JPH0315760B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a waveform editing type speech synthesis device that edits and synthesizes unit speech waveforms such as monosyllables.

Conventionally, unit speech waveforms such as monosyllables (for example, CV and VC waveforms extracted from natural speech waveforms, excluding the beginning and end of the word; here, C represents a consonant and V represents a vowel) are edited and synthesized to create an arbitrary sound. In a speech synthesis method that performs speech synthesis of words, a speech waveform of about 1 pitch at the rear connection part of the temporally preceding unit speech waveform and about 1 pitch worth of speech waveform at the front connection part of the following unit speech waveform are used. A method is known in which a speech waveform is synthesized by interpolating the waveform between the two unit speech waveforms by weighting and adding together the speech waveforms.

However, in the conventional method, although the two unit audio waveforms are smoothly connected through interpolation processing, calculations such as weighting are required during interpolation processing, so different responses are given to multiple channels at the same time. The drawback was that it was relatively difficult to perform a so-called multi-channel response that outputs audio.

An object of the present invention is to provide a waveform editing type speech synthesis device that generates a speech waveform of an arbitrary word by editing and synthesizing unit speech waveforms such as monosyllables, in which interpolation processing between the unit speech waveforms is relatively simple. An object of the present invention is to provide a waveform editing type speech synthesis device that can easily respond to multiple channels simultaneously.

The present invention provides means for storing unit speech waveforms such as monosyllables, means for storing interpolation speech waveforms for interpolating waveforms with different spectra even between the same phonemes, and the preceding unit speech waveforms. and a means for interpolating and editing and synthesizing the subsequent unit audio waveforms using the interpolation audio waveform.

Next, regarding the interpolation method and interpolation waveform in the present invention, we will discuss the interpolation method and interpolation waveform using dyad as a unit audio waveform.
An example of a case where a waveform (CV or VC waveform with several pitches) is used will be explained.

Even if the phoneme is the same, the pitch or spectrum at the rear connection part of the temporally preceding diad waveform may be different from the pitch or spectrum at the front connection part of the following diad waveform. An example of a case where the spectra differ even if the phoneme is the same is that, even if the vowel is the same, there is a difference between a vowel that is nasalized as a result of articulatory combination, a vowel that is not nasalized, and a vowel that is not nasalized.

In order to connect two diad waveforms having different pitches or spectra, an interpolation waveform whose pitch or spectrum gradually changes is prepared in advance, and the two diad waveforms are interpolated.

The interpolation waveform has several pitches including, for example, an intermediate pitch of the diad waveform, and it is possible to obtain a speech segment waveform of about one pitch from a natural speech waveform with a different spectrum such as a nasalized vowel or a non-nasalized vowel. It is obtained by extracting the speech segment waveforms and then weighting and adding the speech segment waveforms to generate a speech segment waveform having an intermediate spectrum.

The interpolation of the two diad waveforms is performed by appropriately selecting and connecting some speech unit waveforms from the speech unit waveforms. Selection of the speech segment waveform is performed by storing interpolation information in advance in accordance with the pitch combination and spectrum combination of the two diad waveforms.

As the waveform for interpolation, prepare in advance a speech waveform of about 1 pitch or more so that changes in pitch and spectrum between the two unit speech waveforms are smooth, rather than the speech unit waveform of about 1 pitch. It is also possible to use it for interpolation of the two unit voices.

As described above, according to the present invention, the above two
It is clear that by interpolating between two unit speech waveforms, a synthesized sound of relatively high quality can be obtained, and since the interpolation process is simple, it is easy to respond to multiple channels simultaneously.

Next, embodiments of the present invention will be described using the drawings.

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

The sentence sequence temporary storage circuit 102 stores character strings input from the character string input terminal 101 representing the speech to be synthesized, such as unit speech waveform names and pitch levels. When a character string is input, information requesting the start of synthesis is sent to the control circuit 121 via the start/end information transmission line 103. When control information instructing the transmission of a character string is sent from the control circuit 121 via the control information transmission path A104, the unit audio waveform name is transmitted via the unit audio waveform name transmission path A105 or B107. The pitch level is transmitted to the waveform address table search circuit A109 or B1 via the pitch level transmission line A106 or B108.
Send it to 10th one after another. After transmitting the last unit audio waveform name and pitch level, information indicating the end is transmitted to the control circuit 121.

The waveform address table search circuit A109 is the control circuit 1
21 via the control information transmission path B112, input the unit audio waveform name and pitch level sent from the character string temporary storage circuit 102, and select the unit audio from the internal table. Find the start address and number of samples where the unit audio waveform corresponding to the waveform name and pitch level is stored. Once the start address and number of samples are obtained, information indicating this is sent to the response information transmission path A111.
to the control circuit 121 via the control circuit 121.
When control information instructing data transmission is received from , the start address and the number of samples are sent to the address generation circuit A113 via the start address transmission path A114 and the sample number transmission path A115, respectively.

The waveform address table search circuit B110 is the control circuit 1
21 via the control information transmission path C117, input the unit audio waveform name and pitch level sent from the character string temporary storage circuit 102, and if it is immediately after initialization, input the unit audio waveform name and pitch level. The preceding unit speech waveform is stored internally as the waveform name and pitch level of the preceding unit speech wave, and if it is the second or subsequent data after initialization, it is stored as the waveform name and pitch level of the subsequent unit speech waveform. The starting address and number of samples of the corresponding interpolated waveform are searched from the internal table according to the waveform name and pitch level of , and the waveform name and pitch level of the following unit audio waveform. Furthermore, it stores the waveform name and pitch level of the following unit audio waveform as a new waveform name and pitch level of the preceding unit audio waveform, and responds to the control circuit 121 with information indicating that the start address and number of samples have been obtained. When control information is sent via the information transmission path B116 and instructs to send data from the control circuit 121, the start address and the number of samples are sent to the address generation circuit B118 via the start address transmission path B119 and the number of samples transmission path B120, respectively. send to

Address generation circuit A113 or B118
are the control information transmission paths D from the control circuit 121, respectively.
122 or the same E124, the start address and number of samples of the unit audio waveform or the start address and number of samples of the interpolation waveform are input, and the address data is sent from the control circuit 121. When the control information instructing the transmission of the unit audio waveform or interpolated waveform is sent to the unit audio waveform data storage circuit 128 or the interpolated waveform data storage circuit 129 via the address transmission path A126 or B127, respectively, from the start address of the unit audio waveform or interpolated waveform. Send addresses for the number of samples one after another. When the sending of addresses for the number of samples is completed, information indicating this is sent to the control circuit 121 via the response information transmission path C123 or D125.

The unit audio waveform data storage circuit 128 or the interpolated waveform data storage circuit 129 edits the waveform data of the address sent from the address generation circuit A113 or address generation circuit B118, respectively, via the waveform data transmission path A131 or the waveform data transmission path B132, respectively. Synthesis circuit 133
send to

The editing/synthesizing circuit 133 inputs the waveform data sent from the unit audio waveform data storage circuit 128 or the interpolated waveform data storage circuit 129 according to the control information sent from the control circuit 121 via the control information transmission line F130, and outputs the synthesized audio output terminal. 134.

A control circuit 121 controls each of the circuits. First, when request information to start synthesis is sent from the character string temporary storage circuit 102, initialization is instructed to each circuit,
Next, the character string temporary storage circuit 102 and the waveform address table search circuits A109 and B110 are instructed to transfer the unit audio waveform name and pitch level. Waveform address table search circuit A109 or B11
When the table search end response information is sent from 0, it is confirmed that the address generation circuit A113 or the address generation circuit B118 is not sending out address data, and the waveform address table search circuit A109 and the address generation circuit A113 or the waveform address are sent. The table search circuit B110 and address generation circuit B118 are instructed to transfer the start address and the number of samples, and then the character string temporary storage circuit 102 and the waveform address table search circuit A109 or B110 are instructed to transfer the next data. . Next, immediately after initialization, since the found start address and the number of samples are related to the initial waveform data of the audio to be synthesized, the address generation circuit A 113 is immediately instructed to send the address, and at the same time, the editing synthesis circuit 1
33, the unit audio waveform data storage circuit 128
From then on, when receiving information from either address generation circuit A113 or address generation circuit B118 indicating that address transmission has been completed, the editing/synthesis circuit simultaneously instructs the other to input the address. 133 to switch between the unit audio waveform data storage circuit 128 or the interpolated waveform data storage circuit 129 to input and output waveform data.

In this embodiment, each circuit operates in parallel, so that when a character string is input to the character string input terminal 101, a synthesized sound begins to be output after a slight delay.

FIG. 2 shows a second embodiment of the present invention, and is a flowchart of a program implemented by program control using a microprocessor or the like. In this example, the synthesized sound is output after editing and synthesis are completed.

