JPH0447840B2 - - Google Patents

Description

Detailed Description of the Invention (Technical Field) The present invention stores waveform data corresponding to one pitch speech element in natural voiced speech as a unit of speech element, and stores the waveform data corresponding to one pitch speech segment in natural voiced speech, and The method relates to a method of synthesizing speech by editing and reproducing the segment unit according to a series of created segment unit selection information, and in particular, a series of pitch periods created according to the input information and the segment unit. The present invention relates to a speech synthesis method in which a sequence of selection information is associated with a sequence while maintaining good phonology and prosody.

(Prior Art) There is an increasing demand for devices that require the output of large vocabularies or arbitrary speech such as people's names, company names, and place names for general purposes. When observing the waveform of speech, it can be seen that similar waveforms are repeated in voiced sections such as vowels. This period is called a pitch period, and the waveform within this one period is called a speech unit of one pitch. Changes in the content of this speech segment represent phonology, and the temporal pattern of this periodic change gives an accent and represents an element of prosody. Waveforms of almost the same shape are repeated in voiced sound sections such as vowels, and waveforms of similar shapes appear in the same type of speech. Therefore, it is possible to synthesize arbitrary continuous speech by storing characteristic speech segment waveforms necessary to create speech among the waveforms that appear in speech in a storage device, and reading and editing them. 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, the pitch, amplitude, and duration of the voice must be changed according to the instructions in the dictionary of control information. It is necessary to create a continuous voice while doing so.

As the basic unit of speech in arbitrary vocabulary speech synthesis methods, it is better to use speech segments in terms of memory capacity and prosody, and units that are larger than the monosyllabic level are considered superior in terms of phonology. .

For this reason, a method has been proposed in which the unit of control is a single syllable in terms of phonology, and the unit of control in terms of prosody is a speech segment. That is, there is a method in which phonetic segments stored in a storage device are sequentially arranged within a single syllable block. 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 vocabulary synthesis method using audio data in the waveform domain, 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 pitch segment waveforms as shown in FIG. 1a. Also, for simplicity, these n piece waveforms are all P
If the pitch period is as follows, the duration of this single syllable PHLa is given by P×n. In order to give naturalness to synthesized speech, appropriate pitch changes must be given to each of these elemental waveforms depending on the situation in which the single syllable is used, but the range of pitch change in speech is quite large. In the case of a female voice, there is a difference of one octave or more between the shortest pitch and the longest pitch. Therefore, for example, if this single syllable is synthesized using the phoneme data in Figure 1a at half the pitch of the standard case, the synthesized waveform will be the first one.
It becomes as shown in Figure b, and its duration PHLb is also halved.

That is, by performing pitch control, the duration time also changes in proportion to it. Originally, changes in pitch represent changes in the vocal fold vibration period, and changes in duration are considered to represent changes in the vocal tract shape, and although they are somewhat related to each other, they can be considered to be largely separate. Therefore, even if the pitch is changed, the duration should be kept constant unless there are other factors that change the duration, and synthesized sounds where the phoneme duration is not appropriate will result in tempo disturbances. Gives an impression of irregularity in timing and rhythm.

Furthermore, the influence of pitch control not only affects changes in duration, but also has a large effect on phonological properties. That is, in FIG. 1a, the state of change in phoneme that changed during the time period PHLa is summarized in half the time period PHLb in FIG. 1b.

Conventionally, in this type of synthesizer, the influence of pitch control on duration time has not generally been taken into consideration. On the other hand, by repeatedly using or deleting the segment at the end of a syllable,
There are methods that keep the duration of the entire syllable constant, but even this method does not solve the problem in terms of the speed of change of phoneme within a syllable.

(Objective of the Invention) The present invention has been made to solve the above-mentioned drawbacks of the prior art. The duration of the synthesized speech is determined by referring to the standard speech data stored in the storage device to determine what kind of phoneme should be expressed in the section, and by selecting the phoneme data to be used during synthesis. The purpose is to maintain appropriate phonology.

(Summary of the Invention) FIGS. 2 and 3 show a speech segment selection method according to the present invention. The upper diagrams in FIGS. 2 and 3 show standard speech data, and it is assumed that a speech unit such as a certain syllable or phoneme is composed of n speech segments S ₁ to _{S o} . These speech segment data are not only a time series of speech segment data, but also represent a time series of phonetic properties that change from moment to moment in the duration T _PH . That is, the speech segment S _i expresses the phonology of the interval from time t _i-1 to _{t i} within the duration T _PH of the entire speech unit, and this time length L _i is defined as the phonology of the speech segment S _i . The phoneme length sequence ₍ L _{1 ~
Lo} ) is stored in the storage device as standard audio data regarding audio units.

As shown in Figure 1a of the prior art, when a single syllable is extracted from natural speech and the segment waveform for each pitch period is used as a speech segment, each pitch waveform corresponds to a speech segment S _i . Furthermore, the pitch period of each one-pitch waveform corresponds to the phoneme length L _i .

When synthesizing speech units, at the start of each speech segment in a given pitch period, what kind of phonology should be expressed in that section is determined by comparing it with standard speech data stored in a storage device. judge and select an appropriate speech segment.

FIG. 2 shows a case where synthesis is performed with a pitch period longer than the standard phoneme length, and FIG. 3 shows a method of selecting speech units when synthesis is performed with a pitch period shorter than the standard phoneme length.

In FIG. 2, since the start time t _P0 of a speech unit corresponds to t ₀ , it is natural that the speech unit S ₁ is used for synthesis. The speech unit S ₁ is given a pitch period P ₁ longer than the phoneme length L ₁ by pitch control during synthesis, and
The start time of the next speech segment is t _p1 . here,
The standard speech data is referred to again in order to select the next speech segment. The time t _p1 is the interval t ₁ to _{t 2} in which the speech unit S ₂ is to be used in the standard speech data.
Therefore, speech segment _S2 is selected. Furthermore, at the start time t _p2 of the next speech segment, in the conventional method, speech segment S ₃ would be selected based on the order, but in the present invention, when referring to the standard speech data, the speech segment S 3 is selected at time t p2. _p2 is the interval t ₃ that should represent the speech segment S ₄
~t ₄ and select S ₄ next to S ₃ . Thereafter, in the same manner, the speech segment to be used is determined at each start time t _p1 of each speech segment.

On the other hand, in the example shown in Fig. 3 in which synthesis is performed with a pitch period shorter than the standard phoneme length, speech segment S ₁ is first selected and synthesized, and then the second segment start point t _{p1 is} selected.
is still in the interval t ₀ to _{t 1} in which the speech unit S ₁ should be selected because the pitch period P ₁ is shorter than the standard phoneme length L ₁ . In such a case, the speech segment S ₁ is used again.

As mentioned above, at the start of each speech unit in the pitch period given during speech unit synthesis,
By comparing the phonological characteristics of the section with standard speech data stored in the storage device and selecting appropriate speech segments, the pitch can be adjusted during synthesis. Even when changing the phonological properties and the duration of the entire speech unit, it is possible to maintain appropriate phonology and the duration of the entire speech unit.

(Embodiment) FIG. 4 shows an example of a speech synthesis device according to the present invention, which is composed of a microprocessor 1, a typewriter 2, a prosody memory 3, a syllable memory 4, a segment memory 5, and a waveform regenerator 6. be done. Fragment memory 5
stores the speech segment data necessary to synthesize any word.

Considering that this type of synthesizer requires mostly female sounds, the data length is 64 samples, which is sufficient to cover the pitch period of female voices with a sampling frequency of 8kHz. It is said that

The syllable memory 4 sequentially stores the start addresses of speech segment data constituting a syllable with one syllable as one block, and also stores the phoneme length of each speech segment in the syllable.

The prosody memory 3 stores coded prosodic control information such as accent and intonation for input words. The operation of the synthesis apparatus will be explained below with reference to FIG.

First, a character string of a word to be output is input to the typewriter 2.

The microprocessor 1 searches the prosody memory 3 for prosodic control information such as accent and intonation for the input word. The control information retrieved from the prosody memory 3 is in the form of accent type, intonation type, etc. codes, and the microprocessor 1 converts this into actual control data, that is, pitch data and amplitude data for each section. .

Furthermore, the microprocessor 1 decomposes the input character string into syllables, and sequentially inputs from the syllable memory 4, for each syllable, the start address of the speech element constituting each syllable and the phoneme length of the speech element. The addresses and control data obtained in this way become data for editing and combining.

Next, the speech segment selection operation which is the gist of the present invention will be explained based on the flowchart of FIG.

The present invention selects phoneme data for synthesis by referring to data representing a standard phoneme change. Corresponding phoneme length data is used as data representing a reference phoneme change.

In Figure 5, PH _NO is the number of syllables when the input character string is broken down into syllables, PH (i) (i = 1 ~
PH _NO ) is an array that temporarily stores the address sequence of the syllable memory 4 corresponding to each decomposed syllable, i is a pointer indicating the syllable address sequence,
PH _AD is the syllable address to read syllable memory 4,
S _NO is a variable that represents the number of phonetic segments that make up the syllable to be synthesized, and S (j) is the syllable memory 4
L(j) is an array that stores the phoneme length sequence read from the syllable memory 4, P(j) is an array that stores the phoneme length sequence read from the syllable memory 4, and P(j) is an array that stores the phoneme length sequence read from the syllable memory 4. An array that stores the generated pitch series, j is the array S
Fragment pointer pointing to (j), L(j), P(j),
S _AD is a reference speech segment address indicating which speech segment should be used for synthesis at the current time, that is, the address of the speech segment to be used. Also,
C _L is a phoneme length counter variable that counts phoneme length using the sampling clock, C _P is a pitch counter variable that counts pitches using the sampling clock, and C _AD is a phoneme length counter variable that counts pitches from the phoneme memory 5.
This is a speech unit readout address counter that reads out each sample.

Step SP1 is a step for performing initial settings at the start of synthesis, in which each variable is initialized and the syllable address series PH(i) and its number PH _NO determined according to the input character string are set.

At step SP2, the end of the phoneme length section is determined by determining whether a phoneme length counter variable C _L that counts the phoneme length for each sampling clock is 0. That is, if C _L =0, the segment pointer j is advanced (step SP3), and the reference speech segment address S _AD and phoneme length counter variable C _L are updated in subsequent steps SP10 and SP11. However, since the end of the phoneme length section may be the end of the syllable section or the word section, step SP4 is performed. At step SP6, the end of each syllable section and word section is determined.

At step SP4, if the syllable section has ended (j>S _NO ), advance the syllable pointer i (step SP4).
SP5), reads the information of the next syllable from the syllable memory 4, calculates the number of segments _SNO , and the speech segment start address string S.
(j) Phoneme length series L(j) and pitch series P(j)
(Steps SP7 to SP9).

Through the above series of processing, the reference speech segment address is
The start address of the standard speech segment to be selected at that time is always set in S _AD .

On the other hand, in the processes after step SP12, the reference speech unit address updating process by the phoneme length counting and the selection of a speech unit at the start of a speech unit for each pitch cycle are performed in parallel within the same sample clock. Processing takes place.

In step SP12, it is determined whether the pitch counter C _P that counts the pitch period for each sampling clock is 0 or not. If C _P = 0, the start address of the speech segment to be used for the next synthesis is referenced. Refer to the address S _AD and set the speech unit read address counter C _AD .

At steps SP15 and SP16, voice segment data is read sample by sample from the segment memory 5 according to the speech segment read address counter C _AD , and the waveform is reproduced.

A series of processes from steps SP12 to SP17 is performed every sample clock, and in the final step SP17, _CAD is incremented and _CL and _CP are decremented.

(Effects of the Invention) As described above in detail, according to the present invention, at the start of a speech segment in each pitch period during phoneme waveform reproduction, it is possible to determine what kind of phoneme is to be expressed in the interval. , by referring to standard data and selecting the appropriate speech segment, so the duration and phonology of the synthesized speech are always kept good and natural, regardless of the pitch period. can obtain synthesized speech. Therefore, by applying the method of the present invention to an arbitrary vocabulary synthesizer in the waveform domain, it is possible to obtain natural synthesized speech.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing the correspondence between phoneticity and prosody in waveform domain synthesis; FIGS. 2 and 3 are explanatory diagrams showing the correspondence between speech units and pitch periods in the present invention; and FIG. The figure is a block diagram showing one embodiment of the present invention, and FIG. 5 is a diagram showing a processing procedure for selecting a speech unit. 1... Microprocessor, 2... Typewriter, 3... Prosodic memory, 4... Monosyllabic memory, 5
...Fragment memory, 6...Waveform regenerator.

Claims (1)

[Scope of Claims] 1. A large number of data corresponding to one pitch unit of speech segment in natural speech are stored as segment units, and the segment unit is a series of speech segments such as phonemes, syllables, or words ( In a speech synthesis method, the speech synthesis method has a function of sequentially extracting data in chronological order for each unit of speech, and synthesizes speech by reading, editing and reproducing data in units of segments, wherein the units of speech are units of segments. The time series S _i (i=1,
), and the phoneme length sequence L _i (i=
1, 2, ...), and during speech synthesis, there is a means for sequentially updating the segment units to be used for synthesis each time the phoneme length L _i is multiplied, and a means for updating the pitch period given at the time of synthesis. 1. A speech synthesis method comprising means for selecting a segment unit to be synthesized for each coefficient.