JPH10207488A - Method of making voice parts, voice part data base, and method of voice synthesis - Google Patents

Abstract

PROBLEM TO BE SOLVED: To allow a synthesized voice to be made by a simple method and be easy to hear with a low parts count. SOLUTION: A voice of reading a sentence at a constant speed 'SHIAWASE HA ARUI-...', of which a row of the phoneme and meter of the character string is clear, is recorded in a medium 8; a time length for reproducing the first two syllables 'SHIA' is measured; at intervals of this time length (for instance 200msec), the reproduced voice on the medium 8 is successively cut to voice waveforms of each two syllables 8a, 8b,...; the character string of the recorded sentence is made to correspond to these voice waveforms 8a, 8b,..., to which a phoneme 'SHIA-' and a meter (ascending type), a phoneme 'WASE' and a meter (ascending type),... are added; and these are stored in addresses of different databases.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech component creation method for creating speech waveforms used for speech synthesis as speech components from continuous human utterance speech, and a speech component database storing various speech waveforms used for speech synthesis. The present invention also relates to a speech synthesis method for generating a synthesized speech corresponding to a Japanese reading sentence using the speech component database.

[0002]

2. Description of the Related Art Heretofore, the following first to third methods have been known as a recording / editing type speech synthesizing method using human speech as a synthetic speech unit. The first method is to cut out the recorded human voice,
Japanese 5 such as "I", "U", "E", "O" ...
Create all Japanese one-syllable voice parts, such as zero sounds, “Kyu”, “Sha”, etc.
This is a method in which this is registered in a database or the like, and the voice parts are combined to synthesize a Japanese reading sentence. For example, Japanese reading sentences are synthesized by combining "Ni", "Ho", "N", and voice parts for each syllable, such as Ni / ho / n.

[0003] However, when Japanese reading sentences include long sounds or prompting sounds such as "Tokyo" or "Roppongi", the speech synthesized by the first method is used. Since the unnaturalness is conspicuous, voice parts of two or more syllables are generally registered in a database as in the second and third methods described below, and the voice parts corresponding to the Japanese reading sentence are taken out. A method of generating a synthesized speech of a word reading sentence has been adopted.

[0004] A second method is to create a speech component for each syllable composed of multiple syllables by cutting out from a recorded human voice, registering these in a database, and compiling a speech component corresponding to a Japanese reading sentence. Is extracted for each phrase and connected to speech-synthesize a Japanese reading sentence. In the speech parts for each of the phrases, the same phoneme sequence (reading sequence) with various prosody changes is stored in a database. For example, speech parts corresponding to a phrase consisting of a combination of the noun "Osaka" and a particle include the following phonemic sequences: "Osaka is", "Osaka is", "Osaka is", "To Osaka", and "Osaka is". Some have added various prosody changes, and these are stored in a database.

[0005] Then, the sentence I want to synthesize is sent to / tomorrow / Osaka / by text analysis. (/
Indicates a phrase break. ), The words "I", "Tomorrow",
Speech parts corresponding to the phoneme sequence and the prosody sequence of each of the phrases “to Osaka” and “go” are extracted from the database, and these are connected to perform speech synthesis.

[0006] A third method is to combine Japanese voice sentences corresponding to all part-of-speech phrases, including adjectives and verb inflected phrases, which are cut out from a recorded human voice and created and registered. This is a method of synthesizing voice. Also in this case, similarly to the second method, speech components having various prosody changes for the same phoneme sequence are stored in the database.

Then, the sentence to be synthesized is analyzed by text, and I / ha / tomorrow / osaka / go / go /. (/ Indicates a part-of-speech delimiter.) When the parts of speech are decomposed for each part of speech as shown in FIG. Is synthesized.

[0008]

On the other hand, the work of cutting out audio components from a recorded human voice is time-consuming because it is necessary to select a portion having desired phonemes and prosody while looking at the voice waveform. There is a demand for a method of creating audio components that requires as few audio components as possible. However, according to the above-mentioned conventional second method, a voice component having various phoneme strings / prosodic strings for each phrase is required, and even if only the number of combinations of nouns and particles and the number of combinations of verbs and auxiliary verbs are considered, Unless a huge number of audio parts must be created and a limited number of Japanese reading sentences are synthesized, it takes time and effort to create the audio parts and it is difficult to put them to practical use.

In the case of the third method, the number of voice parts is small as compared with the case of the second method. In this case, however, at least the number of words of all parts of speech and an adjective,
Considering also inflected phrases such as verbs, etc., it is necessary to have a number of voice parts far exceeding the number of words included in the Japanese dictionary (specifically, 100,000 words or more). It takes time and effort to create audio parts as in the case of the above method.

The present invention requires only a small number of audio components, does not require much time for creation, and can provide an audio component that incorporates human vocal rhythms, and stores the audio component. Was. An object of the present invention is to provide a voice component database and a method for performing voice synthesis using the voice component database.

[0011]

SUMMARY OF THE INVENTION According to the present invention, there is provided a method for producing a voice component, comprising: recording a speech in which a sequence of character strings and a phoneme and a prosody are clearly read out at a constant speed; The first recording and playback time for the number of beats (syllables) based on the utterance rhythm, that is, the number of syllables in the cycle of the utterance rhythm, is set as the unit time, and the recorded voice is sequentially cut from the beginning for each of the unit times. The main task is to create a sound piece by adding a corresponding character string of the sentence and its prosody to the recorded piece in the order cut out one by one. Characteristics.

In the audio component database of the present invention, the number of syllables constituting the various audio waveforms stored is the same,
In addition, the lengths of the voice waveforms are the same, and the voice waveforms are stored at different addresses according to the constituent phonemes and the prosodic types.
Also, the first speech synthesis method of the present invention is characterized in that the speech component database of the present invention, a recorded basic sentence holding means for recording a voice of a base sentence other than an inserted phrase portion in a response sentence, and a voice component for assembling the inserted phrase. And a character information inputting step of inputting a character string applicable to the insertion phrase, and a reference to the insertion word / phrase component combination dictionary. A synthesized voice assembly process of selecting and connecting each voice component corresponding to the input character string from the voice component database, a voice synthesis process of synthesizing the connected voice component as a synthesized voice of the inserted phrase portion, Outputting a voice of a response sentence by inserting the synthesized speech of the inserted phrase portion into the recorded base sentence speech.

Further, in the second speech synthesis method according to the present invention, the speech component database of the present invention and the storage address of each speech component for assembling each fixed sentence with respect to various fixed sentences on the speech component database are provided. A character information inputting step of inputting a character string of the fixed sentence, and a voice corresponding to the input character string by referring to the fixed sentence part combined dictionary. A synthesized voice assembly process for selecting and connecting a component from the voice component database; a voice synthesis process for synthesizing the connected voice component as a synthesized voice; and an output process for outputting the synthesized voice. .

Further, in a third speech synthesis method of the present invention, the speech component database of the present invention and a storage address on the speech component database of each speech component for assembling each phrase for various phrases are described. A character part input dictionary, comprising: an arbitrary character string and a phrase delimiter and a pause position of the arbitrary character string;
A character string synthesis voice assembling step of selecting and connecting each voice component corresponding to the input arbitrary character string from the voice component database with reference to the phrase component connection dictionary, and using the connected voice components as synthesized voice It is characterized by having a voice synthesis process for synthesizing and an output process for outputting the synthesized voice.

As described above, according to the present invention, a part of a human voice recorded at a constant speed is cut out at every unit time based on a human utterance rhythm to generate a voice part necessary for assembling a synthesized voice. Then, a human voice rhythm is incorporated into the voice component so that the human voice rhythm can be incorporated into the synthesized voice generated from the voice component, and the voice component can be easily created.

[0016]

According to the present invention, first, a sentence including various phonemes and prosody and in which the arrangement of character strings is clear is read out at a constant speed and recorded on a recording medium. The sound of this recording medium is reproduced, and the recording and reproduction time corresponding to the first beat (syllabic number) based on the human vocal rhythm is measured while observing the sound waveform, that is, the syllable number is measured from the beginning of the reproduced sound. The time corresponding to one cycle of the human vocal rhythm as a unit is measured, and the measured recording / reproducing time is set as a unit time. The recorded voice is simply mechanically cut every unit time from the beginning to create a plurality of recording pieces each composed of a voice waveform of a fixed time length.

A phoneme and a prosody of a character string included in the recording piece are added to the recording piece (voice waveform) by referring to the sentence in the cut order. Then, recording pieces having various phonemes and prosody are collected to form audio components. This audio component is stored in the audio component database.
Each voice component stored in this database has the same number of syllables and the same length, and is stored at different addresses depending on the combination of the phoneme and the prosody type.
As described below, an appropriate voice component is selected and connected from the voice component database according to the input character string or phrase break to generate a synthesized voice.

For example, when it is desired to output a response sentence including an inserted phrase by voice, a voice of a sentence other than the inserted phrase is recorded in advance as a recording base voice, and a character string applicable to the inserted phrase portion is input. With reference to the inserted word / phrase part combination dictionary based on the extracted character string, a voice part for assembling the inserted phrase part is selected and connected from the voice part database, and the connected voice part is synthesized as a synthesized voice. The synthesized speech of the inserted phrase is inserted into the already-recorded base sentence speech, and a response sentence is output as speech.

For example, when one of several types of fixed phrases is output as speech, the storage addresses in the audio component database of each voice component for assembling each of the fixed phrases are fixed. A voice component for assembling a voice of a standard sentence from a voice component database is selected and connected from the input standard text character string by referring to the standard text component binding dictionary, and the connection is made. The voice component is synthesized as synthesized voice, and a synthesized voice of a fixed sentence is output.

Further, for example, when an arbitrary sentence is to be synthesized by speech, an arbitrary character string to be synthesized and character information of a phrase break and a pause position of the arbitrary character string are input. A speech component for assembling a phrase is selected from the speech component database and connected to each phrase of an arbitrary sentence with reference to the phrase component combination dictionary based on the input character information, and finally all phrases, that is, an arbitrary character string is assembled. The voice component is selected and connected, a voice component obtained by adding a pause to the connected voice component is synthesized as a synthesized voice, and a synthesized voice of the arbitrary character string is output.

[0021]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a flow chart showing an embodiment of a voice component creating method according to the present invention. The voice component creation method of this embodiment includes a database address setting process S1 and
A constant speed recording step S2, a unit time setting step S3, a recording cutting step S4, a recording small piece phoneme prosody provision step S5,
It comprises an address assigning step S6 and a storing step S7.

In the database address setting step S1, an address of the voice component database 1 for storing voice components as a unit for generating a synthesized voice is set in advance. Each speech component to be created is composed of two syllables and a fixed-length speech waveform having various phonemes and prosody required for speech synthesis. FIG. 2 shows a phoneme matrix of a voice part 2 composed of two syllables as a partial configuration of the voice part database 1. On the row side and on the column side, 50 Japanese syllables, muddy sounds, muddy sounds, semi-voiced sounds, murmurs, prompting sounds, prolonged sounds, and silence are arranged, respectively. The second sound is arranged. In the figure, phonemes are indicated by readings. Not all elements of this phoneme matrix are required as units for speech synthesis,
For example, among the phonemes “school” and “ko” of “school”
The number of voice parts 2 required for voice-synthesizing Japanese-read sentences can be reduced to several thousands, for example, "ko" can be sufficiently substituted for "ko".

In the database address setting step S1, each element of the phoneme matrix is assigned "Ah" with "Ah" as an address.
Are set in order, such as "2". Further, each of the two syllables, which are the elements of the phoneme matrix, is given three types of prosody changes (a) flat type, (b) ascending type, and (c) descending type as shown in FIG. The flat type is a prosody type in which the first and second sounds have no change in intonation, the ascending type is a prosody type in which the second sound has a stronger accent than the first sound, and the descending type is a prosody type in which the second sound is less pronounced. Prosodic type with less accent than the first note.

FIG. 5 shows an overall image of the audio component database 1. FIG. 4A shows a storage form of various voice parts 2 formed of a phoneme matrix having a flat prosody, and FIG. 4B shows a storage form of various sound parts 2 formed of a phoneme matrix having an ascending prosody. , C in the same figure shows the storage form of various audio components 2 each composed of a phoneme matrix having a descending prosody.

As shown in the figure, a branch address is set in the voice part 2 according to the prosodic type in addition to the address of the phoneme matrix. If the prosodic type is a flat type, the branch address is 1, the prosodic type is an ascending type, the branch address is 2, and if the prosodic type is a descending type, the branch address is 3. For example, the address 1-1 of the voice part 2 is a flat type of the phoneme "Ah", and the prosody of the first sound "A" and the second sound "A" has no change in intonation. Address 2-3 is the descending type of the phoneme "Ai" and the first sound "A"
It is assumed that the second sound "i" has a weaker accent.

In the constant speed recording step S2, a sentence in which the arrangement of the phonemes and the prosody of the character string are clear is recorded at a constant speed as shown in FIG. Specifically, first, the arrangement of the phoneme sequence and the prosody sequence is clear, and when divided into two syllables, general plural syllables having the phoneme and the prosody required for speech synthesis are obtained. Prepare sentences. Thus 2
In order to separate each syllable, taking into account that the natural biological rhythm and the vocal rhythm of human beings are in two beats, like a march or a cheering song is in two beats, by incorporating these two beats into the audio component 2 The voice synthesized by the voice component 2 also keeps a natural rhythm, so that the voice is natural and easy for the listener to hear. That is, the basic number of beats (one beat is equivalent to one syllable) for taking in the human vocal rhythm is set to two syllables.

The sentence is read by the speaker 9 at a constant speed, and is recorded on the recording medium 8. In this case, it is preferable that the speaker 9 is trained to read at a constant speed. The unit time setting step S3 is based on the recording pieces 8a, 8b,.
・ ・ Set the unit time to cut out. That is, the recording medium 8 is reproduced, and the recording / reproducing time including the first two syllables (the number of rhythms of the vocal rhythm) is measured while observing the audio waveform, and the recording / reproducing time is calculated as a unit time (for example, 200 ms).
Set as

In the recording cutting step S4, the audio waveform recorded on the recording medium 8 is simply cut in order at the set unit time, and the recording pieces 8a, 8b,.・ ・ Cut out. These recording pieces 8a,
8b,... Each include a human vocal rhythm based on two syllables. In the recording small piece phonetic prosody provision step S5, phonemes and prosody are provided to the recording small pieces 8a, 8b,. Since the arrangement of the phoneme sequence and the prosody sequence of the sentence recorded on the recording medium 8 is clear as described above, the phoneme and the prosody of each recording piece are determined in the order in which the recording piece was cut out. Can be easily read.

Then, recording pieces having various phonemes and prosody necessary for generation of a synthesized speech are arranged in a line, and these are designated as speech parts 2 respectively. The address assignment process S6 includes the audio component 2
Component 2 based on the phoneme and prosody information given to
The address of the audio component 2 on the audio component database 1 that stores the

In the storing step S7, based on the address information of the audio component 2, the audio component 2 is inserted into an address on the audio component database 1 shown in FIGS. Next, a specific example of the audio component creation method having the above configuration will be described with reference to FIG. More specifically, as shown in the figure, the sentence "Happiness does not come. I walk." The prosody of the "Shia" part is ascending,
The character string information such as the rising type is clarified in the part of “seed”. This sentence shows a part of the sentence that the speaker 9 reads. ) And the speaker 9 blows at a constant speed. As described above, since the speaker 9 reads at a constant speed, the length of the pause put in the appropriate position of the sentence is n syllables (n: integer) (constant speed recording process S2).

Then, the recording medium 8 is recorded as "Happiness (pause) is not present (pause), so (pause) is going on." The time for reproducing the recording piece 8a including the first two syllables "Shia" of the recorded sentence is measured. For example, if the recording and reproduction time is 200 ms, this 200 ms is set as a unit time (unit time setting). Step S3).

Next, the reproduction sound of the recording medium 8 is sequentially cut every unit time (200 ms) to cut out the recording pieces 8a, 8b,... For every two syllables (recording cutting step S4). These recording pieces 8a, 8b,.
··· Correspond to the character string of the recorded sentence, and assign phonemes and prosody such as phoneme “shia” prosody (ascending type), phoneme “set” prosody (ascending type)... I do.
The recording piece to which the phoneme and the prosody are given is defined as the audio component 2 (recording piece phonetic prosody provision step S5).

The address on the audio component database 1 is specified from the phoneme and the prosody assigned to the audio component 2. For example, the voice part 2 of the first phoneme “Shia” prosody (ascending type) portion is specified as address 741-2 (address assignment process S
6). The voice component 2 to which the address has been assigned is inserted and stored in the address location on the voice component database 1 (storage step S7).

Next, a speech synthesizing apparatus using the speech parts created as described above will be described. As shown in FIG. 4, this speech synthesizer includes a speech component database 1, a character information input unit 3, a component combination dictionary 4, a synthesized speech assembly unit 5, and a speech output unit 7. I have. The voice component database 1 stores voice components 2 having the same various phonemes and prosody as described above.

