System and method for determining pitch contours

• US 5,790,978 A
• Filed: 09/15/1995
• Issued: 08/04/1998
• Est. Priority Date: 09/15/1995
• Status: Expired due to Term

• Current Assignee: Lucent Technologies, Inc. (Nokia Corporation)
• Original Assignee: Lucent Technologies, Inc. (Nokia Corporation)
• Inventors: Olive, Joseph Philip, VanSanten, Jan Pieter
• Primary Examiner(s): MacDonald, Allen R.
• Assistant Examiner(s): COLLINS, ALPHONSO
• Application Number:
  US08/528,576
• Time in Patent Office: 1,054 Days
• Field of Search: 395/2.09, 395/2.14, 395/2.15, 395/2.16, 395/2.18, 395/2.4, 395/2.43, 395/2.45, 395/2.44, 395/2.69
• US Class Current: 704/207
• CPC Class Codes: G10L 13/04 Details of speech synthesis systems, e.g. synthesiser structure or memory management
• CPC Class Codes: G10L 13/08 Text analysis or generation of parameters for speech synthesis out of text, e.g. grapheme to phoneme translation, prosody generation or stress or intonation determination

Abstract

A system and method are provided for automatically computing local pitch contours from textual input to produce pitch contours that closely mimic those found in natural speech. The methodology of the invention incorporates parameterized equations whose parameters can be estimated directly from natural speech recordings. That methodology incorporates a model based on the premise that pitch contours instantiating a particular pitch contour class can be described as distortions in the temporal and frequency domains of a single, underlying contour. After the nature of the pitch contour for different pitch contour classes has been established, a pitch contour can be predicted that closely models a natural speech contour for a synthetic speech utterance by adding the individual contours of the different intonational classes and adjusting the boundaries of these to match the boundaries of the adjacent intonation curves.

Claims

1. A method for determining an acoustical contour for a speech interval having a predetermined duration, said acoustical contour being functionally related to a speech waveform processed by a computerized speech processing application, said method comprising the steps of:

dividing said duration of said speech interval into a plurality of critical intervals;

determining a plurality of anchor times within said speech interval duration, said anchor times being functionally related to said critical intervals;

for each of said anchor times, finding a corresponding anchor value from a look-up table;

representing each of said anchor values as an ordinate in a Cartesian coordinate system having as an abscissa said corresponding anchor time;

fitting a curve to said Cartesian representations of said anchor values; and

multiplying said fitted curve by at least one predetermined numerical constant related to a linguistic factor to create a product curve, said product curve being representative of said acoustical contour;

wherein said acoustical contour is provided as an input to said speech processing application.

2. The method for determining an acoustical contour of claim 1 including the further step of adding said product curve to a pre-computed phrase curve to create an F₀ curve.

3. The method for determining an acoustical contour of claim 1 wherein said speech interval having a predetermined duration comprises an accent group.

4. The method for determining an acoustical contour of claim 1 wherein said acoustical contour is a pitch contour.

5. The method for determining an acoustical contour of claim 3 where said step of dividing said speech interval into a plurality of critical intervals produces three said critical intervals:

a first interval corresponding to the duration for initial consonants in a first syllable of said accent group, hereafter designated D₁, a second interval corresponding to the duration of phonemes in a remainder of said first syllable, hereafter designated D₂, and a third interval corresponding to the duration of phonemes in a remainder of said accent group after said first syllable, hereafter designated D₃.

6. The method for determining an acoustical contour of claim 5 wherein said relationship between said anchor times and said critical intervals is of the form:

space="preserve" listing-type="equation">T.sub.i =α

.sub.ic D.sub.1 +β

.sub.ic D.sub.2 +γ

.sub.ic D.sub.3

where α

, β

&

γ

are alignment parameters, i is an index for an anchor time under consideration and c refers to a phonetic class of said accent group.

7. The method for determining an acoustical contour of claim 6 where said alignment parameters are determined from actual speech data for a multiplicity of phonetic classes, and within each said class, for each of said plurality of anchor times.

8. The method for determining an acoustical contour of claim 1 wherein said plurality of anchor times is set equal to nine.

9. The method for determining an acoustical contour of claim 1 wherein said plurality of anchor times is set equal to fourteen.

10. The method for determining an acoustical contour of claim 1 wherein said anchor values in said look-up table are determined from an average of a plurality of accent curves obtained from natural speech, said averaged curve being divided along a temporal axis into a plurality of intervals corresponding to said plurality of said anchor times, and said anchor values being read from said averaged curve at a point corresponding to a terminal point for each said interval.

11. The method for determining an acoustical contour of claim 10 wherein said averaged curve for determining said anchor values is normalized to limit a numerical value of each of said anchor values to a range of 0 to 1.

12. The method for determining an acoustical contour of claim 1 including the further step of adding to said product curve at least one obstruent perturbation curve corresponding to an obstruent consonant in said speech interval.

13. The method for determining an acoustical contour of claim 12 wherein said obstruent perturbation curves are generated from a set of stored perturbation parameter corresponding to each obstruent consonant.

25. A computer-readable medium encoded with a computer program for estimation of an acoustical contour for a speech interval, said acoustical contour representing a parameter processed by an automated speech processing application, and said program carrying out essentially the steps of the method for determining such an acoustical contour of claim 1.

14. A system for determining an acoustical contour for a speech interval having a predetermined duration, wherein said acoustical contour is functionally related to a speech waveform processed by a computerized speech processing application, said system comprising:

processing means for dividing said duration of said speech interval into a plurality of critical intervals;

processing means for determining a plurality of anchor times within said speech interval duration, said anchor times being functionally related to said critical intervals;

means for finding an anchor value corresponding to each of said anchor times, said anchor values being stored in a storage means, for representing each of said anchor values as an ordinate in a Cartesian coordinate system having as an abscissa said corresponding anchor time, and for fitting a curve to said Cartesian representations of said anchor values; and

means for multiplying said fitted curve by at least one predetermined numerical constant related to a linguistic factor to create a product curve, said product curve being representative of said acoustical contour;

wherein said acoustical contour is provided as an input to said speech processing application.

15. The system for determining an acoustical contour of claim 14 further including summation means for adding said product curve to a pre-computed phrase curve to create an F₀ curve.

16. The system for determining an acoustical contour of claim 14 wherein said speech interval having a predetermined duration comprises an accent group.

17. The system for determining an acoustical contour of claim 14 wherein said acoustical contour is a pitch contour.

18. The system for determining an acoustical contour of claim 16 where said processing means for dividing said speech interval into a plurality of critical intervals operates to produce three said critical intervals:

a first interval corresponding to the duration for initial consonants in a first syllable of said accent group, hereafter designated D₁, a second interval corresponding to the duration of phonemes in a remainder of said first syllable, hereafter designated D₂, and a third interval corresponding to the duration of phonemes in a remainder of said accent group after said first syllable, hereafter designated D₃.

19. The system for determining an acoustical contour of claim 18 wherein said relationship between said anchor times and said critical intervals is of the form:

space="preserve" listing-type="equation">T.sub.i =α

.sub.ic D.sub.1 +β

.sub.ic D.sub.2 +γ

.sub.ic D.sub.3

where α

, β

&

γ

are alignment parameters, i is an index for an anchor time under consideration and c refers to a phonetic class of said accent group.

20. The system for determining an acoustical contour of claim 19 where said alignment parameters are determined from actual speech data for a multiplicity of phonetic classes, and within each said class, for each of said plurality of anchor times.

21. The system for determining an acoustical contour of claim 14 wherein said anchor values stored in said storage means are determined from an average of a plurality of accent curves obtained from natural speech, said averaged curve being divided along a temporal axis into a plurality of intervals corresponding to said plurality of said anchor times, and said anchor values being read from said averaged curve at a point corresponding to a terminal point for each said interval.

22. The system for determining an acoustical contour of claim 21 wherein said averaged curve for determining said anchor values is normalized to limit a numerical value of each of said anchor values to a range of 0 to 1.

23. The system for determining an acoustical contour of claim 14 further including a processing means for generating an obstruent perturbation curve corresponding to an obstruent consonant in said speech interval, and for adding at least one of said generated obstruent perturbation curve to said product curve.

24. The system for determining an acoustical contour of claim 23 wherein said obstruent perturbation curves are generated from a set of stored perturbation parameter corresponding to each obstruent consonant.