The method of claim 7 wherein a graphical indication of the value of r_N is presented to the user along with an indication of the sufficiency of the value.

9. The method of claim 1 wherein the method selects a plurality of segments, each segment comprising a pre-defined number of consecutive digital samples selected to be used to compute a particular clinical metric.

10. The method of claim 9 wherein a plurality of quality measures are used, each quality measure being defined to select a segment suitable for the computation of a particular clinical metric.

11. The method of claim 1 wherein the method computes clinical metrics on the selected segment.

12. The method of claim 11 further comprising a normative database method for storing and analyzing clinical measurements.

13. A non-transitory computer-readable memory having recorded thereon statements and instructions for execution by a processor for selecting a segment of a sound sample of a subject'"'"'s voice that has been digitized by an analog to digital converter, said statements and instructions when executed by a processor cause the processor to perform the steps of:

(a) receiving the sound sample comprising a time series of digital samples;

(b) determining a plurality of segments, each segment comprising a pre-defined number of consecutive digital samples;

(c) for each segment, calculating the value of a quality measure, the quality measure being a real-valued function of at least one parameter of the segment, the parameter reflecting the suitability of the segment for use in computing clinical metrics;

(d) selecting the segment that has the greatest value of the quality measure, the quality measure being greater for segments that are more suitable for use in computing clinical metrics; and

(e) storing the selected segment, or a pointer to the selected segment, in a memory;

wherein the quality measure is the product of functions w₁, w₂, w₃, w₄, and w₅, where;

(a) w₁ is a monotonically increasing, non-negative real-valued function computed on input value r_N, where r_N is the ratio of the sound pressure level of the segment to the estimated background noise, measured in decibels;

(b) w₂ is a monotonically decreasing, non-negative real-valued function computed on input value CV_SPL, where CV_SPL is the ratio of the standard deviation of the sound pressure level for the segment to the mean of the sound pressure level for the segment;

(c) w₃ is a monotonically decreasing, non-negative real-valued function computed on input value r_c, where r_c is the fraction of digital samples in the segment that are clipped by the analog-to-digital converter;

(d) w₄ is a monotonically increasing, non-negative real-valued function computed on input value r_v, where r_v is the fraction of the segment that is voiced; and

(e) w₅ is a monotonically decreasing, non-negative real-valued function computed on input value CV_F0, where CV_F0 is the ratio of the standard deviation of the fundamental frequency for the segment to the mean of the fundamental frequency for the segment.

14. A method comprising transmitting over a communications medium instructions for execution by a processor for selecting a segment of a sound sample of a subject'"'"'s voice that has been digitized by an analog to digital converter, said instructions when executed by a processor cause the processor to perform the steps of:

(a) receiving the sound sample comprising a time series of digital samples;

(b) determining a plurality of segments, each segment comprising a pre-defined number of consecutive digital samples;

(c) for each segment, calculating the value of a quality measure, the quality measure being a real-valued function of at least one parameter of the segment, the parameter reflecting the suitability of the segment for use in computing clinical metrics;

(d) selecting the segment that has the greatest value of the quality measure, the quality measure being greater for segments that are more suitable for use in computing clinical metrics; and

(e) storing the selected segment, or a pointer to the selected segment, in a memory;

wherein the quality measure is the product of functions w₁, w₂, w₃, w₄, and w₅, where;

(a) w₁ is a monotonically increasing, non-negative real-valued function computed on input value r_N, where r_N is the ratio of the sound pressure level of the segment to the estimated background noise, measured in decibels;

(b) w₂ is a monotonically decreasing, non-negative real-valued function computed on input value CV_SPL, where CV_SPL is the ratio of the standard deviation of the sound pressure level for the segment to the mean of the sound pressure level for the segment;

(c) w₃ is a monotonically decreasing, non-negative real-valued function computed on input value r_c, where r_c is the fraction of digital samples in the segment that are clipped by the analog-to-digital converter;

(d) w₄ is a monotonically increasing, non-negative real-valued function computed on input value r_v, where r_v is the fraction of the segment that is voiced; and

(e) w₅ is a monotonically decreasing, non-negative real-valued function computed on input value CV_F0, where CV_F0 is the ratio of the standard deviation of the fundamental frequency for the segment to the mean of the fundamental frequency for the segment.

