Automatic detection of non-stationarity in speech signals

• US 6,535,843 B1
• Filed: 08/18/1999
• Issued: 03/18/2003
• Est. Priority Date: 08/18/1999
• Status: Expired due to Term

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: AT&T Corporation (AT&T, Inc.)
• Inventors: Kapilow, David A., Schroeter, Juergen, Stylianou, Ioannis G.
• Primary Examiner(s): To, Doris H.
• Assistant Examiner(s): Nolan, Daniel A.
• Application Number:
  US09/376,456
• Time in Patent Office: 1,308 Days
• Field of Search: 704/207, 704/206, 704/208, 704/209, 704/265, 704/211, 704/220, 704/266
• US Class Current: 704/207
• CPC Class Codes: G10L 21/04 Time compression or expansion
• CPC Class Codes: G10L 25/48 specially adapted for particular use

Abstract

When necessary to time scale a speech signal, it is advantageous to do it under influence of a signal that measures the small-window non-stationarity of the speech signal. Three measures of stationarity are disclosed: one that is based on time domain analysis, one that is based on frequency domain analysis, and one that is based on both time and frequency domain analysis.

Claims

1. A method for developing a measure of non-stationarity of an input speech signal comprising the steps of:

dividing said input signal into intervals;

evaluating a measure of variability of a selected attribute of said input signal in each of said intervals;

from said measure of variability, developing an analog measure of non-stationarity of said input signal for every one of said intervals.

2. The method of claim 1 where said intervals are uniform, with a length that is on the order of 30 msec.

3. The method of claim 1 where said step of developing an analog measure of non-stationarity of said input signal for each of said intervals develops a measure that is bounded by 0 and 1.

4. The method of claim 1 where said step of evaluating a measure of variability considers a time-domain characteristic of said input signal.

5. The method of claim 1 where said step of evaluating a measure of variability evaluates the RMS value of each interval of said input signal, E_n, in accordance with the relationship $E_n=\sqrt{\frac{1}{N+1}}$

∑

m=-N/2N/2

 



x2

(n+m),

where x represents a sample of said input signal in said interval, and N+1 is the number of such samples in said interval,

developing a measure of non-stationarity of said input signal by evaluating the quotient $\frac{<br><br>E_n-E_{n-1}<br><br>}{E_n+E_{n-1}}$

 

each of said intervals.

6. The method of claim 1 where said step of evaluating a measure of variability considers a frequency-domain characteristic of said input signal.

7. The method of claim 1 where said step of evaluating a measure of variability evaluates $\frac{2}{1+{}^{}}$

-β

1

s

(n)-1,

where β

₁ is a preselected constant and s(n) is a spectral transition rate in interval n of a selected number of spectral lines of said input signal.

8. The method of claim 7 where said s(n) signal is developed in accordance with the relationship $s$

(n)=∑

i=1P

 



(ci

(n))2,

where $c_i<br><br>\left(n\right)=\frac{\underset{m=-M}{\overset{M}{\sum <br><br>}}<br><br>\text{\hspace{1em}<br><br>}<br><br>{\mathrm{my}}_i<br><br>\left(n+m\right)}{\underset{m=-M}{\overset{M}{\sum <br><br>}}<br><br>\text{\hspace{1em}<br><br>}<br><br>m^2},$

and y_i is the i^th spectral line.

9. The method of claim 1 where said step of evaluating a measure of variability considers a time domain and a frequency-domain characteristic of said input signal.

10. The method of claim 9 where said step of evaluating a measure of variability evaluates $\frac{2}{1+{}^{}}$

-β

2

s

(n)-α



 



Cn1-1,

where β

₂ is a preselected constant, α

is another preselected constant, s(n) is a spectral transition rate in interval n of a selected number of spectral lines of said input signal, and $C_n^1=\frac{<br><br>E_n-E_{n-1}<br><br>}{E_n+E_{n-1}}$

where E_n is the RMS value of said input signal within a time interval n, and E<sub>n−

1</sub> is the RMS value of the speech signal within a time interval (n−

1).

11. A method for modifying a speech signal comprising the steps of:

dividing said speech signal into uniform time intervals,

for every interval, computing an analog stationarity measure, ƒ

(n), that is related to energy of said signal within said interval, and

modifying said signal within said interval by a factor that is based on said measure.

12. The method of claim 11 where said measure has a range that approximately spans the interval 0 to 1.

13. The method of claim 11 where $f$

(n)=

En-En-1

En+En-1,

E_n is the a root mean squared value of the speech signal within time interval n, and E<sub>n−

1</sub> is a root mean squared value of the speech signal within time interval (n−

1).

14. The method of claim 13 where $E_n=\sqrt{\frac{1}{N+1}}$

∑

m=-N/2N/2

 



x2

(n+m),

where x(n) is the speech signal over an interval of N+1 samples.

15. The method of claim 11 where said time intervals do not overlap.

16. The method of claim 11 where said time intervals overlap by a preselected amount.

17. The method of claim 11 where said measure is related to a root mean square measure of said signal in said interval.

18. The method of claim 11 where said factor, β

, is β

=1+[1−

ƒ

(n)]b, where b is a preselected constant.

19. The method of claim 11 where said modifying is time scaling of said signal in said time interval.

20. A method for modifying a speech signal comprising the steps of:

dividing said signal into time intervals,

for every interval, n, computing an analog stationarity measure, f(n), that is related to spectral parameters of said signal within said interval, and

modifying said signal within said interval by a scaling factor that is based on said measure.

21. The method of claim 20 where said modifying is time scaling of said signal in said time interval.

22. The method of claim 20 where said spectral parameters measure corresponds to spectral feature transition rate.

23. The method of claim 20 where said spectral parameters measure is related to $s$

(n)=∑

i=1P

 



ci

(n)2,

where $c_i<br><br>\left(n\right)=\frac{\underset{m=-M}{\overset{M}{\sum <br><br>}}<br><br>\text{\hspace{1em}<br><br>}<br><br>{\mathrm{my}}_i<br><br>\left(n+m\right)}{\underset{m=-M}{\overset{M}{\sum <br><br>}}<br><br>\text{\hspace{1em}<br><br>}<br><br>m^2},$

y_i is an i^th spectral parameter about a time window [n−

M, n+M].

24. The method of claim 23 where said scaling factor is $\frac{2}{1+{}^{}}$

-β

1

s

(n)-1,

where β

₁ is a preselected weight factor.

25. The method of claim 23 where said scaling factor is $\frac{2}{1+{}^{}}$

-β

2

s

(n)-α



 



Cn1-1,

where β

₂ and α

are preselected constants, $C_n^1=\frac{<br><br>E_n-E_{n-1}<br><br>}{E_n+E_{n-1}},$

E_n is the a root mean squared value of the speech signal within time interval n, and E<sub>n−

1</sub> is a root mean squared value of the speech signal within time interval (n−

1).