JPS6021098A - Synthesization of voice - 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 (Technical Field) The present invention stores data representing a sample time series of waveforms associated with one-pitch speech segment in natural voiced speech as a unit of segment, and synthesizes the data. It relates to a method of synthesizing speech by editing and reproducing segment units according to a series of segment unit selection information created in response to input information regarding voice speech, and in particular, to The present invention relates to a speech synthesis method in which a sequence and a sequence of the segment unit selection information are associated with each other while maintaining good phonological properties and good prosodic properties.

(Prior art) A large amount of Wr for general purposes: 'There is a high demand for devices that require the output of arbitrary sounds such as bamboo, or names of people, companies, and places. If you observe the waveform of speech, you will see that similar waveforms are repeated in sections of voiced sounds such as vowels. This period is called a pitch period, and the waveform within this one period is called a pitch unit speech element (hereinafter referred to as a phoneme). This change in the content of phonemes represents phonology,
The temporal changes and turns of this cycle give an accent and serve as an element of prosody. In voiced sound sections such as 11 tones, waveforms of almost the same shape as el are repeated, and waveforms of similar shapes appear in the same type of voice. Therefore, it is possible to synthesize arbitrary continuous speech by storing characteristic phoneme waveforms necessary to create speech among the waveforms that appear in speech in a storage device, and editing these basic phoneme waveforms. Conceivable. Additionally, any sentence in Japanese can basically be expressed using over 100 different types of monosyllables. In synthesizing the waveform domain, in order to appropriately control the prosody using speech units stored in the storage device as raw materials, continuous speech is created while changing the pitch, amplitude, and duration of the voice according to the instructions in the dictionary of control information. We need to build everything up. As the basic unit of speech in the arbitrary speech synthesis method, it is better to use speech segments in terms of memory capacity and rhythmicity, and in terms of phonology, it is thought that units larger than the monosyllabic level are better. . For this reason, a method has been proposed in which monosyllables are used as mII control units in terms of phonetics, and speech segments are used as units in prosodic terms. That is, this is a method in which the speech segments recorded in the storage device are sequentially arranged within the unit with single syllables as syllables. In this method, phonetic properties are maintained by sequentially extracting speech segments in chronological order for each single syllable, and speech segments can be used as a unit for prosody control.

(Problems with the Prior Art) By the way, in such an arbitrary word synthesis method using audio data in the waveform region, one problem arises due to pitch control.

Assume that a certain monosyllabic waveform extracted from natural speech and stored in a storage device is composed of a time series of n phonemes as shown in FIG. 1 (,). Also, for simplicity, these n
If all phonemes have a pitch period of P, the time length PHLa of this 'l'111 M) is given by PXn. In order to give naturalness to synthesized sounds, appropriate pitch changes must be given to each of these phonemes depending on the situation in which this single syllable is used, but the range of change in pitch of speech is quite large. In the case of female voices, there is a difference of more than one octave between the shortest pitch and the longest pitch. Therefore, for example, if this single syllable is synthesized using the phoneme data shown in Figure 1(a) at the pitch of the bar in the standard case, the synthesized waveform will be as shown in Figure 1(b), and its continuation. Time PHLb is also A.

That is, by performing pitch control, the duration also changes in proportion to it. Originally, a change in pitch represents a change in the vocal fold vibration period, and a change in duration is thought to represent a change in the shape of the vocal tract, and although they are somewhat related to each other, they can be regarded as largely separate/common things. Therefore, even if the pitch is changed, the duration should be kept constant as long as there are no other factors that change the duration, and synthesized sounds where the duration of the phoneme is not appropriate will cause a disturbance in the ten I. , giving an impression of irregularity in timing and rhythm.

Furthermore, the influence of pitch control is not limited to changes in duration, but also has a large effect on phonological properties. That is, the first
In FIG. 1(b), the change state of the phoneme that changes during the time period PHLaO is summarized in half the time period PHLb in FIG. 1(b).

