Feature sequence generating device, feature sequence generating method, and feature sequence generating program

Patent metadata

Field	Value
Publication number	US-9299338-B2
Application number	US-201113880630-A
Country	US
Kind code	B2
Filing date	Oct 28, 2011
Priority date	Nov 8, 2010
Publication date	Mar 29, 2016
Grant date	Mar 29, 2016

Abstract

Spread level parameter correcting means 501 receives a contour parameter as information representing the contour of a feature sequence (a sequence of features of a signal considered as the object of generation) and a spread level parameter as information representing the level of a spread of the distribution of the features in the feature sequence. The spread level parameter correcting means 501 corrects the spread level parameter based on a variation of the contour parameter represented by a sequence of the contour parameters. Feature sequence generating means 502 generates the feature sequence based on the contour parameters and the corrected spread level parameters.

First claim

What is claimed is:

1. A feature sequence generating device including a processor, comprising:
a spread level parameter correcting unit, implemented by the processor, which corrects a spread level parameter which represents a level of a spread of distribution of features in a feature sequence of the speech signal features based on a contour parameter variation represented by a sequence of contour parameters which represent the contour of the feature sequence respectively; and
a feature sequence generating unit, implemented by the processor, which generates the feature sequence based on the contour parameter and the spread level parameter corrected by the spread level parameter correcting unit.

2. The feature sequence generating device according to claim 1 , wherein the spread level parameter correcting unit corrects the spread level parameter so that a value of the spread level parameter increases with increase in the contour parameter variation.

3. The feature sequence generating device according to claim 1 , wherein the spread level parameter correcting unit determines a provisional correction value based on the contour parameter variation and determines a corrected spread level parameter based on the original spread level parameter and the provisional correction value.

4. The feature sequence generating device according to claim 3 , wherein the spread level parameter correcting unit outputs a not corrected spread level parameter when a difference value between the provisional correction value and a value of the original spread level parameter is less than a prescribed threshold value or when a ratio of the provisional correction value to the value of the original spread level parameter is less than a prescribed threshold value.

5. The feature sequence generating device according to claim 1 , wherein:
the contour parameter is a parameter included in HMM parameters acquired by the modeling of information on features and representing a statistic selected from a mean, a median, a mode, a maximum value and a minimum value of output probability distribution, and
the spread level parameter is a variance parameter included in the HMM parameters and representing variance of the output probability distribution, and
the spread level parameter correcting unit corrects preferentially the variance parameter corresponding to short state duration among variance parameters based on state duration in the HMM.

6. The feature sequence generating device according to claim 1 , wherein:
the feature sequence generating device generates a pitch pattern, formed as a sequence of pitch frequencies of speech, as the feature sequence, and the contour parameter represents a contour of the pitch pattern and the spread level parameter represents a level of a spread of distribution of the pitch frequencies, and
the spread level parameter correcting unit corrects preferentially the spread level parameter corresponding to short duration among spread level parameters based on the contour parameter and duration of each phoneme.

7. The feature sequence generating device according to claim 1 , wherein:
the feature sequence generating device generates a pitch pattern, formed as a sequence of pitch frequencies of speech, as the feature sequence, and the contour parameter represents a contour of the pitch pattern and the spread level parameter represents a level of a spread of distribution of the pitch frequencies, and
the spread level parameter correcting unit sets degree of correction of the spread level parameter lower in parts where the pitch tends to change sharply in comparison with the degrees in the other parts based on language information on the speech.

8. A non-transitory computer readable information recording medium storing a feature sequence generating program for causing a computer to execute:
a process of correcting a spread level parameter representing a level of a spread of distribution of features in a feature sequence of the speech signal features based on a contour parameter variation represented by a sequence of contour parameters which represent the contour of the feature sequence, respectively; and
a process of generating the feature sequence based on the contour parameter and corrected spread level parameter.

9. The feature sequence generating device according to claim 2 , wherein the spread level parameter correcting unit determines a provisional correction value based on the contour parameter variation and determines a corrected spread level parameter based on the original spread level parameter and the provisional correction value.

10. The feature sequence generating device according to claim 2 , wherein:
the feature sequence generating device generates a pitch pattern, formed as a sequence of pitch frequencies of speech, as the feature sequence, and the contour parameter represents a contour of the pitch pattern and the spread level parameter represents a level of a spread of distribution of the pitch frequencies, and
the spread level parameter correcting unit corrects preferentially the spread level parameter corresponding to short duration among spread level parameters based on the contour parameter and duration of each phoneme.

11. The feature sequence generating device according to claim 3 , wherein:
the feature sequence generating device generates a pitch pattern, formed as a sequence of pitch frequencies of speech, as the feature sequence, and the contour parameter represents a contour of the pitch pattern and the spread level parameter represents a level of a spread of distribution of the pitch frequencies, and
the spread level parameter correcting unit corrects preferentially the spread level parameter corresponding to short duration among spread level parameters based on the contour parameter and duration of each phoneme.

12. The feature sequence generating device according to claim 4 , wherein:
the feature sequence generating device generates a pitch pattern, formed as a sequence of pitch frequencies of speech, as the feature sequence, and the contour parameter represents a contour of the pitch pattern and the spread level parameter represents a level of a spread of distribution of the pitch frequencies, and
the spread level parameter correcting unit corrects preferentially the spread level parameter corresponding to short duration among spread level parameters based on the contour parameter and duration of each phoneme.

13. The feature sequence generating device according to claim 5 , wherein:
the feature sequence generating device generates a pitch pattern, formed as a sequence of pitch frequencies of speech, as the feature sequence, and the contour parameter represents a contour of the pitch pattern and the spread level parameter represents a level of a spread of distribution of the pitch frequencies, and
the spread level parameter correcting unit corrects preferentially the spread level parameter corresponding to short duration among spread level parameters based on the contour parameter and duration of each phoneme.

14. The feature sequence generating device according to claim 2 , wherein:
the feature sequence generating device generates a pitch pattern, formed as a sequence of pitch frequencies of speech, as the feature sequence, and the contour parameter represents a contour of the pitch pattern and the spread level parameter represents a level of a spread of distribution of the pitch frequencies, and
the spread level parameter correcting unit sets degree of correction of the spread level parameter lower in parts where the pitch tends to change sharply in comparison with the degrees in the other parts based on la

Assignees

• Kato Masanori
• Nec Corp

Inventors

• Kato Masanori

Classifications

• G10L13/033 Primary

  Voice editing, e.g. manipulating the voice of the synthesiser · CPC title

• G10L15/10

  using distance or distortion measures between unknown speech and reference templates · CPC title

• G10L13/08 Primary

  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 · CPC title

• G10L13/0335

  Pitch control · CPC title

• G10L15/04

  Segmentation; Word boundary detection · CPC title