First, in the preparation procedure 201, the output buffer (referred to as OUTB) is emptied, the counter (referred to as COUNT) representing the number of samples of waveform data in the output buffer is set to zero, and the unit audio waveform in the input buffer (referred to as INB) is Set the pointer (referred to as POINT) indicating the name and pitch level to 1.

Next, in a character string input procedure 202, character strings such as the unit speech waveform name and pitch level representing the speech to be synthesized are read into the input buffer (INB).

Next, in the unit audio waveform address table search procedure 203, the unit audio waveform name and pitch level indicated by the pointer (POINT) of the input buffer (INB) are set as the unit audio waveform address table (TADR).
Find the corresponding start address (assumed to be SADR1) and number of samples (assumed to be NSPL1).

Next, in the unit audio waveform data transfer procedure 204, the waveform data for the number of samples (referred to as THAKEI) from the start address of the unit audio waveform obtained in the unit audio waveform address table search procedure 203 is transferred to the output buffer (OUTB) and its counter Transfer from the next address indicated by (COUNT).

Next, in a counter update procedure 205, the counter (COUNT) is incremented by the number of samples transferred in the unit audio waveform data transfer procedure 204.

Next, in the end determination procedure 206, it is determined whether character strings representing unit audio waveform names and pitch levels remain in the input buffer (INB), and if so, the next interpolation waveform address table search procedure 20
7, if there is no remaining synthesized voice output procedure 210
transfer control to

In the interpolation waveform address table search procedure 207, the preceding unit audio waveform name and pitch level indicated by the pointer (POINT) of the input buffer (INB) and the next succeeding unit audio waveform name and pitch level indicated by the pointer (POINT) are used as the interpolation waveform address. Search in the table (assumed to be HADR) and obtain the corresponding start address (assumed to be SADR2) and number of samples (assumed to be NSPL2).

Next, in the interpolated waveform data transfer procedure 208,
Waveform data for the number of samples (called HHAKE1) from the top address of the interpolated waveform obtained in the interpolated waveform address table search procedure 207 is transferred to the output buffer (OUTB) from the address following the address indicated by the counter (COUNT).

Next, in the counter/pointer update procedure 208, the counter (COUNT) of the output buffer is advanced by the number of samples transferred, the pointer (POINT) in the input buffer is advanced to the beginning of the next unit audio waveform name and pitch level, and the unit audio waveform is Control is transferred to address table search procedure 203.

The synthesized speech output procedure 210 outputs the data stored in the output buffer (OUTB) and indicated by the output buffer counter (COUNT) as synthesized speech waveform data.

In this embodiment, the above-mentioned series of procedures are applied to the memory that stores unit audio waveforms as a means for storing unit audio waveforms, and the memory that stores interpolated waveforms as means that stores audio waveforms for interpolation. The processing device and output buffer correspond to means for editing and combining.

Note that the speech synthesis device of the present invention generates speech of any word by a process similar to that of conventional speech response devices that output speech waveforms of fixed words such as certain specific sentences using a random access method. It is possible to synthesize waveforms, and it is particularly effective in application to voice response devices that output sounds of fixed words and arbitrary words.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing one embodiment of the present invention;
FIG. 2 is a diagram showing the flow of a program when a microprocessor or the like is used as a second embodiment of the present invention. In the figure, 101 is a character string input terminal, 102
109 is a character string temporary storage circuit, 109 is a waveform address table search circuit A, 110 is a waveform address table search circuit B, 113 is an address generation circuit A, 118 is an address generation circuit B, 121 is a control circuit, and 128 is unit audio waveform data. 129 is an interpolated waveform data storage circuit; 133 is an editing/synthesizing circuit; 134 is a synthesized audio output terminal; 201 is a preparation procedure; 202
203 is a unit audio waveform address table search procedure, 204 is a unit audio waveform data transfer procedure, 205 is a counter update procedure, 206 is an end determination procedure, 207 is an interpolated waveform address table search procedure, and 208 is an interpolated waveform A data transfer procedure, 209 a counter and pointer update procedure, and 210 a synthesized voice output procedure.

Claims (1)

[Claims] 1 In a waveform editing type speech synthesis device that edits and synthesizes unit speech waveforms such as monosyllables, there is a means for storing unit speech waveforms such as monosyllables, and a means for interpolating waveforms with different spectra even between the same phoneme. a means for storing an interpolation speech waveform for use in the interpolation speech waveform; and a means for editing and synthesizing a speech waveform of an arbitrary word by interpolating between the preceding unit speech waveform and the following unit speech waveform using the interpolation speech waveform. A waveform editing type speech synthesis device comprising: