Voice quality conversion apparatus, pitch conversion apparatus, and voice quality conversion method

1. A voice quality conversion apparatus that converts voice quality of an input speech, said apparatus comprising:

a fundamental frequency converting unit configured to calculate a weighted sum of a fundamental frequency of an input sound source waveform and a fundamental frequency of a target sound source waveform at a predetermined conversion ratio as a resulting fundamental frequency, the input sound source waveform representing sound source information of an input speech waveform, and the target sound source waveform representing sound source information of a target speech waveform;

a low-frequency spectrum calculating unit configured to calculate a low-frequency sound source spectrum by mixing a level of a harmonic of the input sound source waveform and a level of a harmonic of the target sound source waveform at the predetermined conversion ratio for each order of harmonics including fundamental, using an input sound source spectrum and a target sound source spectrum in a frequency range equal to or lower than a boundary frequency determined depending on the resulting fundamental frequency calculated by said fundamental frequency converting unit, the low-frequency sound source spectrum having levels of harmonics in which the resulting fundamental frequency is set to a fundamental frequency of the low-frequency sound source spectrum, the input sound source spectrum being a sound source spectrum of an input speech, and the target sound source spectrum being a sound source spectrum of a target speech;

a high-frequency spectrum calculating unit configured to calculate a high-frequency sound source spectrum by mixing the input sound source spectrum and the target sound source spectrum at the predetermined conversion ratio in a frequency range larger than the boundary frequency;

a spectrum combining unit configured to combine the low-frequency sound source spectrum with the high-frequency sound source spectrum at the boundary frequency to generate a sound source spectrum for an entire frequency range; and

a synthesis unit configured to generate a synthesized speech waveform using the sound source spectrum for the entire frequency range.

2. The voice quality conversion apparatus according to claim 1 ,

wherein the boundary frequency is set higher as the resulting fundamental frequency is higher.

3. The voice quality conversion apparatus according to claim 2 ,

wherein the boundary frequency is a frequency corresponding to a critical bandwidth matching a value of the resulting fundamental frequency, the critical bandwidth being a frequency bandwidth (i) which varies depending on a frequency and (ii) in which two sounds at different frequencies in a same frequency range are perceived by a human ear as a single sound obtained by adding intensities of the two sounds.

4. The voice quality conversion apparatus according to claim 1 ,

wherein said low-frequency spectrum calculating unit is further configured to hold rule data for determining a boundary frequency using a fundamental frequency, and to determine, using the rule data, a boundary frequency corresponding to the resulting fundamental frequency calculated by said fundamental frequency converting unit.

5. The voice quality conversion apparatus according to claim 4 ,

wherein the rule data indicates a relationship between a frequency and a critical bandwidth, and

said low-frequency spectrum calculating unit is further configured to determine, as a boundary frequency and using the rule data, a frequency corresponding to a critical bandwidth matching a value of the resulting fundamental frequency calculated by said fundamental frequency converting unit.

6. The voice quality conversion apparatus according to claim 1 ,

wherein said low-frequency spectrum calculating unit is further configured to calculate a level of a harmonic by mixing the level of the harmonic of the input sound source waveform and the level of the harmonic of the target sound source waveform at the predetermined conversion ratio for each order of the harmonics including the fundamental in the frequency range equal to or lower than the boundary frequency, and to calculate the low-frequency sound source spectrum by determining the calculated level of the harmonic as the level of the harmonic of the low-frequency sound source spectrum at a frequency of a harmonic calculated using the resulting fundamental frequency.

7. The voice quality conversion apparatus according to claim 1 ,

wherein said low-frequency spectrum calculating unit is further configured to calculate the low-frequency sound source spectrum in the frequency range equal to or lower than the boundary frequency by interpolating a level of the low-frequency sound source spectrum at a first frequency other than a frequency of a harmonic calculated using the resulting fundamental frequency, using a level of a harmonic at a frequency adjacent to the first frequency in the low-frequency sound source spectrum.

8. The voice quality conversion apparatus according to claim 1 ,

wherein said low-frequency spectrum calculating unit is further configured to calculate the low-frequency sound source spectrum in the frequency range equal to or lower than the boundary frequency by transforming the input sound source spectrum and the target sound source spectrum into another input sound source spectrum and an output sound source spectrum, respectively, so that each of the fundamental frequency of the input sound source spectrum and the fundamental frequency of the target sound source spectrum matches the resulting fundamental frequency, and mixing the other input sound source spectrum and the output sound source spectrum at the predetermined conversion ratio.

9. The voice quality conversion apparatus according to claim 1 ,

wherein said high-frequency spectrum calculating unit is configured to calculate the high-frequency sound source spectrum by calculating a weighted sum of a spectral envelope of the input sound source spectrum and a spectral envelope of the target sound source spectrum at the predetermined conversion ratio in the frequency range larger than the boundary frequency.

