JPS5936273B2 - Fragment editing type speech synthesizer - Google Patents

Abstract

PURPOSE:To make it possible to edit and compose element waveforms with a small number of multiplier circuits by controlling the amplitude of the elongation wavelength, whose element waveform in the preceding pitch period overlaps with that in the following period, by mean-amplitude data in the following pitch period.

Description

[Detailed description of the invention] The present invention relates to a segment editing type speech synthesis device.

In advance, various speech segment waveforms with time lengths ranging from one pitch section to several pitch sections are prepared by extracting them from natural speech waveforms, and the segment number, pitch period data, and segment number given as synthetic data are prepared. A segment editing type speech synthesis method is known in which a synthesized speech waveform is generated by editing and synthesizing the speech segment waveform according to amplitude data that gives an average amplitude value of a segment waveform. In addition, as a method for editing and synthesizing the speech segment waveforms, after performing amplitude processing on each segment waveform according to the amplitude data given as synthesis data, the segment waveform is shifted by a pitch period and continues. A method is known in which a composite waveform is generated by adding elemental waveforms. As shown in Fig. 1, the above synthesis method uses the sample value of a segment waveform specified in a certain pitch section as fl(nT) (101 in Fig. 1), and uses the sample value of the segment waveform specified in a certain pitch section as fl(nT) (101 in FIG. Sample values of the waveform are f2 (nT) (102 in Fig. 1) from those closest to the pitch section.
, f3(nT) (103 in Figure 1), ...
・(Here, n■1, 2, ......, NN represents the number of sample values of the elemental waveform, T represents the sampling period), and the pitch period of each pitch section is expressed as P1T, P
2T, P3T... (Pi is a positive integer), and the average amplitude values are respectively expressed as Al, A2, As,..., the composite waveform n (iT) (1041 in Figure 1 is Next (given in 1 wipe.V(iT)=A1f1(iT)+A2f
2((i+p2)T)+・・・・−・・・・・・・・・(1
) The above synthesis method provides better continuity of the synthesized waveform and better sound quality than the method of cutting off the segment waveforms for each pitch period without adding the segment waveforms used in the preceding pitch section. As is clear from the above, it has the disadvantage of a large number of multiplications and a large amount of calculation when generating a synthesized waveform, especially when performing multi-channel synthesis processing using the same synthesis circuit by time-division processing. It will be disadvantageous.

An object of the present invention is to edit and synthesize speech segment waveforms in the segment editing type speech synthesis with almost no deterioration in sound quality and with a relatively small amount of calculations, thereby generating multi-channel speech segment waveforms in real time. It is an object of the present invention to provide a segment editing type speech synthesis device suitable for performing editing synthesis. The segment editing type speech synthesis device of the present invention combines a segment waveform selected in a temporally preceding pitch section with a waveform of a portion where a segment waveform selected in a temporally preceding pitch segment overlaps with a temporally subsequent pitch segment, and a segment waveform selected in the temporally preceding pitch segment. an adding circuit that calculates an added value by adding the waveforms, and a value obtained by multiplying the added value output from the adding circuit by the average amplitude value near the subsequent pitch section given in the subsequent pitch section. It consists of a multiplication circuit that outputs a synthesized speech waveform.

A feature of the present invention is that the amplitude of an extended waveform in which a segment waveform used in a temporally preceding pitch section overlaps with a subsequent pitch section is controlled by average amplitude data given in the subsequent pitch section. .

Therefore, when the present invention is used, discontinuities at the boundaries between adjacent pitch sections are relatively small, and elemental waveforms can be edited and synthesized with a relatively small number of multiplications. In the present invention, by utilizing the property that a small discontinuity in the waveform at the boundary point of the pitch interval has almost no effect on the sound quality of the synthesized sound, the temporally preceding After adding the segment waveforms F2 (NT), F3 (NT), ... selected in the pitch interval and the segment waveform fl (NT) selected in the pitch interval, the waveforms given in the pitch interval are added. The composite waveform (y'(□T)) in the pitch interval is calculated by multiplying the average amplitude data A1 obtained by A discontinuity that did not occur in the waveform y(IT) occurs at the boundary between adjacent pitch sections. However,
Since the value of the average amplitude data given in adjacent pitch sections changes small in most cases and does not become significant, the above-mentioned discontinuity is small and hardly affects the sound quality of the synthesized sound. In the editing and synthesis method according to the present invention expressed by equation (2), the number of multiplications per each time slot is one, and (1)
It is clear that the number of multiplications is reduced compared to the conventional method expressed by Eq. Therefore, the method of the present invention requires a relatively short amount of time to calculate a composite waveform for one time slot, and is suitable for a so-called multi-channel voice response method that simultaneously generates different composite waveforms for a plurality of channels.
Next, an embodiment of the present invention shown in FIG. 2 will be described in detail.

