JPH07319497A - Speech synthesizer - Google Patents

Abstract

(57) [Abstract] [Purpose] To provide a speech synthesis apparatus configured to reduce the storage amount of natural speech data necessary for speech synthesis while shortening the processing time while ensuring the naturalness of speech. A preprocessing unit 12 divides an input character string into phonemes, and a selection reference parameter setting unit 13 sets a prosody parameter (selection reference parameter) as a reference for selecting an optimum segment.
The segment selection unit 14 searches through the segment file management unit 15 whether there is a waveform segment corresponding to each phoneme unit, and if there is, a waveform with a minimum error from the selection reference parameter. Select and extract the pieces. On the other hand, when the waveform segment does not exist, the single syllable corresponding to the phoneme unit is regarded as the optimum segment and selectively extracted from the single syllable file. The prosodic parameters of the selected waveform segment or monosyllabic are transformed by the segment transforming unit 16 and then sequentially coupled by the segment connecting unit to obtain a synthetic voice corresponding to the input character string.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesizing device, and more particularly, to a synthetic speech from an input character string by using a waveform segment obtained by cutting out a natural speech as a synthesizing parameter and a monosyllabic speech whose prosody parameter is known. The present invention relates to a speech synthesizer to generate.

[0002]

2. Description of the Related Art FIG. 6 is a diagram showing an example of a functional block diagram of a conventional general speech synthesizer, in which 61 is an input terminal, 62 is a preprocessing section, 63 is a selection reference parameter setting section,
64 is a segment selection unit, 651 is a segment parameter table, 6
52 is a segment file, 66 is a segment transformation part, 67 is a segment connection part, and 68 is an output terminal.

In the speech synthesizer having this configuration, an input character string consisting of phoneme symbols and accent symbols is input to the input terminal 61 and then divided into phoneme units in the preprocessing unit 62. The selection reference parameter setting unit 63 sets selection reference parameters (average pitch frequency, pitch inclination, time length, average power) from the divided phoneme units and accent symbols, and outputs them to the segment selection unit 64. The selection criterion parameter is a prosody parameter used as a criterion for selecting a waveform segment that is a synthesis parameter.

The segment file 652 stores a plurality of waveform segments obtained by cutting out a natural voice such as a novel or essay. The segment parameter table 651 stores the prosody parameters of each waveform segment. Has been done. The element selection unit 64
The element parameter table 651 is searched based on the selection reference parameters, and the optimal element for phoneme concatenation is selected from the element file 652. The segment transforming unit 66 appropriately transforms the segment to match the optimum segment selected by the segment selecting unit 64 with the selection reference parameter. In the segment connecting unit 67, the deformed segments are connected according to the input character string and output through the output terminal 68.

[0005]

By the way, in the above speech synthesis processing, if the waveform segment corresponding to the phoneme unit does not exist in the segment file 652, the voice for the phoneme unit is not generated. Therefore, in order to obtain a natural synthesized speech, it is necessary to prepare at least one waveform segment corresponding to the vocabulary of the character string to be synthesized. However, when performing synthetic speech processing on a vocabulary having a high degree of arbitraryness, all the corresponding waveform segments are previously stored in the segment file 6.
It is difficult to prepare the data in No. 52 and analyze the prosody parameter in advance, and there is a certain limit in improving the voice quality. Although it is possible to newly add a waveform segment to the segment file 652, it is necessary to collect a considerably large number of waveform segments in either case, which is a problem that requires a lot of labor.

Further, as described above, storing a large number of waveform segments in the segment file 652 inevitably increases the file capacity. Therefore, there is also a problem that the time required for the combining process increases in proportion to the file size.

The present invention solves the above problems, reduces the amount of natural voice data required for voice synthesis, shortens the processing time, and is capable of obtaining a natural synthesized voice. The purpose is to provide.

[0008]

A speech synthesis apparatus provided by the present invention comprises at least a preprocessing section for dividing an input character string corresponding to a speech into phoneme units, and a plurality of waveform segments cut out from natural speech. A segment information storage means for storing prosodic parameters of each waveform segment; a monosyllabic information storage means for storing a plurality of monosyllabic and prosodic parameters of each monosyllabic;
It is searched whether or not a waveform segment corresponding to the divided phoneme unit exists in the segment information storage means, and when it exists, an error from a prosody parameter previously determined as a segment selection criterion is minimized. A waveform segment corresponding to the prosody parameter is selectively extracted. On the other hand, when the waveform segment does not exist, a single syllable corresponding to the phoneme unit is selectively extracted from the single syllable information storage unit, and the extracted waveform segment. Or a segment connecting section for connecting a single syllable in the order of the input character string to generate a synthesized voice. That is, two types of parameters, that is, a waveform segment and a monosyllabic, are used as a composite synthetic parameter.

According to another aspect of the present invention, there is provided a voice synthesizing device according to the above voice synthesizing device, wherein the prosodic parameter of a waveform segment or a single syllable selected and extracted by the segment selecting means is further modified. It is characterized in that it is configured to provide a phoneme deforming unit that brings an error with a prosodic parameter defined as a selection criterion close to zero value, and guide the waveform phoneme or monosyllable of the deformed prosodic parameter to the phoneme connecting unit. To do.

The error is preferably the sum of squares of values obtained by dividing the difference between the prosody parameter defined as the selection criterion and the extracted individual prosody parameter by the variation width of each prosody parameter.

[0011]

In the speech synthesizer of the present invention, the input character string is divided into phonological units by the preprocessing unit, and prosody parameters (selection reference parameters) which are selection references of waveform segment for each phonological unit are set. The selection criterion parameters are, for example, a desired average pitch frequency, pitch slope, time length, and average power. The segment selection unit first searches the segment information storage unit for a waveform segment corresponding to each phoneme unit. If it exists, it is selectively extracted from the element information storage means. If a plurality of waveform elements exist, for example, a squared error is calculated between the selection reference parameter and the prosody parameter of each waveform element, and the waveform element having the smallest squared error is selected and extracted. On the other hand, when the waveform segment does not exist, the single syllable corresponding to the phoneme unit is regarded as the optimum segment and selectively extracted from the single syllable information storage means. The waveform segment or the single syllable selected in this way is sequentially combined for each phonological unit according to the input character string to generate a synthetic voice.

In the speech synthesizer having another configuration of the present invention, the prosodic parameter of the waveform segment or the monosyllabic segment selected and extracted by the segment selection means is modified so that an error from the prosody parameter defined as the segment selection reference is generated. It approaches zero. Then, the waveform segment or monosyllabic segment of the transformed prosody parameter is guided to the segment segment connection part. As a result, the prosody of the synthesized voice is adjusted.

[0013]

Embodiments of the present invention will now be described in detail with reference to the drawings. FIG. 1 is a functional block diagram of a speech synthesizer according to an embodiment of the present invention, in which 11 is an input terminal, 12 is a preprocessing unit, 13 is a selection reference parameter setting unit, 14 is a segment selection unit, and 15 is a unit selection unit. A segment file management unit, 16 is a segment transformation unit, 17 is a segment connection unit, and 18 is an output terminal. The segment selection unit 14 and the segment file management unit constitute a segment selection unit of the present invention. The functions of the respective units except the segment file management unit 15 are basically the same as those of the conventional apparatus shown in FIG. 6, and thus detailed description thereof will be omitted.

FIG. 2 shows the segment file management unit 1 of this embodiment.
5 shows a configuration example of No. 5. The unit file management unit 15 includes a phoneme information management table 22, a unit file 241, a unit parameter table 231, and a single syllable parameter table 23.
2, a single syllable file 242, and a file control unit (CPU) not shown. It should be noted that the CPU, the segment file 241, and the segment parameter table 231 constitute the segment information storage means of the present invention, and similarly, CP
U, the monosyllabic file 242, and the monosyllabic parameter table 232 constitute the monosyllabic information storage means of the present invention.

The segment file 241 stores a plurality of segments extracted from words / sentences in advance in consideration of the phonological environment and the like in units of phonemes. On the other hand, the single syllable file 242 stores single syllables such as 50 syllables and consecutive vowels as they are. The waveform segment and the monosyllabic piece stored in both of these files 241 and 242 are used as a composite synthesis parameter described later. The phoneme information management table 22 has a phoneme file 241 belonging to each phoneme unit.
The number of waveform segments inside is described as, for example, numerical data. The segment parameter table 231 stores the prosody parameters of each waveform segment, and the monosyllabic parameter table 232 stores the prosody parameters of each monosyllabic. Examples of the contents of these tables 231 and 232 are shown in FIGS.

FIG. 3 shows an example of the contents of the fragment parameter table 231 for / ka /, where 30 is the table name and 31
Is a file number, 32 is a segment extraction environment, 33 is an average pitch frequency, 34 is a pitch slope, 35 is a duration, and 36 is an average power. The table name 30 functions as an index of the segment extraction environment 32, for example. "#Ka"
Is the phoneme at the beginning of the natural speech (word / sentence)
Means that no accent is given. The file number 31 is a file identification code in the segment file 241 and is selected by the segment selection unit 14. The segment extraction environment 32 indicates the words that can be extracted from the waveform segment. The illustrated example shows that the waveform segment of "# ka-" can be extracted from a word such as "# kakaeru #". The prosody parameters 33 to 36 corresponding to each segment extraction environment are values analyzed at the time of waveform cutout, and are fixed numerical data.

Further, FIG. 4 shows a monosyllabic parameter table 2
32 is an example of contents of 32, 41 is a file number, 42 is a single syllable, 43 is an average pitch frequency, 44 is a pitch slope, and 45 is a pitch slope.
Is the duration and 46 is the average power. File number 41
Is a file identification code in the monosyllabic file 242, and is a target of selection in the segment selection unit 14, like the file number 31. The monosyllabic 42 corresponds to the monosyllabic stored in the monosyllabic file 242. The illustrated example shows that 110 single syllables such as a, i, ... Wa are stored in the single syllable file 242. It is preferable that there are as many types of monosyllables as possible. The prosody parameters 43 to 46 corresponding to each monosyllabic
Is fixed numerical data analyzed in advance.

Next, the contents of the voice synthesizing process in the voice synthesizing apparatus having the above configuration will be described. An input character string consisting of phonetic symbols and accent symbols is input from the input terminal 11 and then divided into phoneme units in the preprocessing unit 12.
In the present embodiment, the phoneme unit is, in addition to phonemes such as / a / and / ka /, in consideration of the smoothness of segment connection / ai /
It includes continuous vowels such as and compound syllables such as / aN /. The selection criterion parameter setting unit 13 sets a prosody parameter that is a criterion for selecting an optimum waveform segment (optimal segment) corresponding to a divided phoneme unit, that is, a selection criterion parameter. In this embodiment, all of the above-mentioned average pitch frequency, pitch slope, time length, and average power are used as the selection reference parameter, but this selection reference parameter may be appropriately changed according to the prosodic condition of the synthetic speech. .

In the segment selection section 14, the segment file 241
The optimum segment for each phonological unit is selected by using each prosody parameter of the waveform segment belonging to (1) or the single syllable belonging to the single syllable file 242 and the selection reference parameter. The detailed operation of the segment selection unit 14 will be specifically described with reference to FIG. Here, FIG. 5 is a fragment selection unit 1 according to the present invention.
4 is a flowchart showing the operating principle of No. 4, in which S represents each processing step.

First, the number of waveform segments for each phoneme unit is obtained by referring to the phoneme information management table 22 (S31). Then, the contents of the selection process are determined as follows based on the obtained number of waveform segments (S32). When the number of waveform segments is 0, that is, when the waveform segment does not exist, the single syllable corresponding to the phoneme unit is regarded as the optimum segment and selected (S331). For example, when there is no waveform segment corresponding to the phoneme unit / di /, the monosyllabic parameter table 232 is searched and the monosyllabic di of the file number 110 (see FIG. 4) is automatically regarded as the optimum segment. On the other hand, when the number of waveform segments is not 0, the optimum segment is selected using the selection reference parameter and the prosody parameter of the waveform segment (S332). The number of waveform segments is 1
In the case of, the waveform element is set as the optimum element. An arbitrary selection method may be used when the number of waveform segments is two or more. For example, a squared error is calculated between the selection reference parameter and each prosody parameter, and the squared error is minimized. If the object is the optimal segment, there is an advantage that the optimal segment of the prosody parameter closer to the selection criterion parameter can be obtained. The file corresponding to the optimal segment selected in this way and its prosody parameter are output to the segment transformation unit (S34).

For example, the file number 074 in FIG.
If the waveform segment belonging to 1 is the optimal segment, the file corresponding to the file number is assigned to the segment file 241.
Extracted and output, and its prosody parameter 33
~ 36 is output. Similarly, when the monosyllabic di belonging to the file number 110 in FIG. 4 is regarded as the optimum segment, the file corresponding to the file number is extracted from the monosyllabic file 242 and output, and the prosody parameter 43 of the file is extracted. ~ 46 is output.

The segment transforming unit 16 transforms the prosody parameters of the file derived from the segment selecting unit 14 so as to approach the selection reference parameters. This change process is
When the file content is a waveform element, a conventional general method can be used. On the other hand, in the case of a single syllable, for example, the process is to thin out or interpolate the time length of the prosody parameter so as to approach the time length of the corresponding selection reference parameter. Other prosodic parameters can be modified by pitch control, frequency control, or the like. The reason why the transformation process is also performed on the prosodic parameters of the monosyllabic is that the naturalness of the synthesized speech cannot be improved simply by extracting and connecting the corresponding monosyllabic file from the monosyllabic file 242. It is also for completely ensuring the substitutability for the corrugated element.

In the element piece connecting portion 17, the element piece deforming portion 16 is provided.
Synthesized speech is generated by sequentially combining the waveform segment or the single syllable transformed in (3) for each phoneme unit according to the input character string. Through the series of processes described above, the output terminal 18
In, a natural synthetic voice with no omission corresponding to the input character string is output.

As described above, according to this embodiment, even when there is no waveform segment corresponding to the divided phoneme unit,
It is possible to perform speech synthesis by selecting an appropriate corresponding monosyllable. Therefore, unlike the conventional case, it is not necessary to prepare waveform segments for all phoneme units in advance, and voice synthesis can be performed even from a small amount of natural voice data.

[0025]

As is apparent from the above description, the speech synthesizer of the present invention uses two types of parameters, a waveform segment and a monosyllabic, as a composite synthesis parameter, and a waveform segment corresponding to a phoneme unit. Exists in the segment information storage means, the waveform segment corresponding to the prosody parameter that minimizes the error from the prosody parameter previously determined as the segment selection criterion is selected and extracted. Even if the number of vocabularies to be synthesized increases, the singular syllable corresponding to the unit is selectively extracted from the single syllable information storage means, and the extracted waveform segments or single syllables are connected in the order of the input character string. It is possible to generate natural synthetic speech without omissions by using only pre-created natural speech data, that is, the waveform segment and the monosyllabic, and its prosody parameters without newly adding the corresponding waveform segment. That. Moreover, since it is not necessary to add the natural voice data as described above, the file capacity can be reduced. This makes it simple and low cost,
In addition, it is possible to realize a voice synthesizing device having a configuration in which the voice synthesizing processing time is shortened and a high quality synthetic voice is generated within a fixed time.

A speech synthesizer according to another configuration of the present invention modifies the prosodic parameter of the waveform segment or the single syllable selected by the segment selection unit to obtain a prosodic parameter as a reference for segment selection. Since the elemental piece deforming unit to be brought close to is provided and the deformed waveform elemental pieces are connected to generate the synthetic speech, it is possible to generate the synthetic speech of a desired prosody at high speed.

[Brief description of drawings]

FIG. 1 is a functional block diagram of a voice synthesizer according to an embodiment of the present invention.

FIG. 2 is a configuration diagram of a segment file management unit in the speech synthesizer of this embodiment.

FIG. 3 is an explanatory diagram of contents of a segment parameter table that constitutes the segment file management unit.

FIG. 4 is an explanatory diagram of the contents of a monosyllabic parameter table that constitutes the segment file management unit.

FIG. 5 is a flowchart showing the operation of a phoneme selection unit in the speech synthesizer of this embodiment.

FIG. 6 is a functional block diagram of a conventional speech synthesizer.

[Explanation of symbols]

 12 preprocessing unit 13 selection reference parameter setting unit 14 unit selection unit 15 unit file management unit 16 unit transformation unit 17 unit connection unit 22 phoneme information management table 231 unit parameter table 232 monosyllabic parameter table 241 unit file 242 monosyllabic files

Claims (3)

[Claims] 1. A preprocessing unit for dividing an input character string corresponding to a voice into phonological units, a plurality of waveform units cut out from a natural voice, and a unit information storage unit for storing a prosody parameter of each waveform unit. A monosyllabic information storage means for storing a plurality of monosyllabic and prosodic parameters of each monosyllabic, and whether or not a waveform segment corresponding to the divided phonological unit exists in the segment information storage means. When it exists, the waveform segment corresponding to the prosody parameter that minimizes the error from the prosody parameter previously determined as the segment selection criterion is selected and extracted, and when it does not exist, the single segment corresponding to the phoneme unit is extracted. At least a segment selection unit that selectively extracts syllables from the single syllabic information storage unit, and a segment connection unit that connects the extracted waveform segments or single syllables in the order of the input character string to generate a synthesized voice. Characterized by having Speech synthesis device that. 2. The speech synthesizer according to claim 1, wherein an error from a prosodic parameter defined as the phoneme selection reference is obtained by modifying a prosodic parameter of a waveform phoneme or a single syllable selected and extracted by the phoneme selection unit. A speech synthesizer characterized in that it is provided with a segment transformation unit that brings the element closer to a zero value, and leads the waveform segment or monosyllabic segment of the transformed prosody parameter to the segment connection unit. 3. The speech synthesizer according to claim 1, wherein the error is obtained by dividing a difference between a prosody parameter defined as the selection criterion and an extracted individual prosody parameter by a variation width of each prosody parameter. A speech synthesizer characterized by being a sum of squared values.