JPH04234100A - Voice synthesizing method and device - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesis method and apparatus.

0002

[Prior Art] Generally, the same waveform appears periodically in an audio signal such as a human voice, so this period is called a pitch, and the waveform that appears at each pitch is called a segment waveform. It is used in synthesis equipment. In other words, in a speech synthesis device, the data of the required segment waveform is set in advance in a memory, which is a waveform storage means, and the data of the predetermined segment waveform is sequentially selected and edited and synthesized in the time domain. Synthesized speech is output according to the edited and synthesized segment waveform data. However, in such a speech synthesis device, if the phonemes of the segment waveforms of continuously generated speech signals differ greatly, the discontinuity will result in the output of a harsh synthesized speech.

[0003] Therefore, JASA VOL1.45 (196
9) “Speech Analysis-Synth
esis System Based on
Homorphic Filtering"A.V.
There are those disclosed in Oppenheim and Japanese Patent Application Laid-Open No. 61-256398. For example, JASA
In the speech synthesis device disclosed in VOL 1.45, data of two different segment waveforms are linearly interpolated on the time axis in order to improve the auditory continuity of synthesized speech. Furthermore, in the speech synthesis device disclosed in Japanese Patent Application Laid-Open No. 61-256398, the structure of the device is simplified by limiting the interpolated segment waveform data to be of the same phase, in substantially the same way as the above device.

[0004] By doing so, even when audio signals having greatly different segmental waveforms are continuously generated, the auditory continuity is good due to the complemented audio signals.

[0005]

SUMMARY OF THE INVENTION In the above-mentioned speech synthesis apparatus, the perceptual characteristics are improved by supplementing a predetermined speech signal between speech signals having significantly different segmental waveforms.

[0006] Here, the calculation formula for linearly interpolating the segment waveform in this way is, for example, F(nT)=F1(nT)+{F2(nT)-
F1(nT)}·m/M (m=1, M). Note that M in the above equation is a complementary interval, which means that M segment waveforms are inserted between the segment waveform F1 and the segment waveform F2. However, such an operation that includes division complicates the structure of the device, so in the speech synthesis device of Japanese Patent Application Laid-Open No. 61-256398, m/M (m=1, M) is used. The structure of the device is simplified by calculating the calculation results in advance and recording them in a table.

[0007] In a speech synthesis device that linearly interpolates elemental waveforms as described above, the interpolation interval M is set to a large value for vowels and the like whose elemental waveforms change slowly, and for consonants and the like whose elemental waveforms change rapidly. Therefore, it is desirable to set it to a small value. However, in the above-mentioned speech synthesis device, the complementary interval M is fixedly set, and it is not assumed that the complementary interval M is changed in response to the rate of change in phoneme as described above.

[0008] In other words, in the conventional speech synthesis device, the
As illustrated in a), the interpolation interval M is set in advance to be small corresponding to the part where the change in the elemental waveform F is sudden, and the interpolation interval M is set small in advance,
As illustrated in b), in parts where the waveform change F is slow, the segment waveform F' is complemented at small interpolation intervals M.
This is undesirable because the amount of data of this elemental waveform F' increases.

[0009]

[Means for solving the problem] The invention according to claim 1 includes:
The data of a plurality of segment waveforms obtained by extracting the waveform of the audio signal in pitch units is set in advance in the waveform storage means, and the data selection means sequentially selects data of a predetermined segment waveform from the waveform storage means. In a speech synthesis method, a data editing means edits and synthesizes data of a segment waveform generated immediately before in the time domain, and a speech output means outputs synthesized speech according to the edited and synthesized segment waveform data. The complementary data generating means superimposes the data of the segment waveform that has been obtained and the data of the segment waveform newly selected by the data selection means at a predetermined ratio.

[0010] The invention according to claim 2 provides a waveform storage means in which data of a plurality of segment waveforms obtained by extracting the waveform of an audio signal in units of pitches is set in advance, and a predetermined segment waveform is stored from the waveform storage means. A data selection means for sequentially selecting data is provided, a data editing means is provided for editing and synthesizing the data of the segment waveform selected by the data selection means in the time domain, and data of the segment waveform edited and synthesized by the data editing means is provided. In a speech synthesizer equipped with a voice output means for outputting synthesized speech according to The means were set up.

[0011]

[Operation] By superimposing the data of the segment waveform generated immediately before and the data of the segment waveform newly selected by the data selection means at a predetermined ratio,
Changes in synthesized speech can be freely set according to the selection of segment waveform data.

0012

DESCRIPTION OF THE PREFERRED EMBODIMENTS A speech synthesis apparatus according to the present invention will be explained based on FIGS. 1 and 2. First, in this speech synthesis device 1, as illustrated in FIG. 1, data of a plurality of elemental waveforms are set in advance in a prosody generation section 2 that outputs waveform selection data via a selector 3. A waveform memory 4 is connected, and a counter 5 is connected to the waveform memory 4 and the selector 3, thereby forming a waveform selection section 6 which is data selection means. The waveform selection section 6 is connected to a smoothing section 10 which is complementary data generating means having a structure in which multipliers 7 and 8 and a shift register 9 are combined in a loop. , audio output means (not shown) consisting of an amplifier, a speaker, etc. are connected via a superimposition circuit 11, which is a data editing means for superimposing and editing and synthesizing waveform data at a predetermined pitch.

[0013] In the smoothing unit 10, (α-1) and α are set as the multiplier values of the multipliers 7 and 8 from the complementary coefficient α, and the number of stages of the shift register 9 is set as the value of the multipliers 7 and 8. It is the same as the number of fixed-length samples of data. Further, this shift register 9 can also be replaced with a ring buffer or the like.