First, the audio output command data is input to the audio output command input terminal 201.
The signal is input to the control circuit 202 via.

The control circuit 202 outputs the synthetic data storage circuit control data to the synthetic data storage circuit 206 via the synthetic data storage circuit control zeta transmission line 203 according to the audio output command data, and outputs the segment waveform output circuit control zeta as a segment waveform. It is output to the elemental piece waveform output circuit 208 via the circuit control data transmission line 204. According to the synthetic data storage circuit control data, the composite data storage circuit 206 stores the segment number data, pitch data, and amplitude data specifying the segment waveform stored in advance in the segment waveform address register 209, respectively, and the pitch period data. Memory circuit 21
1 and amplitude data are output to the storage circuit 212. Note that the segment number data mentioned above is stored in the segment address register 20.
The old segment number data, which was temporarily stored in the segment address register 209 before being output to the segment address register 210, is output to the segment address register 210. Fragment address register 209
, 210 is output to the segment waveform output circuit 208. Piece waveform output circuit 2
08 indicates two numbers corresponding to the segment number data outputted from the segment address registers 209 and 210 from among the segment waveforms previously stored in the segment waveform storage circuit 20T according to the segment waveform output circuit control data. The segment waveforms are accessed via the switch 219, and the segment waveform data is temporarily stored in the memory circuit 213 or 214 via the switch 219. The adder circuit 215 is the pitch period data storage circuit 2.
According to the pitch period data outputted from the unit 11, the two unit waveform data outputted from the unit waveform data storage circuits 213 and 214, respectively, are shifted by the pitch period specified by the pitch period data and added. It is output to the multiplication circuit 216. The multiplication circuit 216 multiplies the amplitude data output from the amplitude data storage circuit 212 by the waveform data output from the addition circuit 215, and outputs the digital value to the DA conversion circuit 21T. The 0A conversion circuit 21T performs digital-to-analog conversion on the digital value output from the multiplication circuit 216, and outputs a composite waveform via the composite waveform output terminal 218.

By repeating the above operations for each pitch period, a composite waveform is obtained from the composite waveform output terminal 218.

In the above embodiment, the addition value is calculated by the addition circuit 215 on the assumption that only the segment waveform specified in the immediately preceding pitch interval and the two segment waveforms specified in the subsequent pitch interval overlap. It is clear that it is also possible to realize a device that also uses a segment waveform specified in a temporally preceding pitch interval.

[Brief explanation of drawings]

FIG. 1 is a diagram for explaining the editing and synthesis method of segment waveforms, and 101 is the segment waveform f selected in the pitch section.
1 (NT), 102 represents the segment waveform F2 (NT) selected one pitch interval ago, and 103 represents 2
Piece waveform F3 (NT) selected before pitch interval
, and 104 represents the composite waveform V(IT). FIG. 2 is a block diagram showing an embodiment of the present invention.
202 is an audio output command input terminal, 202 is a control circuit, 203 is a synthetic data storage circuit control data transmission line, 204 is a segmental waveform output circuit control data transmission line, 206 is a synthetic data storage circuit, 20T is a segmental waveform storage circuit, 208 is a segment waveform output circuit, 209 and 210 are segment waveform address registers, 211 is a pitch period data memory circuit, 212 is an amplitude data memory circuit, 213 and 214 are segment waveform data memory circuits, and 215 is a segment waveform data memory circuit. 216 is a multiplication circuit, 211 is a DA conversion circuit, 218 is a composite waveform output terminal, and 219 is a switch.

Claims (1)

[Claims] 1. In a segment-editing speech synthesizer that edits and synthesizes speech segment waveforms having a time length equivalent to a plurality of pitch segments prepared in advance, a segment waveform selected in a temporally preceding pitch segment is an adding circuit that outputs an added value obtained by adding a waveform of a portion that overlaps with a subsequent pitch section and an elemental waveform selected in the subsequent pitch section; an added value output from the adding circuit and the subsequent pitch; 1. A segment editing type speech synthesis device comprising: a multiplication circuit that outputs a value multiplied by an amplitude value given in an interval as a synthesized speech waveform.