Speech signal distortion measurement which varies as a function of the distribution of measured distortion over time and frequency

• US 5,794,188 A
• Filed: 04/04/1996
• Issued: 08/11/1998
• Est. Priority Date: 11/25/1993
• Status: Expired due to Term

• Current Assignee: Psytechnics Ltd. (NetScout Systems Incorporated)
• Original Assignee: British Telecommunications PLC (BT Group PLC)
• Inventors: Hollier, Michael Peter
• Primary Examiner(s): Eisenzopf, Reihard J.
• Assistant Examiner(s): Smits, Talivaldis Ivars
• Application Number:
  US08/624,489
• Time in Patent Office: 859 Days
• Field of Search: 395/2.37, 395/2.42, 704/228, 704/233
• US Class Current: 704/228
• CPC Class Codes: G01R 23/20 Measurement of non-linear distortion
• CPC Class Codes: G10L 25/69 for evaluating synthetic or decoded voice signals
• CPC Class Codes: H04B 1/665 using psychoacoustic properties of the ear, e.g. masking effect
• CPC Class Codes: H04B 2203/5408 using protocols
• CPC Class Codes: H04B 3/46 Monitoring; Testing
• CPC Class Codes: H04B 3/542 the information being in digital form
• CPC Class Codes: H04M 1/24 Arrangements for testing
• CPC Class Codes: H04M 2201/40 using speech recognition speech recognition per se G10L15/00
• CPC Class Codes: H04M 3/2236 Quality of speech transmission monitoring

Abstract

Telecommunications testing apparatus includes an analyzer arranged to receive a distorted signal which corresponds to a test signal when distorted by telecommunications aparatus to be tested. The analyzer periodically derives, from the distorted signal, a plurality of spectral component signals responsive to the distortion in each of a plurality of spectral bands, over a succession of time intervals. The analyzer generates a measure of the subjective impact of the distortion due to the telecommunications apparatus, the measure of subjective impact being calculated to depend upon the spread of the distortion over time and/or over the spectral bands.

Claims

1. Telecommunications testing apparatus arranged to receive a distorted signal which corresponds to a test signal when distorted by telecommunications apparatus to be tested, the testing apparatus comprising:

means for periodically deriving, from the distorted signal, a plurality of spectral component signals responsive to the distortion in each of a plurality of spectral bands, over a succession of time intervals,

means for generating a measure of the subjective impact of the distortion due to the telecommunications apparatus, said measure of subjective impact changing as a function of the distribution of the distortion over time and over said spectral bands, and

means for generating a measure of the total amount of said distortion over a predetermined time segment, and to provide a quantitative measure of said subjective impact based on both said measure of distribution of the distortion and said total amount of distortion.

2. Apparatus as in claim 1, in which a measure of distortion distribution E_E is related to the sum over all time intervals (i) and spectral bands (j) of the value:

-a(i,j)·

ln (a(i,j)), where a(i,j) is the absolute magnitude of distortion in a predetermined time interval (i), and spectral band (j), expressed as a proportion of the total distortion over all time intervals (i) and spectral bands (j).

3. Apparatus as in claim 1, in which the means for generating a measure of the subjective impact of distortion estimates the extent to which the distortion will be perceptible to a human listener.

4. Apparatus as in claim 3 in which the means for generating a measure of the subjective impact of distortion performs spectral and temporal masking calculations.

5. Apparatus as in claim 1, in which the means for generating a measure of the subjective impact of distortion performs a pitch analysis upon said distorted signal, in which said spectral bands comprise pitch bands.

6. Apparatus as in claim 1, further comprising a signal generator for supplying a test signal which has a spectral resemblance to human speech.

7. Apparatus as in claim 6, in which said test signal does not correspond to a single speaker conveying intelligent content speech.

8. Apparatus as in claim 6 in which:

the signal generator generates a test signal which comprises a sequence formed of a predetermined, small, number of speech segments,

the speech signal comprising several different portions including said segments such that said segments are represented in several different temporal contexts within said sequence, so as to vary the effects on each segment of time varying distortions in the telecommunications apparatus.

9. Apparatus as in claim 1, in which the means for generating a measure of the subjective impact of distortion analyzes the distorted signal, and forms, for each spectral band over each time interval, a measure of the difference between the distorted signal and the test signal.

10. Telecommunications testing apparatus arranged to receive a distorted signal which corresponds to a test signal when distorted by telecommunications apparatus to be tested, the testing apparatus comprising:

means for periodically deriving, from the distorted signal, a plurality of spectral component signals responsive to the distortion in each of a plurality of spectral bands, over a succession of time intervals,

means for generating a measure of the subjective impact of the distortion due to the telecommunications apparatus, said measure of subjective impact changing as a function of the distribution of the distortion over time and over said spectral bands, and

means for generating a measure of temporal correlation between distortion and the test signal, and to generate said subjective impact measure in dependence upon said temporal correlation measure.

11. Apparatus as in claim 10 wherein said means for generating a measure of temporal correlation utilizes a temporally displaced version of the test signal.

12. Telecommunications testing apparatus arranged to receive a distorted signal which corresponds to a test signal when distorted by telecommunications apparatus to be tested, the testing apparatus comprising:

means for periodically deriving, from the distorted signal, a plurality of spectral component signals responsive to the distortion in each of a plurality of spectral bands, over a succession of time intervals,

means for generating a measure of the subjective impact of the distortion due to the telecommunications apparatus, said measure of subjective impact changing as a function of the distribution of the distortion over time and over said spectral bands, and

said time intervals being longer for lower frequency spectral component signals than for higher frequency spectral component signals.

13. A method of analysing the output of a speech signal handling apparatus to derive a measure of the audibility of distortion generated thereby, the method comprising:

providing a predetermined test signal;

analysing the distorted signal corresponding to the test signal when distorted by the apparatus, using a digital electronic distortion calculation apparatus; and

generating an indication of the subjective impact of said distortion based on said analysis;

said step of analysing including deriving a distribution measure of the spectral and temporal distribution of said distortion, deriving a total measure of the total amount of distortion, and said distribution measure and said total measure being both used to derive said subjective impact measurement.

14. A method as in claim 13 in which a measure of the extent to which the distortion will be perceptible to a human listener is derived.

15. A method as in claim 14 in which the analysis step comprises spectral and temporal masking calculations.

16. A method as in claim 13 in which a pitch analysis is performed on the distorted signal.

17. A method as in claim 13 in which a signal having a spectral resemblance to human speech is used as a test signal.

18. A method as in claim 17 in which the test signal does not correspond to a single speaker conveying intelligent content speech.

19. A method as in claim 17 in which:

the test signal comprises a sequence formed of a predetermined, small, number of speech segment, and

the speech signal comprises several different portions including said segments such that said segments are represented in several different temporal contexts within said sequence, so as to vary the effects on each segment of time varying distortions in the telecommunications apparatus.

20. A method as in claim 13 in which the measure of distortion is determined by analysing the distorted signal, and forming, for each spectral band over each time interval, a measure of the difference between the distorted signal and the test signal.

21. A method of analyzing the output of a speech signal handling apparatus to derive a measure of the audibility of distortion generated thereby the method comprising:

providing a predetermined test signal;

analyzing the distorted signal corresponding to the test signal when distorted by the apparatus using a digital electronic distortion calculation apparatus;

generating an indication of the subjective impact of said distortion based on said analysis;

said step of analyzing including deriving a measure of the spectral and temporal distribution of said distortion, and said distribution measure being used to derive said subjective impact measurement;

measure of distortion E_E being determined as the sum over all time intervals (i), and spectral bands (j), of the value;

-a(i,j)·

ln (a(i,j)), where a(i,j) is the absolute magnitude of distortion in a predetermined time interval (i), and spectral band (j), expressed as a proportion of the total distortion over all time intervals (i) and spectral bands (j).

22. A method of analyzing the output of a speech signal handling apparatus to derive a measure of the audibility of distortion generated thereby, the method comprising:

providing a predetermined test signal;

analyzing the distorted signal corresponding to the test signal when distorted by the apparatus, using a digital electronic distortion calculation apparatus;

generating an indication of the subjective impact of said distortion based on said analysis;

said step of analyzing including deriving a measure of the spectral and temporal distribution of said distortion and said distribution measure being used to derive said subjective impact measurement; and

deriving a measure of the temporal correlation between the distortion and the original signal.

23. A method of analyzing the output of a speech signal handling apparatus to derive a measure of the audibility of distortion generated thereby, the method comprising:

providing a predetermined test signal;

analyzing the distorted signal corresponding to the test signal when distorted by the apparatus, using a digital electronic distortion calculation apparatus;

generating an indication of the subjective impact of said distortion based on said analysis;

said step of analyzing including deriving a measure of the spectral and temporal distribution of said distortion and said distribution measure being used to derive said subjective impact measurement;

deriving a measure of the total amount of distortion over a predetermined period of time, and

the input signal being assessed over a plurality of time intervals, said time intervals being longer for lower frequency spectral component signals than for higher frequency spectral component signals.

24. A method for quantitatively measuring distortion in speech handling circuits, said method comprising the steps of:

inputting a test signal into a speech handling circuit;

generating an intermediate quantitative measure of respectively corresponding speech signal distortion in an output of the speech handling circuit as a multi-dimensional function of both time and frequency;

generating a distortion error entropy measurement based on said intermediate multi-dimensional measure, said distortion error entropy measurement also varying as a function of the distribution of measured distortion over time and frequency even when the total aggregate of measured distortion over a time and frequency remains constant.