With such a configuration, in this speech synthesis device 1, when the waveform selection data corresponding to the synthesized speech to be generated is sequentially output from the prosody generation section 2, the waveform memory of the waveform selection section 6 is stored in accordance with this waveform selection data. Data of predetermined elemental waveforms are sequentially selected from 4. Then, the smoothing unit 10 to which the data of these segment waveforms is transferred assumes that the data of the segment waveform is X and the data of the segment waveform to be complemented is Y.

Complementing elemental waveform data is generated according to an arithmetic expression such as [Equation 1]. Note that α in the above equation is a complementary coefficient of 0 or more and less than 1, and the smoothness of the synthesized speech changes according to its magnitude. In other words, this smoothing unit 10 uses the data of the elemental waveform as

When [Equation 2] is newly input, this and the data of the segment waveform edited and synthesized previously are

[Equation 3] is superimposed at a ratio of (1-α)/α to create new segment waveform data.

[Formula 4] is generated. Note that this smoothing section 10 has an initial value of

[Math 5] can be set in advance.

Therefore, in this speech synthesis device 1, the segment waveform data sequentially generated by the smoothing unit 10 as described above is superimposed at a predetermined pitch in the superimposing circuit 11 and edited and synthesized, and this edited and synthesized Synthesized speech is output from the speech output means according to the segment waveform data.

[0016] In this manner, the speech synthesis device 1 converts the waveform selection data in which the segment waveform data ■ to ■ whose phonemes differ in stages, as illustrated in FIG. 2(a), into a waveform. Even if the input is made to the selection unit 6, as shown by the solid line in the figure, the smoothing unit 10 outputs elemental waveform data that has an intermediate value between the elemental waveform data ■ and ■ and has good smoothness. It will be output. In addition, as illustrated in Figure 2(b), by forming waveform selection data in which data ■ and ■ of two segment waveforms with different phonemes are arranged so that the density changes sequentially, the solid line in the figure As shown, it is possible to generate segment waveform data with good smoothness in which phonemes change sequentially. Furthermore, as illustrated in FIG. 2(c), even when synthesizing vowel stationary parts, it is possible to give fluctuations to the synthesized speech by arranging the segment waveform data ■ to ■ of a predetermined phoneme at a predetermined timing. It is possible.

In other words, in this speech synthesis device 1, changes in the synthesized speech can be freely set according to the configuration contents of the waveform selection data, and its complementation characteristics are not limited to linear complementation in which the complementation interval is fixedly set. Therefore, various complementary characteristics can be expressed approximately.

[0018] As a method of generating the waveform selection data as described above, CV (Consonant Vowel) is used.
For within the section, waveform selection data stored in advance in CV units is directly used, and for the connection between CVs, waveforms are selected based on change patterns set in advance according to the type of phoneme. It is possible to generate data.

Furthermore, in this speech synthesis device 1, the structure of the smoothing section 10 can be simplified by setting the complementary coefficient α fixedly. For example, if α=0.5, the multiplication operation only requires a 1 (bit) shift, and the multiplier 7 that multiplies the data of the elemental waveform by (1-α) can be omitted. In this case, the data of the segment waveform output from the smoothing section 10 will be multiplied by 1/(1-α). Note that it is also possible to vary the complementary coefficient α as appropriate according to the phoneme of the synthesized speech.

[0020]

Effects of the Invention As described above, in the present invention, data of a plurality of segment waveforms obtained by extracting the waveform of an audio signal in pitch units is set in advance in the waveform storage means, and the data selection means selects a predetermined data from the waveform storage means. The data of the segment waveforms are sequentially selected, the data editing means edits and synthesizes the selected segment waveform data in the time domain, and the audio output means generates synthesized speech according to the edited and synthesized segment waveform data. In the speech synthesis method and device, the complementary data generation means superimposes the data of the segment waveform generated immediately before and the data of the segment waveform newly selected by the data selection means at a predetermined ratio. As a result, it is possible to freely set changes in the synthesized speech according to the selected content of the segment waveform data, so that it is possible to complement various characteristics with a high degree of freedom.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the present invention.

FIG. 2 is a characteristic diagram.

FIG. 3 is a characteristic diagram showing a conventional example.

[Explanation of symbols]

Claims (2)

[Claims] Claim 1: Data of a plurality of segment waveforms obtained by extracting the waveform of an audio signal in pitch units is set in advance in a waveform storage means, and a data selection means sequentially selects data of a predetermined segment waveform from the waveform storage means. The data editing means edits and synthesizes the data of the selected segment waveform in the time domain, and the speech output means outputs synthesized speech according to the edited and synthesized segment waveform data. A speech synthesis method characterized in that the complementary data generating means superimposes the data of the segment waveform generated immediately before and the data of the segment waveform newly selected by the data selection means at a predetermined ratio. 2. A waveform storage means in which data of a plurality of segment waveforms obtained by extracting the waveform of an audio signal in pitch units is provided, and data of a predetermined segment waveform is sequentially selected from the waveform storage means. A selection means is provided, a data editing means is provided for editing and synthesizing the segmental waveform data selected by the data selecting means in the time domain, and a synthesized speech is output in accordance with the segmental waveform data edited and synthesized by the data editing means. In a speech synthesis device provided with a voice output means, a complementary data generation means is provided for superimposing the data of the segment waveform generated immediately before and the data of the segment waveform newly selected by the data selection means at a predetermined ratio. A speech synthesis device featuring: