Segment information generation device, speech synthesis device, speech synthesis method, and speech synthesis program

• US 9,401,138 B2
• Filed: 05/10/2012
• Issued: 07/26/2016
• Est. Priority Date: 05/25/2011
• Status: Active Grant

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Kato, Masanori
• Primary Examiner(s): Guerra-Erazo, Edgar
• Application Number:
  US14/114,891
• Publication Number:
  US 20140067396A1
• Time in Patent Office: 1,538 Days
• Field of Search: 704/260, 704/261, 704/258, 704/254, 704/267, 704/266, 704/268, 704/269
• US Class Current: 1/1
• CPC Class Codes: G10L 13/00 Speech synthesis; Text to speech systems
• CPC Class Codes: G10L 13/06 Elementary speech units used in speech synthesisers; Concatenation rules

Abstract

A segment information generation device includes a waveform cutout unit cuts out a speech waveform from natural speech at a time period not depending on a pitch frequency of the natural speech. A feature parameter extraction unit extracts a feature parameter of a speech waveform from the speech waveform cut out by the waveform cutout unit. A time domain waveform generation unit generates a time domain waveform based on the feature parameter.

Claims

1. A segment information generation device comprising:

a waveform cutout unit implemented at least by hardware including a processor that cuts out a speech waveform from natural speech at a time period not depending on a pitch frequency of the natural speech, continuously;

a feature parameter extraction unit implemented at least by hardware including a processor that extracts a feature parameter of a speech waveform from the speech waveform cut out by the waveform cutout unit;

a time domain waveform generation unit implemented at least by hardware including a processor that generates a time domain waveform based on the feature parameter;

a spectrum shape change degree estimation unit implemented at least by hardware including a processor that estimates a degree of change in spectrum shape indicating a degree of change in spectrum shape of natural speech; and

a period control unit implemented at least by hardware including a processor that determines a time period to cut out a speech waveform from the natural speech based on the degree of change in spectrum shape.

2. The segment information generation device according to claim 1, wherein the period control unit determines the time period to cut out a speech waveform from natural speech based on attribute information of the natural speech.

3. The segment information generation device according to claim 1, wherein when a degree of change in spectrum shape is determined to be small, the period control unit sets a time period to cut out a speech waveform from natural speech to be longer than a time period during normal time.

4. The segment information generation device according to claim 1, wherein when a degree of change in spectrum shape is determined to be large, the period control unit sets a time period to cut out a speech waveform from natural speech to be shorter than a time period during normal time.

7. The segment information generation device according to claim 1, further comprising:

a spectrum shape change degree estimation unit comprising a processor that estimates a degree of change in spectrum shape indicating a degree of change in spectrum shape of natural speech.

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

a natural speech storage unit comprising a storage device that stores information indicating a natural speech waveform of the natural speech.

9. The segment information generation device according to claim 8, further comprising:

an attribute information storage unit comprising a storage device that stores, as attribute information, linguistic information indicating character strings corresponding to the natural speech, and prosody information of the natural speech.

10. The segment information generation device according to claim 9, wherein the linguistic information comprises information on at least one of pronunciation, syllable string, phoneme string, accent position, accent phrase separation and morphemic word class.

11. The segment information generation device according to claim 9, wherein the prosody information comprises at least one of pitch frequency, amplitude, short-time power time series, and duration of respective syllables, phonemes and pauses contained in the natural speech.

12. The segment information generation device according to claim 1, further comprising:

a segment information storage unit comprising a storage device that stores segment information indicating a segment and comprising the time domain waveform.

5. A speech synthesis device comprising:

a waveform cutout unit implemented at least by hardware including a processor that cuts out a speech waveform from natural speech at a time period not depending on a pitch frequency of the natural speech, continuously;

a feature parameter extraction unit implemented at least by hardware including a processor that extracts a feature parameter of a speech waveform from the speech waveform cut out by the waveform cutout unit;

a time domain waveform generation unit implemented at least by hardware including a processor that generates a time domain waveform based on the feature parameter;

a segment information storage unit implemented by a storage device that stores segment information indicating a segment and containing the time domain waveform;

a segment information selection unit implemented at least by hardware including a processor that selects segment information corresponding to an input character string;

a waveform generation unit implemented at least by hardware including a processor and that generates a speech synthesis waveform by use of the segment information selected by the segment information selection unit;

a spectrum shape change degree estimation unit implemented at least by hardware including a processor that estimates a degree of change in spectrum shape indicating a degree of change in spectrum shape of natural speech; and

a period control unit implemented at least by hardware including a processor that determines a time period to cut out a speech waveform from the natural speech based on the degree of change in spectrum shape.

6. A segment information generating method, implemented by a processor, comprising:

cutting out a speech waveform from natural speech at a time period not depending on a pitch frequency of the natural speech, continuously;

extracting a feature parameter of the speech waveform from the speech waveform;

generating a time domain waveform based on the feature parameter;

estimating a degree of change in spectrum shape indicating a degree of change in spectrum shape of natural speech; and

determining a time period to cut out a speech waveform from the natural speech based on the degree of change in spectrum shape.