25. A method as in claim 24 further comprising:

generating an estimate of subjective mean opinion scores for the measured distortion by forming a weighted sum of the distortion error entropy measurement and the total aggregate measured distortion over time and frequency.

26. A method as in claim 24, further comprising the step of deriving a measure of the total amount of distortion over a predetermined period of time.

27. A method as in claim 26 in which a measure of distribution E_E is determined as the sum over all time intervals (i), and spectral bands (j), of the value:

-a(i,j)·

ln (a(i,j)), where a(i,j) is the absolute magnitude of the distortion in a predetermined time interval (i), and

spectral band (j), expressed as a proportion of the total distortion over all time intervals (i) and spectral bands (j).

28. A method as in claim 24 further comprising the step of deriving a measure of the temporal correlation between the distortion and the original signal.

29. A method as in claim 24 in which a measure of the extent to which the distortion will be perceptible to a human listener is derived.

30. A method as in claim 29 in which the analysis step comprises spectral and temporal masking calculations.

31. A method as in claim 24 in which a pitch analysis is performed on the distorted signal.

32. A method as in claim 24 in which a signal having a spectral resemblance to human speech is used as a test signal.

33. A method as in claim 32 in which the test signal does not correspond to a single speaker conveying intelligent content speech.

34. A method as in claim 32 in which:

the test signal comprises a sequence formed of a predetermined, small, number of speech segment, and

the speech signal comprises several different portions including said segments such that said segments are represented in several different temporal contexts within said sequence, so as to vary the effects on each segment of time varying distortions in the telecommunications apparatus.

35. A method as in claim 24 in which the measure of distortion is determined by analyzing the distorted signal, and forming, for each spectral band over each time interval, a measure of the difference between the distorted signal and the test signal.

36. A method as in claim 24 in which the input signal is assessed over a plurality of time intervals, said time intervals being longer for lower frequency spectral component signals than for higher frequency spectral component signals.

37. Apparatus for quantitatively distortion measuring in speech handling circuits, said apparatus comprising:

means for inputting a test signal into a speech handling circuit;

means for generating an intermediate quantitative measure of respectively corresponding speech signal distortion in an output of the speech handling circuit as a multi-dimensional function of both time and frequency;

means for generating a distortion error entropy based on said intermediate multi-dimensional measure, said distortion error entropy measurement also varying as a function of the distribution of measured distortion over time and frequency even when the total aggregate of measured distortion over a time and frequency remains constant.

38. Apparatus as in claim 37 further comprising:

means for generating an estimate of subjective mean opinion scores for the measured distortion by forming a weighted sum of the distortion error entropy measurement and the total aggregate measured distortion over time and frequency.

39. Apparatus as in claim 37 further comprising means for generating a measure of the total amount of said distortion over a predetermined time segment, and to provide a quantitative measure of said subjective impact based on both said measure of distribution of the distortion and said total amount of distortion.

40. Apparatus as in claim 39 in which a measure of distortion distribution E_E is related to the sum over all time intervals (i) and spectral bands (j) of the value:

-a(i,j)·

ln (a(i,j)), where a(i,j) is the absolute magnitude of distortion in a predetermined time interval (i), and spectral band (j), expressed as a proportion of the total distortion over all time intervals (i) and spectral bands (j).

41. Apparatus as in claim 37 further comprising means for generating a measure of temporal correlation between distortion and the test signal, and to generate said subjective impact measure in dependence upon said temporal correlation measure.

42. Apparatus as in claim 41 wherein said means for generating a measure of temporal correlation utilizes a temporally displaced version of the test signal.

43. Apparatus as in claim 37, in which the means for generating a measure of the subjective impact of distortion estimates the extent to which the distortion will be perceptible to a human listener.

44. Apparatus as in claim 43 in which the means for generating a measure of the subjective impact of distortion performs spectral and temporal masking calculations.

45. Apparatus as in claim 37 in which the means for generating a measure of the subjective impact of distortion performs a pitch analysis upon said distorted signal, in which said spectral bands comprise pitch bands.

46. Apparatus as in claim 37 further comprising a signal generator for supplying a test signal which has a spectral resemblance to human speech.

47. Apparatus as in claim 46, in which said test signal does not correspond to a single speaker conveying intelligent content speech.

48. Apparatus as in claim 46 in which:

the signal generator generates a test signal which comprises a sequence formed of a predetermined, small, number of speech segments,

the speech signal comprising several different portions including said segments such that said segments are presented in several different temporal contexts within said sequence, so as to vary the effects on each segment of time varying distortions in the telecommunications apparatus.

49. Apparatus as in claim 37 in which the means for generating a measure of the subjective impact of distortion analyzes the distorted signal, and forms, for each spectral band over each time interval, a measure of the difference between the distorted signal and the test signal.

50. Apparatus as in claim 37 in which the said time intervals are longer for lower frequency spectral component signals than for higher frequency spectral component signals.