10. The voice quality conversion apparatus according to claim 9 , further comprising

a sound source spectrum calculating unit configured to calculate an input sound source spectrum and a target sound source spectrum using a waveform obtained by multiplying a first window function by the input sound source waveform and a waveform obtained by multiplying a second window function by the target sound source waveform, respectively, and to calculate the spectral envelope of the input sound source spectrum and the spectral envelope of the target sound source spectrum using the calculated input sound source spectrum and the calculated target sound source spectrum, respectively.

11. The voice quality conversion apparatus according to claim 10 ,

wherein the first window function is a window function having a length that is double a fundamental period of the input sound source waveform, and

the second window function is a window function having a length that is double a fundamental period of the target sound source waveform.

12. The voice quality conversion apparatus according to claim 1 ,

wherein said high-frequency spectrum calculating unit is configured to calculate the high-frequency sound source spectrum in the frequency range larger than the boundary frequency by calculating a difference between a spectral tilt of the input sound source spectrum and a spectral tilt of the target sound source spectrum, and transforming the input sound source spectrum using the calculated difference.

13. The voice quality conversion apparatus according to claim 1 ,

wherein the input speech waveform and the target speech waveform are speech waveforms of a same phoneme.

14. The voice quality conversion apparatus according to claim 13 ,

wherein the input speech waveform and the target speech waveform are the speech waveforms of the same phoneme and at a same temporal position within the same phoneme.

15. The voice quality conversion apparatus according to claim 1 , further comprising

a fundamental frequency calculating unit configured to extract feature points repeatedly appearing at fundamental period intervals of each of the input sound source waveform and the target sound source waveform, and to calculate the fundamental frequency of the input sound source waveform and the fundamental frequency of the target sound source waveform using corresponding ones of the fundamental period intervals of the extracted feature points.

16. The voice quality conversion apparatus according to claim 15 ,

wherein each of the feature points is a glottal closure instant (GCI).

17. A voice quality conversion method of converting voice quality of an input speech, said method comprising:

calculating a weighted sum of a fundamental frequency of an input sound source waveform and a fundamental frequency of a target sound source waveform at a predetermined conversion ratio as a resulting fundamental frequency, the input sound source waveform representing sound source information of an input speech waveform, and the target sound source waveform representing sound source information of a target speech waveform;

calculating a low-frequency sound source spectrum by mixing a level of a harmonic of the input sound source waveform and a level of a harmonic of the target sound source waveform at the predetermined conversion ratio for each order of harmonics including fundamental, using an input sound source spectrum and a target sound source spectrum in a frequency range equal to or lower than a boundary frequency corresponding to the resulting fundamental frequency calculated in said calculating a weighted sum, the low-frequency sound source spectrum having levels of harmonics in which the resulting fundamental frequency is set to a fundamental frequency of the low-frequency sound source spectrum, the input sound source spectrum being a sound source spectrum of an input speech, and the target sound source spectrum being a sound source spectrum of a target speech;

calculating a high-frequency sound source spectrum by mixing the input sound source spectrum and the target sound source spectrum at the predetermined conversion ratio in a frequency range larger than the boundary frequency;

combining the low-frequency sound source spectrum with the high-frequency sound source spectrum at the boundary frequency to generate a sound source spectrum for an entire frequency range; and

generating a synthesized speech waveform using the sound source spectrum for the entire frequency range.

18. A program for converting voice quality of an input speech recorded on a non-transitory computer-readable recording medium, said program causing a computer to execute:

calculating a weighted sum of a fundamental frequency of an input sound source waveform and a fundamental frequency of a target sound source waveform at a predetermined conversion ratio as a resulting fundamental frequency, the input sound source waveform representing sound source information of an input speech waveform, and the target sound source waveform representing sound source information of a target speech waveform;

calculating a low-frequency sound source spectrum by mixing a level of a harmonic of the input sound source waveform and a level of a harmonic of the target sound source waveform at the predetermined conversion ratio for each order of harmonics including fundamental, using an input sound source spectrum and a target sound source spectrum in a frequency range equal to or lower than a boundary frequency corresponding to the resulting fundamental frequency calculated in the calculating a weighted sum, the low-frequency sound source spectrum having levels of harmonics in which the resulting fundamental frequency is set to a fundamental frequency of the low-frequency sound source spectrum, the input sound source spectrum being a sound source spectrum of an input speech, and the target sound source spectrum being a sound source spectrum of a target speech;

calculating a high-frequency sound source spectrum by mixing the input sound source spectrum and the target sound source spectrum at the predetermined conversion ratio in a frequency range larger than the boundary frequency;

combining the low-frequency sound source spectrum with the high-frequency sound source spectrum at the boundary frequency to generate a sound source spectrum for an entire frequency range; and

generating a synthesized speech waveform using the sound source spectrum for the entire frequency range.