The character information input section 3 inputs character information corresponding to a voice to be synthesized. As the character information, in addition to the character string, a paragraph break and a pause position are input as necessary. The component combination dictionary 4 describes combinations of voice components 2 for various types of input character information so that voice synthesis is performed based on the character information input from the character information input unit 3, and each voice component 2 has a voice component database. The upper address is designated.

The synthesized speech assembling unit 5 converts the speech component 2 for generating a synthesized speech based on the input character information such as a character string or a phrase segment input from the character information input unit 3 or a pause position into the speech component database 1. Choose from and connect.
The voice synthesizer 6 processes the voice component 2 connected by the synthesized voice assembler 5 so as to be a smooth continuous voice and synthesizes it as a synthesized voice.

The voice output unit 7 outputs the synthesized voice. Next, a specific example of this embodiment having the above configuration will be described. (Specific Example-1) The inserted phrase is made into a synthesized speech. The character information input by the user is used as an insertion phrase, and a fixed portion of the response sentence other than the insertion phrase is recorded in advance as a recording base sentence voice. Then, based on the character information input to the character information input unit 3, a synthesized speech of the inserted phrase is generated, and this synthesized speech is inserted into the recorded base sentence speech to output a response sentence.

For example, when inquiring of a telephone number,
After outputting the sentence "Please enter the name of the prefecture or city where you want to look up.", And the user responds with the name of the prefecture or city. If you want to output the response sentence "Is it okay with ○○○○○?" As a response sentence to confirm this response, pre-record the fixed part of the sentence "Is it okay?" In advance, "XX
The voice of "○○○" is synthesized as an insertion phrase of the recording base sentence based on character information of a specific prefecture name or city / ward / county name input by the user.

On the other hand, as shown in FIG. 7, the combinations of the voice parts 2 necessary for generating the synthesized voice and the addresses of the respective voice parts in the voice part database are shown in FIG. Are described in an insertion phrase part combination dictionary as the part combination dictionary 4. In the figure, both the "kanji" character string and the "kana" character string are received as keywords as input character information of the inserted phrase.

For example, when the user inputs a character string such as "Okayama Prefecture" or "Okayamaken" to the character information input section 3, the synthesized speech assembling section 5 inserts the character string shown in FIG. 286-
The audio component of the "oka" portion at address 2, the audio component of the "yama" portion at address 2482-2, and the audio component of the "ken" portion at address 610-2 are selected and connected.

The speech synthesizing unit 6 synthesizes the speech parts connected to the inserted phrase as a synthesized speech of smooth continuous speech. Then, the voice output unit 7 fits the synthesized voice of the inserted phrase portion into the recording base voice and outputs a response text. In the case where the place names of prefectures and municipalities in Japan are speech-synthesized as an insertion phrase as in the specific example-1, if there are 1175 types of speech parts 2 for about 4000 place names actually present, speech synthesis is performed. Experiments have shown that this is sufficient for the formation of (Specific Example-2) Output a fixed phrase.

In the case of the above-mentioned specific example-1, the words in the inserted phrase are synthesized by voice, but in the specific example-2, the fixed phrase is synthesized by voice. In this case, speech synthesis such as "Please enter the name of the prefecture where the person you want to look for" or "Thank you." The combination of the audio components 2 to be performed and a pause at an appropriate position, and the address of each audio component on the audio component database are determined by the fixed-text component connection dictionary as the component connection dictionary 4 shown in FIG. It is described in. The fixed phrase component combination dictionary shown in FIG. 3 is configured to accept either a Japanese character string or a kana character string of a fixed phrase as a keyword.

The synthesized speech assembling unit 5 refers to the fixed phrase component combination dictionary based on the character information of the fixed phrase character string input from the user to the character information input unit 3 and outputs the voice corresponding to the fixed phrase. Select and connect the part 2 and in the speech synthesis unit 6,
The connected voice component is synthesized into synthesized voice, and the voice output unit 7
Outputs this synthesized voice. Specifically, if you want to output the fixed phrase "This is a telephone number guide.", For example, enter "This is a telephone number" in the character information input unit 3 using a kana character string. Based on the input character string, the synthesized voice assembling unit 5 refers to the fixed phrase component combination dictionary to select the voice component 2 at address 716-1 for the phoneme "Kochi" and 2603-2 for the phoneme "Rawa". The audio component 2 at the address is selected, and the audio component 2 is selected and connected from the audio component database 1 and so on.

In this case, when the user wants to output the fixed phrase by voice, the fixed phrase is assigned a number in advance without inputting the entire character string of the fixed phrase as a Japanese character string or a kana character string as described above. It is also possible to output a synthesized voice of a fixed sentence by simply inputting the fixed sentence number to the character information input unit 3 by using the fixed sentence number as a keyword. (Specific Example-3) Output an arbitrary sentence.

In the case of synthesizing an arbitrary sentence, a character string of this arbitrary sentence, a segment break and a pause position are input to the character information input unit 3. On the other hand, as shown in FIG. 9, the phrase component combination dictionary serving as the component combination dictionary 4 describes the voice component 2 for assembling a phrase in phrase units and addresses of the respective voice components in the voice component database. Also in the case of the phrase part combination dictionary shown in FIG. 6, keywords are set so that either a Japanese character string or a kana character string can be accepted as a phrase. For example, the phrase string "I am" or "I am"
Is the voice part 2 whose phoneme "wa" is at address 2901-2
It is described in the phrase part connection dictionary that the phoneme "wrinkle" part is assembled from the voice part 2 at address 780-2.

For the creation of the phrase part combination dictionary,
As shown in the second conventional method, a large number of phrase phrases are recorded in the dictionary, and as in the second conventional method, the huge number of voice parts are It is not a hard work to cut out while observing the waveform. In this case, for this huge number of phrase words,
Since the address of each audio component 2 for assembling a phrase is simply stored and stored in a predetermined storage medium, the work of creating the dictionary is easily performed.

In the synthesized speech assembling section 5, the character information input section 3
Based on the character string of an arbitrary sentence input to the phrase and a plurality of phrase words recognized from the phrase delimiter, a speech component database necessary for speech synthesis is obtained for each phrase by referring to a phrase component combination dictionary. Choose from and connect them. For example, if the optional sentence “I go to school.” Is divided into phrases, “I will go to school / goes.” (/ Indicates a phrase break). Since the pause position is inserted between "go to school", the character strings of the phrase "I am", the phrase "go to school" and the phrase "go" and the character information of the pause position are input to the character information input unit 3. .

The synthesized speech assembling section 5 selects a speech component 2 for assembling a phrase from the speech component database 1 with reference to a phrase component combination dictionary for each phrase, and connects them. For the phrase “I am”, the voice part 2 of the phoneme “Wata” at address 2901-2 and the voice part 2 of the phoneme “Wrinkle” at address 780-2 are selected from the voice part database 1, and the phrase “School” The part “He” is the sound part 2 of the phoneme “ga” part at address 2520-2, the sound part 2 of the phoneme “ko” part at address 623-1, and the phoneme “e (silence) at address 48-1. )) Part of the audio part database 1
, The audio component 2 is selected for each phrase in order, and these selected audio components 2 are connected.

Further, the synthesized voice assembling section 5 gives a pause of, for example, 100 ms to the connected body of the voice parts 2 for assembling all the phrases according to the designated pause position. For example, in the example of the preceding sentence, a pause is given between the phonemes "Wata" and "Wrinkle" and the phonemes "Gatsu" and "Kou" because there are punctuation marks. The voice synthesizing unit 6 synthesizes the connected body of the voice parts 2 to which the pause has been given as a smoothly continuous synthesized voice, and the voice output unit 7 outputs the synthesized voice.

As described above, according to this embodiment, the recording medium 8 is reproduced to observe the sound waveform of the first two syllables, which is the basis of the human vocal rhythm, and the recording / reproducing time at this time is used as a unit. If the time is set, the audio component 2 composed of two syllables can be cut out by simply cutting the reproduction signal of the recording medium 8 for each unit time. Therefore, compared with the conventional operation of cutting out all the audio components while reproducing the audio component from the recording medium and observing the audio waveform, it can be performed very easily.

Also, the number of audio components 2 does not exceed 100,000 types as in the second and third prior arts, but can be several thousand types. Can be omitted. Furthermore, according to this embodiment, since the voice component 2 composed of two syllables incorporating the natural vocal rhythm of the human in two beats is used as a unit for generating a synthesized voice, the synthesized voice obtained by combining the voice components 2 is also a human voice. The natural utterance rhythm is adopted, and the synthesized speech becomes easy to hear.

The present invention is not limited to the above-described embodiment, but may have various embodiments. For example, in each specific example of the above-described embodiment, the input of a character string is made to accept either a Japanese character string including a kanji character or a kana character string, but a keyword is set so that only an input of a kana character string is accepted. Is also good.

Further, in this embodiment, the audio component 2 stored in the audio component database 1 has three types of intonation changes, but may have more abundant intonation changes.

[0054]

As is apparent from the above description, when the voice component creation method of the present invention is used, a voice component is cut out from a recording medium in which a human voice is recorded at a constant speed by a simple method, which is cut out every unit time. Can be created. Compared with the conventional method of cutting out all the audio components while observing the audio waveform, it is not necessary to create the audio components.

Further, since the number of audio components can be reduced, the time and effort for creating the components are further reduced. Furthermore, since the unit time cut out from the recording medium is based on the human vocal rhythm, the natural voice rhythm of the human is incorporated into the voice component cut out for each unit time. The natural utterance rhythm of human beings is also incorporated into the synthesized speech, and the synthesized speech becomes easy to hear.

[Brief description of the drawings]

FIG. 1 is a diagram showing a processing procedure of an embodiment of a voice component creation method according to the present invention.

FIG. 2 is a diagram showing an example of a phoneme matrix of a voice component database.

FIG. 3 is a diagram showing three types of prosody changes.

FIG. 4 is a block diagram showing a functional configuration example of a speech synthesis device to which the speech synthesis method of the present invention is applied.

FIG. 5 is a diagram showing an example of an overall image of a sound component database according to the present invention.

FIG. 6 is an explanatory diagram showing a method for creating a voice component.

FIG. 7 is a diagram illustrating an example of an insertion phrase part combination dictionary.

FIG. 8 is a diagram showing an example of a fixed phrase component combination dictionary.

FIG. 9 is a diagram showing an example of a phrase part combination dictionary.

Claims (5)

[Claims] 1. A method of recording a speech in which a sentence in which the arrangement of character strings and phonology and prosody are clear is read out at a constant speed, and the first recording of the reproduced sound of the recorded voice for the number of syllables in the period of the human utterance rhythm. A playback time is set as a unit time, and the recorded voice is cut in order from the beginning for each unit time to create a recording piece, and each of the recording pieces is written in the cut-out order of the sentence. A speech component creation method, wherein a speech component is created by adding a corresponding character string and its prosody. 2. A database storing various voice waveforms, wherein each voice waveform has the same number of constituent syllables and the same length, and each voice waveform has its constituent phonemes and prosody. A voice component database stored at different addresses according to types. 3. A voice component database according to claim 2, a recording base sentence holding means for recording a base voice other than an insertion phrase portion in a response sentence, and an address on the voice component database of each voice component for assembling the insertion phrase. And a character information inputting step of inputting a character string applicable to the insertion phrase, and referring to the insertion word / parts combination dictionary to correspond to the input character string. A voice synthesis process for selecting and connecting each voice component to be performed from the voice component database; a voice synthesis process for synthesizing the connected voice component as a synthesized voice of the inserted phrase portion; and a synthesized voice of the inserted phrase portion. Outputting a voice of the response sentence by being fitted into the recorded base sentence voice. 4. A speech part database according to claim 2, a fixed phrase part connection dictionary in which addresses of the respective speech parts for assembling each of the fixed parts for the various fixed parts are described on the speech part database. And a character information inputting step of inputting a character string of the fixed sentence; and selecting and connecting each of the sound components corresponding to the input character string from the sound component database with reference to the fixed sentence component combination dictionary. A voice synthesis process for synthesizing the connected voice component as a synthesized voice, and an output process for outputting the synthesized voice. 5. An audio component database according to claim 2, and a phrase component connection dictionary in which addresses of the audio components for assembling each phrase for various phrases on the audio component database are provided. A character information inputting step of inputting a character string and a phrase delimiter and a pause position of the arbitrary character string; and referring to the phrase part combination dictionary to input each of the audio parts corresponding to the input arbitrary character string from the audio part database. A speech synthesis method comprising: a character string synthesis speech assembly process of selecting and connecting; a speech synthesis process of synthesizing the connected speech component as a synthesized speech; and an output process of outputting the synthesized speech.