15. A non-transitory computer-readable memory storing instructions for execution by a processor for selecting a segment of a sound sample of a subject'"'"'s voice that has been digitized by an analog to digital converter, said instructions when executed by a processor cause the processor to perform the steps of:

(a) receiving the sound sample comprising a time series of digital samples;

(b) determining a plurality of segments, each segment comprising a pre-defined number of consecutive digital samples;

(c) for each segment, calculating the value of a quality measure, the quality measure being a real-valued function of at least one parameter of the segment, the parameter reflecting the suitability of the segment for use in computing clinical metrics;

(d) selecting the segment that has the greatest value of the quality measure, the quality measure being greater for segments that are more suitable for use in computing clinical metrics; and

(e) storing the selected segment, or a pointer to the selected segment, in a memory;

wherein the quality measure is the product of functions w₁, w₂, w₃, w₄, and w₅, where;

(a) w₁ is a monotonically increasing, non-negative real-valued function computed on input value r_N, where r_N is the ratio of the sound pressure level of the segment to the estimated background noise, measured in decibels;

(b) w₂ is a monotonically decreasing, non-negative real-valued function computed on input value CV_SPL, where CV_SPL is the ratio of the standard deviation of the sound pressure level for the segment to the mean of the sound pressure level for the segment;

(c) w₃ is a monotonically decreasing, non-negative real-valued function computed on input value r_c where r_c is the fraction of digital samples in the segment that are clipped by the analog-to-digital converter;

(d) w₄ is a monotonically increasing, non-negative real-valued function computed on input value r_v, where r_v is the fraction of the segment that is voiced; and

(e) w₅ is a monotonically decreasing, non-negative real-valued function computed on input value CV_F0, where CV_F0 is the ratio of the standard deviation of the fundamental frequency for the segment to the mean of the fundamental frequency for the segment.

16. A voice segment selection system for selecting a segment of a sound sample of a subject'"'"'s voice that has been digitized by an analog to digital converter into a time series of digital samples, the system comprising:

(a) a processor and computer readable memory;

(b) a segmentation module for determining, using the processor, a plurality of segments from the sound sample, each segment comprising a pre-defined number of consecutive digital samples;

(c) a quality module for calculating the value of a quality measure for each segment, the quality measure being a real-valued function of at least one parameter of the segment, the parameter reflecting the suitability of the segment for use in computing clinical metrics; and

(d) a selection module for selecting the segment that has the greatest value of the quality measure and storing the selected segment, or a pointer to the selected segment, in a memory, the quality measure being greater for segments that are more suitable for use in computing clinical metrics;

wherein the quality measure is the product of functions w₁, w₂, w₃, w₄, and w₅, where;

(a) w₁ is a monotonically increasing, non-negative real-valued function computed on input value r_N, where r_N is the ratio of the sound pressure level of the segment to the estimated background noise, measured in decibels;

(b) w₂ is a monotonically decreasing, non-negative real-valued function computed on input value CV_SPL, where CV_SPL is the ratio of the standard deviation of the sound pressure level for the segment to the mean of the sound pressure level for the segment;

(c) w₃ is a monotonically decreasing, non-negative real-valued function computed on input value r_c, where r_c is the fraction of digital samples in the segment that are clipped by the analog-to-digital converter;

(d) w₄ is a monotonically increasing, non-negative real-valued function computed on input value r_v, where r_v is the fraction of the segment that is voiced; and

(e) w_s is a monotonically decreasing, non-negative real-valued function computed on input value CV_F0, where CV_F0 is the ratio of the standard deviation of the fundamental frequency for the segment to the mean of the fundamental frequency for the segment.

17. The system of claim 16 wherein the function w₁ is defined to be

18. The system of claim 17 wherein k₂ is 5, k₃ is 30, k₄ is 5, and k₅ is 20.

19. The system of claim 18 wherein the sound pressure level is calibrated to produce an absolute sound pressure level in decibels and ρ

, x₁ and x₂ are chosen so that 0.3<

ρ

<

0.6, 20 dB<

x₁<

34 dB and 32 dB<

x₂<

46 dB.

20. The system of claim 19 wherein ρ

is 0.4, x₁ is 30 dB and x₂ is 42 dB.

21. The system of claim 16 wherein the sound pressure level is calibrated to produce an absolute sound pressure level in decibels.

22. The system of claim 21 wherein the value of r_N is calculated in real-time as the subject presents a sound to the analog-to-digital converter, based on the last N digital samples, where N is a pre-defined number, and an indication of the value of r_N is presented to the subject.

23. The system of claim 22 wherein a graphical indication of the value of r_N is presented to the user along with an indication of the sufficiency of the value.

24. The system of claim 16 wherein the system selects a plurality of segments, each segment comprising a pre-defined number of consecutive digital samples selected to be used to compute a particular clinical metric.

25. The system of claim 24 wherein a plurality of quality measures are used, each quality measure being defined to select a segment suitable for the computation of a particular clinical metric.

26. The system of claim 16 wherein the system computes clinical metrics on the selected segment.

27. The system of claim 26 wherein the system further comprises a normative database subsystem for storing and analyzing clinical measurements using the stored selected segment.

28. The system of claim 16 wherein the system acts as a server that receives from a client a sound sample and a request to select a segment, wherein the system returns to the client an indication of the selected segment.