Text-to-speech device, text-to-speech method, and computer program product

• US 9,830,904 B2
• Filed: 06/17/2016
• Issued: 11/28/2017
• Est. Priority Date: 12/20/2013
• Status: Active Grant

• Current Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation), Toshiba Digital Solutions Corporation (Toshiba Corporation)
• Original Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Inventors: Nasu, Yu, Tamura, Masatsune, Morinaka, Ryo, Morita, Masahiro
• Primary Examiner(s): AZAD, ABUL K
• Application Number:
  US15/185,259
• Publication Number:
  US 20160300564A1
• Time in Patent Office: 529 Days
• Field of Search: 704258-269
• CPC Class Codes: G10L 13/033 Voice editing, e.g. manipulating the voice of the synthesiser
• CPC Class Codes: G10L 13/06 Elementary speech units used in speech synthesisers; Concatenation rules
• CPC Class Codes: G10L 13/10 Prosody rules derived from text; Stress or intonation

Abstract

According to an embodiment, a text-to-speech device includes a context acquirer, an acoustic model parameter acquirer, a conversion parameter acquirer, a converter, and a waveform generator. The context acquirer is configured to acquire a context sequence affecting fluctuations in voice. The acoustic model parameter acquirer is configured to acquire an acoustic model parameter sequence that corresponds to the context sequence and represents an acoustic model in a standard speaking style of a target speaker. The conversion parameter acquirer is configured to acquire a conversion parameter sequence corresponding to the context sequence to convert an acoustic model parameter in the standard speaking style into one in a different speaking style. The converter is configured to convert the acoustic model parameter sequence using the conversion parameter sequence. The waveform generator is configured to generate a voice signal based on the acoustic model parameter sequence acquired after conversion.

Claims

1. A text-to-speech device comprising:

one or more processors configured to;

acquire a context sequence that is an information sequence affecting fluctuations in voice;

acquire an acoustic model parameter sequence corresponding to the context sequence, the acoustic model parameter sequence representing a standard speaking style of a target speaker;

acquire a conversion parameter sequence corresponding to the context sequence, the conversion parameter sequence being used in converting an acoustic model parameter in the standard speaking style into one in a speaking style different from the standard speaking style;

convert the acoustic model parameter sequence using the conversion parameter sequence; and

generate a voice signal based on the acoustic model parameter sequence acquired after conversion.

2. The device according to claim 1, wherein the context sequence includes at least a phoneme sequence.

3. The device according to claim 1, further comprising:

an acoustic model parameter storage configured to store a plurality of acoustic model parameters classified according to contexts and store first classification information used in determining one of the acoustic model parameters corresponding to a given context; and

a conversion parameter storage configured to store a plurality of conversion parameters classified according to contexts and store second classification information used in determining one of the conversion parameters corresponding to a given context,

wherein the one or more processors is configured to;

determine, based on the first classification information stored in the acoustic model parameter storage, the acoustic model parameter sequence corresponding to the acquired context sequence, and

determine, based on the second classification information stored in the conversion parameter storage, the conversion parameter sequence corresponding to the acquired context sequence.

4. The device according to claim 3, wherein the conversion parameter is created using voice samples uttered by a certain speaker in a standard speaking style and voice samples uttered by the same speaker in a different speaking style from the standard speaking style.

5. The device according to claim 3, wherein

the acoustic model parameter is created using voice samples uttered by the target speaker, and

the conversion parameter is created using voice samples uttered by a speaker different from the target speaker.

6. The device according to claim 3, wherein

the acoustic model parameter is created using voice samples uttered by the target speaker in a speaking style expressing neutral feeling, and

the conversion parameter represents information used in converting an acoustic model parameter of the speaking style expressing neutral feeling into one expressing a feeling other than neutral.

7. The device according to claim 1, wherein

the acoustic model is a probabilistic model in which output probabilities of respective phonetic parameters that represent characteristics of a voice are expressed using Gaussian distribution,

the acoustic model parameter includes a mean vector representing a mean of an output probability distribution of each phonetic parameter,

the conversion parameter represents a vector having the same dimensionality as the mean vector included in the acoustic model parameter, and

the one or more processors is further configured to add a conversion parameter included in the conversion parameter sequence to a mean vector included in the acoustic model parameter sequence to generate a post-conversion acoustic model parameter sequence.

8. The device according to claim 1, further comprising:

a plurality of conversion parameter storages configured to store conversion parameters corresponding to mutually different speaking styles,

wherein the one or more processors is further configured to;

select one of the plurality of conversion parameter storages, and

acquire the conversion parameter sequence from the selected conversion parameter storage.

9. The device according to claim 1, further comprising:

a plurality of conversion parameter storages configured to store conversion parameters corresponding to mutually different speaking styles,

wherein the one or more processors is further configured to;

select two or more of the plurality of conversion parameter storages, wherein

acquire the conversion parameter sequence from each of the selected two or more conversion parameter storages, and

convert the acoustic model parameter sequence using the two or more conversion parameter sequences.

10. The device according to claim 9,

wherein the one or more processors is further configured to:

control ratios at which the respective conversion parameters acquired from the selected two or more of the conversion parameter storages are to be reflected in the acoustic model parameters.

11. The device according to claim 1, further comprising:

a plurality of acoustic model parameter storages configured to store the acoustic model parameters corresponding to mutually different speakers,

wherein the one or more processors is further configured to;

select one of the plurality of acoustic model parameter storages, and

acquire the acoustic model parameter sequence from the selected acoustic model parameter storage.

12. The device according to claim 11,

wherein the one or more processors is further configured to convert the acoustic model parameter stored in one of the acoustic model parameter storages into the acoustic model parameter corresponding to a specific speaker using speaker adaptation, and write the acoustic model parameter acquired by conversion in the acoustic model parameter storage corresponding to the specific speaker.

13. A text-to-speech method comprising:

acquiring by one or more processors, a context sequence that is an information sequence affecting fluctuations in voice;

acquiring by the one or more processors, an acoustic model parameter sequence corresponding to the context sequence, the acoustic model parameter sequence representing an acoustic model in a standard speaking style of a target speaker;

acquiring by the one or more processors, a conversion parameter sequence corresponding to the context sequence, the conversion parameter sequence being used in converting an acoustic model parameter in the standard speaking style into one in a speaking style different from the standard speaking style;

converting by the one or more processors, the acoustic model parameter sequence using the conversion parameter sequence; and

generating by the one or more processors, a voice signal based on the acoustic model parameter sequence acquired after conversion.

14. A computer program product comprising a non-transitory computer-readable medium containing a program executed by a computer, the program causing the computer to execute:

acquiring a context sequence that is an information sequence affecting fluctuations in voice;

acquiring an acoustic model parameter sequence corresponding to the context sequence, the acoustic model parameter sequence representing an acoustic model in a standard speaking style of a target speaker;

acquiring a conversion parameter sequence corresponding to the context sequence, the conversion parameter sequence being used in converting an acoustic model parameter in the standard speaking style into one in a speaking style different from the standard speaking style;

converting the acoustic model parameter sequence using the conversion parameter sequence; and

generating a voice signal based on the acoustic model parameter sequence acquired after conversion.