Conventionally, in this type of synthesis apparatus, the influence of pitch control on duration time has generally not been taken into consideration. On the other hand, there are methods in which the duration of the entire syllable is kept constant by repeating or deleting the elemental fragment at the end of the syllable, but these (Objective of the Invention) The present invention does not solve the problem in terms of the speed of change of phonemes within a natural voiced sound. It belongs to a type of data that uses attenuated waveforms by converting the above, and stores data that can reproduce the Zannor time series as a unit of fragments.

Furthermore, the present invention allows the pitch cycle series and the segment unit selection information series, which are respectively created based on input information regarding speech to be synthesized, to be generated on a time axis measured by a sample clock of real time or calculation time. It belongs to the format where each is set independently.

An object of the present invention is to set the leaf unit selection information and the pitch period on the time axis, and set the elemental unit selection information every time a number of sample clocks corresponding to the number of samples of the corresponding elemental piece are counted. and the pitch period is updated every time a corresponding number of sample clocks are counted, thereby maintaining the phonology and prosody of the synthesized sound independently of each other, and The purpose of this method is to avoid discontinuities in waveform joints that occur due to this by providing a plurality of playback channels.

(Summary of the invention) An outline of the present invention will be explained using FIGS. 2 to 4.

Figures 2 and 3 show a case where one monosyllable is selected based on input information regarding the speech to be synthesized, and the phoneme sequence ELi in that monosyllable is created based on the input information. It shows the correspondence relationship with the pitch period series PTj.

Note that each phoneme EL in Figures 2 and 3 corresponds to the standard natural voiced sounds ``ebi'' and ``chi'', and even when the same phoneme is consecutive, it is distinguished by a different number i. It shows. Each phoneme ELi is 1
The preceding phonemes ELi, -
Every time a number of sample clocks corresponding to the number of samples of [ are counted, the next phoneme ELi is selected and set. Therefore,
Monosyllabic continuation on the time axis 1 Samurai 1) IILl,
The arrangement of the phonemes ELi reproduces that of a standard monosyllable stored in advance. Each pitch period PTj is updated and set every time a number of sample clocks corresponding to the preceding pitch period PTj-1 are counted. Then, the composite output is generated every time the pitch period PTj is updated. , tl, ·"ntm-1, the phoneme PHi set at the time tj is selected, and the phonemes are combined to create the phoneme.

Figure 2 shows a case where the pitch period of the speech to be synthesized is longer than the phoneme length of the phoneme EL□ (therefore the pitch period of the standard phoneme) adopted as the standard one, for example, the phoneme EL3. By not reproducing the phonology, especially the monosyllabic duration PHL.

and prosodic characteristics, especially the sequence of pitch cycles, maintains its uniqueness.

Figure 3 shows a case where the pitch period of the synthesized output is shorter than the phoneme length of the phoneme PHi.
1. Maintain the uniqueness of phonology and prosody by reproducing EL4 repeatedly.

1 Figure 4 shows the method of superimposing each reproduced phoneme. A predetermined phoneme ELi in a specified channel
By playing back samples of , and then superimposing the outputs of all playback channels, discontinuities at joints between adjacent phonemes or between adjacent monosyllables in the synthesized output are avoided.

(Example) Next, an example will be described using FIGS. 5 to 9.

FIG. 5 is a block diagram showing the speech synthesis device, and FIG. 6 is a flowchart showing the functions and control procedures executed by the microprocessor 1. Input information regarding the speech to be synthesized is input from the transcription writer 2 to the microprocessor 1 in the form of a series of character codes indicating monosyllables for each word with appropriate pauses and the like. The prosody memory unit 3 stores prosodic control information such as accent type, intonation type, and duration regarding input words, as well as single sounds 11j.
These are searched by the microprocessor 1, and a series of monosyllabic information and a series of pitch periods PTj are created according to the input information. Further, the monosyllable information is created to include monosyllable selection information that specifies the start address of each monosyllable in the monosyllable memory 4 and a monosyllable length. The syllable memory 4 stores a series of long selection information in units of elements that constitute a single syllable, with each single syllable as one block. The segment unit selection information is stored as information specifying the start address of each segment unit in the segment memory 5. The segment memory 5 stores the segment units necessary to reproduce standard arbitrary phonemes, and the segment units are DP.
It is created using an appropriate compression technology such as CM, and all the fragment units are stored in an amount of data that can reproduce a sample time series of a predetermined length of 128 samples. The waveform regenerator 6 has a configuration compatible with the compression technology adopted, and reproduces the corresponding sample time series based on the data in units of fragments, and the pitch period in the synthesized output is short. There are also 4 playback channels RG1 to RG4 so that 128 samples can be played back.
has been established. Waveform playback is performed by setting the pitch period sequence PTY and the sequence of segment unit selection information in correspondence on the real-time time axis set by interrupt activation of the sample clock, and changing the time axis every time the pitch period is updated. By selecting the element unit ELj of the element unit selection information corresponding to the above and selecting one reproduction channel sequentially and cyclically every time the pitch period is updated, the element unit to be reproduced and the reproduction channel are made to correspond. , is executed by reproducing one sample at a time in all reproduction channels and superimposing them.

7 to 9 show the waveform generation flow executed by the microprocessor 1. FIG.

FIG. 7 shows the flow of waveform reproduction. Address variables ADj to AD4 for four channels, which are initialized and updated as described later, are prepared, and step 5P
In II, the address of the elemental piece memory 5 is specified with the first address number ADi, one piece of data of one elemental piece is read out, the data is sent to the first reproduction channel RG1 of the waveform regenerator 6, and one sample is read out. to play. In steps 5P12 to 5P14, the addresses of the segment memory 5 are specified using the second to fourth address variables AD2 to AD4, respectively, and the corresponding second to fourth reproduction channels RG2 are
~Play one sample each on RG4. In step 3P15, all playback channels RG1 to RG
By adding and superimposing the outputs of 4, that is, 4 samples, one sample of the composite output is reproduced and output. The synthesized output is reproduced by reproducing one sample each time the sample clock is activated.

FIG. 8 shows a flowchart for setting segment unit selection information on the time axis of the sample clock.

In FIG. 7, it is started by the interruption of sample cursor, and it is determined in step 5T21 whether the monosyllable currently being processed remains. Single note length variable PH7 = 0
Then, in step 5P22, the monosyllable number is incremented to extract the next 11'L syllable information from the previously created series of monosyllable information, and in step 5P23, the monosyllable length PHLa is set as the monosyllable length variable PH. . Subsequently, in steps 5P24 and 5P25, the address of the monosyllabic memory 4 is designated by the % head address, the first address of the segment unit ELi is read out, and it is worded to the address variable ADe of the segment memory 5. In this step SP25, the first segment unit selection information in one single syllable is set in correspondence with "zamburukuro" and "ri". Further, in step 5P26, the remaining length of the segment unit ELj currently being processed is determined, and if the number of segment long hairs ELt=O, the content of the address variable AD is incremented in step 5T27 to specify the next address. , Steps 5P24 to 5P25, the start address of the next elemental unit ELj+1 in one single syllable, that is, the next elemental unit selection information, is set on the time axis of the siple clock. Step 5P28.5P
29 is a reed in the process of counting the segment length on the time axis of the sample clock, and the number of samples for all segment units is the same 12
Since it is set to 8 pieces, the segment length variable E is set in step 5P28.
This is executed by loading 128 as LA and decrementing it in synchronization with the sample clock in step SP29. Also, in step 5P30, the single note length variable PH6 is decremented in synchronization with the sample clock. .

FIG. 9 shows a flowchart for setting the pitch period in real time and associating four segment units simultaneously read out from the segment memory 5' with predetermined reproduction channels RGj to RG4. Referring to FIG. 9, each time an interrupt is activated by the sample clock, the remaining length of the pitch period PTj currently being set is determined in step 5P40, and if the pitch period variable p'rt-〇 is set, the remaining length of the pitch period PTj is determined. ~5P
After the processing in step 43, in step 44, the pitch number J is incremented to extract the next pinch period PTj from the previously created pitch period series, and the result is Stellar fsP4.
5, load it into the pitch period variable PH6-\,
Further, in step SP46, each pitch period PTj is set on the time axis of the sample clock by decrementing the pitch period variable PH,6 in synchronization with the Sansol clock. In steps 5P-4-, 1 to 43, which are executed in synchronization with the update of the pitch period PTj, first, in step SP41, the 4-channel specified value CH is incremented to 1 in a cyclical manner.
Select one channel, and in step SP42 address variables ADj to AD4 for four channels in the segment memory 5 are selected.
Address variable AD corresponding to the specified value CH from among
Channel is selected, and in step 5P43, the selection information of the currently set segment unit EL□, that is, the address variable AD indicating the start address of the segment unit, is loaded into the address variable ADoh of the selected channel. By doing this, each elemental unit is assigned to each channel. In step 5P47, the address variable AD of each channel
By incrementing 1 to AD4, each address variable AD1 to AD4 addresses data for reproducing samples in different unit of fragments and in different orders in the same X1jL-c sequential manner as the sample clock. I can go.

In the above embodiment, since a real-time sample clock is used, only four address variables are prepared.
Although it is possible to play back one sample at a time, if you want to play back based on the sample rate of the calculation time, it is necessary to play back all the data for each segment corresponding to the time axis every time the pitch period is updated. Read and play the sample time series in each playback channel once, and then store the sample time series of the subsequent adjacent pitch family jυ for those with 1 μr adjacent pitch and chi period with the pitch period. A composite waveform can also be obtained by adding related samples. Also, even when the number of samples played from a segment unit, that is, the segment length, is not constant, the segment length is stored in the monosyllabic memory together with the segment unit selection information, and the segment length is used as the segment length variable instead of a constant. This can be handled by loading the length of the segment into .

(Effects of the Invention) As is clear from the above explanation, the present invention selects phoneme pieces in accordance with each pitch period given at the time of synthesis. Time and phonology are always maintained well, and natural synthesized sounds can be obtained in the waveform domain arbitrary word zero synthesizer.

[Brief explanation of the drawing]

FIG. 1 is an explanatory diagram showing the correspondence between phoneticity and prosody in waveform domain synthesis, and FIGS. 2 and 3 are explanatory diagrams showing the correspondence between phonemes and pitch periods in the present invention. Fig. 4 is an explanatory diagram showing the relationship of waveform superimposition in the present invention, Fig. 5 is a diagram showing an embodiment of the present invention, Fig. 6 is a diagram showing the overall flow, and Figs. FIG. 9 is a diagram showing a part of the detailed flow. 1...Microprocessor, 2...Typewriter,
3... Prosodic memory section, 4... Monosyllabic memory section, 5.
... Fragment memory unit, 6... Waveform regenerator. Procedural amendment (master) 58.11. 1 Director General of the Japan Patent Office, 1939 1. Indication of the case 1988 Special A'r i No. 127999 2. Name of the invention Speech synthesis method 3. Person making the amendment Relationship with the case Patent applicant 5. Subject of the amendment Details The "Detailed Description of the Invention" section is included in the book.

Claims (1)

[Scope of Claims] a) Data representing a sample time series of a waveform associated with a speech unit of one pitch unit in natural voiced speech is stored as a unit of unit, and b) A plurality of playback channels are prepared to play sample time series of the waveform, and C) a pitch period series created based on human information about the speech to be synthesized is measured using a real-time or calculation-time sample clock. Monitoring on the time axis and updating and setting the next pitch period in the pitch period series every time a number of sample clocks corresponding to one pitch period set on the time axis is counted; d) Synthesis. A segment created based on the input information regarding the voice to be used. (i) Selecting the first place (monitoring the rJ report series on the time axis and selecting one segment unit set on the time axis) Every time a number of sample clocks corresponding to the number of Sansols in the elemental piece unit corresponding to the information are counted, the elemental piece unit selection information in the elemental piece unit selection information series is updated and set, e) synchronized with the update of the pitch period. select the corresponding piece unit selection information on the time axis, and sequentially and cyclically select one of the playback channels in synchronization with the update of the pitch period, thereby selecting each piece unit selection information and each playback. f) giving each playback channel a corresponding fragment unit of the fragment unit selection information and playing the sample time series respectively; g) superimposing the sample time series of all playback channels; A speech synthesis method characterized by generating speech output based on the superimposed sample time series.