JP2015172625A - Voice synthesizer, synthesized voice editing method, and synthesized voice editing computer program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a voice synthesizer capable of preventing accent errors due to the edition of boundaries between accentual phases in synthesized voice.SOLUTION: The voice synthesizer comprises: a language processing unit 10 performing language processing on an original of synthesized voice and reading of the original while referring to a language dictionary, identifying a part of speech of each word included in the original, and generating an intermediate notation representing a position at which an accent and an accent combination occur in each accentual phrase into which the original is divided; an accentual-phrase-boundary-candidate extraction unit 21 extracting a boundary between an independent word out of the words and a word just before the independent word as an accentual phrase boundary candidate; and a correction-target-candidate determination unit 22 identifying a candidate for which accentual phrases before and after the candidate do not have incorrect accents as a correction target candidate to be displayed even if changing determination as to whether the boundary is an accentual phrase boundary depending on whether an accent of an accentual phrase before each accentual phrase boundary candidate and the candidate out of the accentual phrase boundary candidates are at a position of accent combination.

Description

  The present invention relates to a speech synthesizer that synthesizes a speech signal from text data, for example, a synthesized speech editing method and a synthesized speech editing computer program used in the speech synthesizer.

  In recent years, speech synthesis technology for automatically synthesizing speech has been developed. Since speech synthesis technology has the advantage that it can create desired speech in a short time, such speech synthesis technology has been adopted in applications that have used pre-recorded speech by professional narrators. There are also things. In particular, in an application in which information provided at a short time interval is updated, such as a guidance broadcast in a commercial facility, a highway radio, a highway telephone, or a weather forecast broadcast, the speech synthesis technology having the above-described advantages is useful.

  In order to generate a speech signal to be synthesized, text data mixed with kanji characters is input to the speech synthesizer via, for example, a keyboard. Then, the speech synthesizer uses a word dictionary in which kanji kana notation representing a word in kanji and kana and a phonetic character string representing pronunciation of the word are registered with respect to the text data, and morphological analysis or Performs language processing such as receiving analysis. The speech synthesizer then generates a phonetic character string of the text data and an intermediate notation in which the phonetic character string is added with a prosodic symbol representing the prosody such as accent position, accent strength, or inflection magnitude. . Then, the speech synthesizer generates a synthesized speech signal based on the intermediate notation.

  In synthesized speech used as a narration for an exhibition or e-Learning, the user may adjust the prosody of the synthesized speech in order to obtain high-quality speech that is closer to a natural utterance. Therefore, a user interface for editing the break position of an accent phrase is provided, and when a speech segment representing an accent phrase of a word string is edited, a phonetic symbol string is regenerated based on the edited speech segment. Techniques have been proposed (see, for example, Patent Document 1).

JP-A-5-11797

  In the technique disclosed in Patent Literature 1, since boundaries of all accent phrases are presented, the user determines whether the boundaries of all accent phrases are correct or not and whether or not editing has been performed. However, in order to change the boundary of the accent phrase so that the accent of the synthesized speech does not become unnatural, it is required that the user himself has knowledge about the accent. If a user who does not have enough knowledge about accents performs editing work such as removing or adding accent phrase boundaries, the correct accent position will differ depending on the presence or absence of the accent phrase boundaries. It may become a thing.

  Accordingly, an object of one aspect of the present specification is to provide a speech synthesizer capable of preventing an accent error caused by editing a boundary of an accent phrase of a synthesized speech.

  According to one embodiment, a speech synthesizer is provided. The speech synthesizer includes an input unit that obtains original text of synthesized speech and data representing reading of the original text, a storage unit that stores a word dictionary in which part-of-speech and accent positions for each word are registered, and a word dictionary To identify the part of speech of each word contained in the original text, and to divide the original text into accent phrases, and to accent and accent each accent phrase. A language processing unit that generates an intermediate notation that indicates the position where the coupling occurs, and a word that is an independent word by referring to the part of speech of each word included in the original text, and between the independent word and the word immediately before the independent word An accent phrase boundary candidate extraction unit that sets each boundary as a candidate for an accent phrase boundary, and among the accent phrase boundary candidates, an accent phrase preceding the candidate and its candidates Correction target that identifies candidates whose accent phrases before and after the candidate do not become incorrect accents as correction target candidates even if the boundary of the accent phrase is changed depending on whether or not it is in an accent-joined position A candidate determination unit; and a display control unit that displays correction target candidates on the display unit.

The objects and advantages of the invention will be realized and attained by means of the elements and combinations particularly pointed out in the appended claims.
It should be understood that both the foregoing general description and the following detailed description are exemplary and explanatory and are not restrictive of the invention as claimed.

  The speech synthesizer disclosed in this specification can prevent accent errors due to editing of accent phrase boundaries in synthesized speech.

It is a schematic block diagram of a speech synthesizer. It is a functional block diagram of the process part which a speech synthesizer has. It is a figure which shows an example of the candidate of an accent phrase boundary. It is a figure which shows an example of the relationship between the candidate of an accent phrase boundary, and a candidate for correction. It is a figure which shows an example of the display screen of a display part which displays a correction object candidate. It is an operation | movement flowchart of the synthetic | combination audio | voice editing process performed by the synthetic | combination audio | voice edit part. It is a functional block diagram of the synthetic | combination audio | voice edit part by 2nd Embodiment. It is a figure which shows an example of the difference in the accent position by the presence or absence of an accent coupling | bonding. It is a table which shows the relationship between the combination of the presence / absence of accent connection, the accent type of the prefix accent phrase and the presence / absence of change in accent position, and the edit attribute, boundary attribute and change attribute. (A) is a figure which shows an example of the display screen of the display part which displays a correction object candidate by 2nd Embodiment, (b) shows the reading and accent of the original text shown by (a). . It is an operation | movement flowchart of the synthetic | combination audio | voice editing process performed by the synthetic | combination audio | voice edit part by 2nd Embodiment.

Hereinafter, the speech synthesizer will be described with reference to the drawings.
When this speech synthesizer allows the user to edit the accent phrase boundaries of the synthesized speech, the correct accent candidate can be edited even if the accent phrase boundary is changed or not. To do. On the other hand, this speech synthesizer changes the accent phrase boundary by hiding candidates for which the accent is inappropriate among the accent phrase boundary candidates when the accent phrase boundary is changed. Prevent accent errors due to editing.

  FIG. 1 is a schematic configuration diagram of a speech synthesizer according to one embodiment. In the present embodiment, the speech synthesizer 1 includes an operation unit 2, a display unit 3, a communication interface unit 4, an output unit 5, a storage unit 6, and a processing unit 7.

  The operation unit 2 includes, for example, a keyboard and a pointing device such as a mouse. The operation unit 2 is an example of an input unit that acquires text data that is an original sentence of synthesized speech and is a kanji-kana mixed sentence and text data that represents the reading. The operation unit 2 passes the text data input by the user to the processing unit 7.

  The display unit 3 includes a display device such as a liquid crystal display. Then, the display unit 3 displays the text data of the input synthesized speech original text, editable accent phrase boundary candidates set in the original text, and the like. The operation unit 2 and the display unit 3 may be integrally formed as a touch panel display.

The communication interface unit 4 has an interface circuit for connecting the speech synthesizer 1 to a communication network, and acquires various information via the communication network. The communication interface unit 4 is another example of an input unit, which is a synthesized speech original text data that is a kanji kana mixed text and text data representing the reading thereof via a communication network. You may acquire from other apparatuses connected with.
Further, the communication interface unit 4 may output the synthesized speech signal received from the processing unit 7 to another device connected to the speech synthesizer 1 via the communication network.

  The output unit 5 outputs the synthesized voice signal received from the processing unit 7 to the speaker 8. For this purpose, the output unit 5 includes, for example, an audio interface circuit for connecting the speaker 8 to the speech synthesizer 1.

The storage unit 6 includes, for example, at least one of a semiconductor memory circuit, a magnetic storage device, and an optical storage device. And the memory | storage part 6 memorize | stores the various computer program used by the process part 7, the various data used for a speech synthesis process or a synthetic | combination speech edit process.
The storage unit 6 stores, for example, a prosodic model and a speech waveform dictionary as data used for speech synthesis processing. Furthermore, the storage unit 6 stores a word dictionary. In the word dictionary, for various words, a notation of the word, a phonetic character string, an accent unique to the word, and points indicating ease of accent combination are registered. The point indicating the ease of combining the accents of a word may be different between the case where the word is positioned before another word and the case where the word is positioned after another word. The word dictionary may further register part-of-speech information and a utilization form of each registered word.

The processing unit 7 includes one or more processors, a memory circuit, and a peripheral circuit. Then, the processing unit 7 creates a synthesized speech signal based on the input text data.
FIG. 2 is a functional block diagram of the processing unit 7. The processing unit 7 includes a language processing unit 10, a speech synthesis unit 11, and a synthesized speech editing unit 12.
Each of these units included in the processing unit 7 is, for example, a functional module realized by a computer program that runs on a processor included in the processing unit 7. Or these each part which the process part 7 has may be mounted in the speech synthesizer 1 as one integrated circuit which implement | achieves the function of each part.

  The language processing unit 10 generates a phonetic character string corresponding to the original text data that is the input kanji kana mixed sentence, and further generates an intermediate notation based on the phonetic character string. Here, the intermediate notation is obtained by adding a prosodic symbol representing a prosody to a phonetic character string. The prosodic symbols include, for example, symbols that represent “accent position”, “accent strength”, “pitch pitch”, “inflection magnitude”, “speech speed”, “volume level”, and “separation”. Therefore, the intermediate notation excluding the prosodic symbols matches the phonetic character string.

  The language processing unit 10 reads the word dictionary stored in the storage unit 6 in order to generate an intermediate notation from the input text data of the original text and text data representing the reading. For example, the language processing unit 10 uses the word dictionary to perform morphological analysis and dependency analysis on the text data, and the order and reading of each word appearing in the original sentence, the position of the accent, and the accent phrase Delimiter positions such as borders and exhalation paragraph boundaries are determined. Furthermore, the language processing unit 10 refers to the word dictionary, calculates the sum of points of ease of combining when consecutive accent phrases are combined, and the total value is equal to or greater than a predetermined threshold. , And join these accent phrases together.

The language processing unit 10 can use, for example, a method using dynamic programming as the morphological analysis. The language processing unit 10 can use a syntax analysis technique such as a prefetch LR parser or an LL method, for example, as dependency analysis. The language processing unit 10 creates an intermediate notation according to the order of each word, the reading, the position of the accent, and the position of the break including the boundary of the accent phrase.
The language processing unit 10 stores the generated intermediate notation in the storage unit 6.

  The speech synthesizer 11 creates a synthesized speech signal based on the intermediate notation of the input text data.

  The speech synthesizer 11 generates a target prosody for generating a synthesized speech signal based on the intermediate notation. For this purpose, the speech synthesis unit 11 reads a plurality of prosodic models from the storage unit 6. This prosodic model represents a position in which the voice is raised and a position in which the voice is lowered in time order. Then, the speech synthesizer 11 selects a prosodic model that most closely matches the position of the accent indicated by the intermediate notation among a plurality of prosodic models. Then, the speech synthesizer 11 creates a target prosody by setting a position where the voice becomes high or low, a position where the voice becomes low, a voice inflection, a pitch, and the like according to the selected prosodic model and synthesis parameters. . The target prosody includes a phoneme length and a pitch frequency for each phoneme as a unit for determining a speech waveform. Note that the phoneme can be, for example, one vowel or one consonant.

The speech synthesizer 11 creates a synthesized speech signal according to the generated target prosody, for example, by an HMM (Hidden Markov Model) synthesis method, a phoneme connection method, or a corpus-based method.
For example, for each phoneme, the speech synthesizer 11 selects a speech waveform closest to the phoneme length and pitch frequency of the target prosody from a plurality of speech waveforms registered in the speech waveform dictionary by pattern matching, for example. For this purpose, the speech synthesis unit 11 reads a speech waveform dictionary from the storage unit 6. The speech waveform dictionary records a plurality of speech waveforms and an identification number of each speech waveform. The voice waveform is, for example, a waveform signal extracted in units of phonemes from voice signals obtained by recording various voices in which one or more narrators read various texts.
Furthermore, the speech synthesizer 11 connects the selected speech waveform and the waveform pattern of the corresponding phoneme indicated in the target prosody so that the speech waveform selected for each phoneme can be connected along the target prosody. The deviation amount may be calculated as waveform conversion information.
The speech synthesizer 11 creates waveform generation information including the identification number of the speech waveform selected for each phoneme. The waveform generation information may further include waveform conversion information.

The speech synthesizer 11 reads from the storage unit 6 a speech waveform signal corresponding to the speech waveform identification number of each phoneme included in the waveform generation information. Then, the speech synthesizer 11 creates a synthesized speech signal by connecting each speech waveform signal continuously. When the waveform conversion information is included in the waveform generation information, the speech synthesizer 11 continuously connects the speech waveform signals by correcting each speech waveform signal according to the waveform conversion information obtained for the corresponding phoneme. By doing so, a synthesized speech signal is created.
The voice synthesizer 11 outputs the synthesized voice signal to the output unit 5.

  The synthesized speech editing unit 12 displays a boundary between words that can be an accent phrase boundary (hereinafter referred to as an accent phrase boundary candidate for convenience) even if the user corrects the accent, Part 3 is displayed. When the accent phrase boundary candidate is changed through the operation unit 2 as to whether or not it is an accent phrase boundary, the synthesized speech editing unit 12 corrects the intermediate notation according to the change. For this purpose, the synthesized speech editing unit 12 includes an accent phrase boundary candidate extraction unit 21, a correction target candidate determination unit 22, a display control unit 23, and a correction unit 24.

  When the processing unit 7 receives an operation signal indicating that intermediate notation editing is to be performed from the operation unit 2, the processing unit 7 activates the synthesized speech editing unit 12. When the synthesized speech editing unit 12 is activated, it reads from the storage unit 6 the original text data of the synthesized speech and the part-of-speech information and intermediate notation of each word corresponding to the text data.

  The accent phrase boundary candidate extraction unit 21 identifies independent words such as nouns, pronouns, or conjunctions included in the original synthesized speech based on the part of speech information of each word. Then, the accent phrase boundary candidate extraction unit 21 extracts a boundary between each independent word and the immediately preceding word as an accent phrase boundary candidate.

  FIG. 3 is a diagram illustrating an example of accent phrase boundary candidates. In FIG. 3, the synthesized speech original sentence 300 is decomposed for each word by the language processing unit 10, and the part of speech of each word is specified. Among these, as indicated by the double line 301, the words “era” (noun), “speech” (noun), “composite” (noun), “progress” (noun) and the word immediately before that The boundary is a candidate accent phrase boundary.

  The correction target candidate determination unit 22 identifies candidates for which the accent does not become inappropriate even if the user changes the accent phrase boundary among the accent phrase boundary candidates as the correction target candidates. In the present embodiment, the correction target candidate determination unit 22 sets an accent phrase boundary candidate in which the position of the accent phrase before and after the accent phrase does not change even if the user changes the accent phrase boundary as a correction target candidate. .

There are the following two conditions that do not change the position of the accent phrase before and after the boundary of the accent phrase even if it is changed.
(1) Candidates that are located in the middle of one accent phrase because of accent coupling (that is, candidates that are not accent phrase boundaries in intermediate notation)
(2) Candidates that are the boundary between two accent phrases even in intermediate notation, and the previous accent phrase is a flat accent

  Therefore, the correction target candidate determination unit 22 refers to the intermediate notation and sets a candidate for an accent phrase boundary that satisfies one of the two conditions as a correction target candidate.

  FIG. 4 is a diagram illustrating an example of a relationship between an accent phrase boundary candidate and a correction target candidate. In FIG. 4, the synthesized speech original sentence 400 is decomposed for each word by the language processing unit 10, and the part of speech of each word is specified. Further, in the original text 400, “this era”, “speech synthesis”, and “progress” are each one accent phrase 401. The accent of each accent phrase 401 is indicated by a broken line 403 representing the height of each individual sound in the notation 402 representing the reading of the original text 400. Further, the accent phrase boundary candidates 404-1 to 404-4 extracted by the accent phrase boundary candidate extraction unit 21 are indicated by double lines.

Among the accent phrase boundary candidates, the candidates 404-1 and 404-3 are located in the middle of one accent phrase, and therefore satisfy the above condition (1).
Therefore, the candidates 404-1 and 404-3 are correction target candidates.

Further, since the accent phrase “in this era” immediately before the candidate 404-2 is a flat plate, the above condition (2) is satisfied.
Therefore, the candidate 404-2 becomes a correction target candidate.

On the other hand, the candidate 404-4 is not positioned in the middle of the accent phrase, and the accent of the immediately preceding accent phrase “speech synthesis” is not a flat plate.
Therefore, the candidate 404-4 is not a correction target candidate.

  The correction target candidate determination unit 22 notifies the display control unit 23 of the position of each accent phrase boundary candidate and information indicating whether or not each accent phrase boundary candidate is a correction target candidate.

  The display control unit 23 causes the display unit 3 to display the accent phrase boundary candidate as the correction target candidate so that it can be determined whether or not the user can change the accent phrase boundary.

FIG. 5 is a diagram illustrating an example of a display screen of the display unit 3 that displays correction target candidates.
On the display screen 500, an original sentence 510 and correction target candidates 501 to 503 are displayed. Among them, the correction target candidates 501 and 503 are located in the middle of one accent phrase for accent combination, and thus are not accent phrase boundaries in the state before correction. Therefore, in this example, the correction target candidates 501 and 503 are indicated by dotted lines. On the other hand, the correction target candidate 502 is an accent phrase boundary in the intermediate notation generated by the language processing unit 10. Therefore, in this example, the correction target candidate 502 is indicated by a solid line.

  The display control unit 23 may display the accent phrase boundary candidate that is not the correction target candidate on the display unit 3 so that the user cannot change whether or not the accent phrase boundary candidate is set as the accent phrase boundary. For example, the display control unit 23 may display an accent phrase boundary candidate that is not a correction target candidate as a line having a color different from that of the line representing the correction target candidate or a line having a different luminance at a corresponding position in the original text.

  For example, the user can change whether or not the correction target candidate is an accent phrase boundary by positioning the cursor on the displayed correction target candidate and clicking via the operation unit 2. Then, a signal corresponding to the operation is output to the processing unit 7.

  The correction unit 24 adds a symbol indicating an accent phrase boundary to the position of the correction target candidate in the intermediate notation in response to an operation via the operation unit 2 having the correction target candidate as an accent phrase boundary. Conversely, the correction unit 24 determines that the correction target candidate is an accent phrase boundary at the position of the correction target candidate from the intermediate notation in response to an operation through the operation unit 2 that makes the correction target candidate not an accent phrase boundary. Remove the indicated symbol.

  In the present embodiment, the correction unit 24 does not correct the accent of the accent phrase that follows the correction target candidate even if the correction target candidate is corrected. This is because, in this embodiment, even if the correction target candidate is corrected, it is assumed that the accent position of the accent phrase that follows the correction target candidate does not result in an unnatural utterance.

  FIG. 6 is an operation flowchart of the synthesized speech editing process executed by the synthesized speech editing unit 12. The synthesized speech editing process is started, for example, when an operation for executing the synthesized speech editing process is performed via the operation unit 2 on the synthesized speech original text for which the intermediate notation has already been generated.

The accent phrase boundary candidate extraction unit 21 refers to the part of speech of each word included in the original text, and sets the boundary between the independent word and the immediately preceding word as an accent phrase boundary candidate (step S101).
The correction target candidate determination unit 22 refers to the intermediate notation, and sets, among the accent phrase boundary candidates, an accent phrase boundary candidate that satisfies one of the above conditions (1) and (2) as a correction target candidate. (Step S102).

The display control unit 23 displays the correction target candidate on the display unit 3 so that it can be seen that the correction target candidate can be edited (step S103).
The correction unit 24 corrects the notation of the accent phrase boundary at the position corresponding to the correction target candidate in which whether or not the accent phrase boundary is changed in the intermediate notation in accordance with the operation via the operation unit 2 (step S104). . Then, the correction unit 24 stores the corrected intermediate notation in the storage unit 6. Then, the synthesized speech editing unit 12 ends the synthesized speech editing process.

  As described above, this speech synthesizer presents to the user those accent phrase boundary candidates that do not become inappropriate utterances even if the accents of the preceding and following accent phrases are not changed along with the correction. . Therefore, this speech synthesizer can prevent accent errors due to editing of accent phrase boundaries of synthesized speech.

  Next, a speech synthesizer according to the second embodiment will be described. The speech synthesizer according to the second embodiment makes the display of the correction target candidate whose accent position can be changed when the accent phrase boundary is corrected differently from the display of other possible correction target candidates.

FIG. 7 is a functional block diagram of a synthesized speech editing unit included in the speech synthesizer according to the second embodiment. The synthesized speech editing unit 12 according to the second embodiment includes an accent phrase boundary candidate extraction unit 21, a correction target candidate determination unit 22, a compound word determination unit 25, an accent position change determination unit 26, and a display control unit 23. And a correction unit 24.
Compared with the synthesized speech editing unit according to the first embodiment, the synthesized speech editing unit 12 according to the second embodiment includes a compound word determination unit 25 and an accent position change determination unit 26, and a display control unit 23. The processing of is different. Therefore, hereinafter, the display control unit 23, the compound word determination unit 25, the accent position change determination unit 26, and the related parts will be described. For other components of the speech synthesizer according to the second embodiment, refer to the description of the corresponding components of the first embodiment.

  In compound words, especially compound words generated by combining multiple consecutive nouns, the accent position of the word included in the accent phrase before and after the joint position between nouns is changed from the accent position of the original noun. It may change. Further, since nouns are independent words, positions where nouns are combined, that is, boundaries between nouns, included in compound words are candidates for accent phrase boundaries. Therefore, if the accent phrase boundary candidate at the joining position of nouns included in the compound word is changed as an accent phrase boundary, there is a possibility that the utterance is natural when the accent position is changed.

Therefore, the compound word determination unit 25 determines whether or not each accent phrase boundary candidate is at a combined position of nouns in a compound word in which a plurality of nouns are continuous. For this purpose, the compound word determination unit 25 refers to the part-of-speech information of each word included in the original sentence, and confirms the part-of-speech of the words before and after each accent phrase boundary candidate. Then, the compound word determination unit 25 determines that, among the candidates for each accent phrase boundary, candidates whose preceding and following words are nouns are in the combined position of the compound words.
The compound word determination unit 25 stores, in the storage unit 6, information indicating whether or not each accent phrase boundary candidate is at the combined word combination position.
In addition, according to the modified example, editing of accent phrase boundary candidates that are not correction target candidates is not permitted in principle, so that the compound word determination unit 25 determines only compound word candidates from among accent phrase boundary candidates. You may determine whether it exists in a coupling | bonding position inside.

  The accent position change determination unit 26 refers to the compound word accent combination rule and determines whether or not the accent position changes due to accent combination for each of the accent phrase boundary candidates. Then, the accent position change determination unit 26 assigns different attributes to an accent phrase boundary candidate whose accent position changes depending on whether or not the accent phrase boundary is used, and an accent phrase boundary candidate whose accent position does not change. This attribute is called an accent position change attribute.

The compound word accent combining rule is that the accent phrase before the combining position is a flat plate, and the accent phrase after the combining position is not changed when the accent is not a flat plate. Is. Hereinafter, for convenience of explanation, an accent phrase before the joining position is referred to as a prefix accent phrase, and an accent phrase after the joining position is referred to as a post-accent phrase.
If the unique accent of the noun included in the prefix accent phrase is not a flat plate or the unique accent of the noun included in the postfix accent phrase is a flat plate according to the above accent combination rule, the accent is determined depending on the presence or absence of the accent combination. The position of changes. The unique accent is an accent when the noun is uttered alone.

  Therefore, the accent position change determination unit 26 sets the accent position change attribute to “none” for the accent phrase boundary candidate not located in the middle of the compound word if the accent of the prefix accent phrase is a flat plate. On the other hand, as for the accent phrase boundary candidate that is not located in the middle of the compound word, the accent position change determination unit 26 is not allowed to change the accent phrase boundary in principle if the accent of the prefix accent phrase is other than a flat plate. The accent position change attribute is “undefined”.

  Further, the accent position change determination unit 26 specifies the unique accent of each noun included in the compound word with respect to the accent phrase boundary candidate located in the middle of the compound word with reference to the word dictionary. The accent position change determination unit 26 sets the accent position change attribute if the unique accent of the noun included in the prefix accent phrase is a flat plate and the unique accent of the noun included in the postfix accent phrase is other than a flat plate. “None”. On the other hand, if the unique accent of the noun included in the prefix accent phrase is other than a flat plate, or if the unique accent of the noun included in the postfix accent phrase is a flat plate, the accent position change determination unit 26 changes the accent position. The attribute is “present”.

  For compound words in which the accent position has changed due to accent coupling, the accent position can be changed to match the unique accent of each noun as the accent coupling is resolved, or the accent position is maintained. But it can be a natural voice.

FIG. 8 is a diagram illustrating an example of a difference in accent position depending on the presence or absence of accent coupling. In FIG. 8, each of a circle and a triangle represents one sound. In the compound word 800 in which the accent combination is generated, as indicated by the broken line 801 representing the pitch of the sound, the accent of the prefix accent phrase 802 is a flat plate, and the accent of the suffix accent phrase 803 is a so-called head height or middle height. Become.
Here, it is assumed that the accent phrase boundary candidate 804 between the prefix accent phrase 802 and the prefix accent phrase 803 is corrected to the accent phrase boundary. In this case, as indicated by the broken line 811, the accent positions of the front accent phrase 802 and the post accent phrase 803 may be maintained as they are when the accents are combined. Alternatively, as indicated by the broken line 812, the position of the accent may be corrected according to the unique accent of the noun included in the prefix accent phrase 802 and the noun included in the prefix accent phrase 803.
Thus, whether or not to correct the accent position can be determined by the user's selection.

  Therefore, the accent position change determination unit 26 is in the position where the nouns in the compound word are combined, the accent combination is performed, and the accent phrase boundary candidate for which the accent position change attribute is determined to be “present” is used. The accent position change attribute may be “selected”. For accent phrase boundary candidates whose accent position change attribute is “Select”, when the accent phrase boundary is changed, whether the accent position of the preceding or following accent phrase is changed or maintained is determined by the user's selection Is done.

  The accent position change determination unit 26 stores the accent position change attribute for each accent phrase boundary candidate in the storage unit 6.

  Note that, according to the modification, the accent position change determination unit 26 determines whether or not the accent position changes depending on the accent phrase boundary only for the correction target candidate, and sets the accent position change attribute. Good. This is because, as a rule, accent phrase boundary candidates that are not correction target candidates are not changed by the user as to whether or not they are accent phrase boundaries.

  The display control unit 23 changes the display of each correction target candidate according to the accent position change attribute of the correction target candidate.

  FIG. 9 is a table showing the relationship between the presence / absence of accent coupling, the combination of the accent type of the prefix accent phrase and the presence / absence of change in accent position, and the edit attribute, boundary attribute, and accent position change attribute. The edit attribute represents whether or not the candidate is a correction target candidate. The boundary attribute represents whether or not an accent phrase boundary is set in the intermediate notation obtained as a result of language processing.

  In the table 900, one category is shown in one row, and the combination of presence / absence of accent coupling, accent type of the prefix accent phrase, and presence / absence of accent position change is defined for each category. The symbol “-” indicates that the item indicated by the symbol is not referred to. Categories 1 to 5 correspond to the case where the accent phrase boundary candidate is located in the middle of the compound word, and categories 6 and 7 correspond to the case where the position of the accent phrase boundary candidate is not in the middle of the compound word.

  As shown in category 1, when the accent phrase boundary candidate is at the accent coupling position, the accent phrase boundary candidate is not set as the accent phrase boundary. In addition, accent phrase boundary candidates are editable (that is, correction target candidates). In Category 1, since the accent position does not change due to the accent connection, the accent position change attribute is “none”. In this case, from the accent combination rule, the unique accent of the prefix accent phrase is always a flat plate.

  As shown in Category 2, even if the accent phrase boundary candidate is not the accent joint position and the accent position does not change due to the accent joint, the unique accent of the prefix accent phrase is always a flat plate from the accent joint rule. is there. Therefore, the accent phrase boundary candidate is editable, and the accent position change attribute is “none”. Further, in this case, since the accent phrase boundary candidate is not the accent coupling position, the accent phrase boundary candidate is the accent phrase boundary.

  Further, as shown in category 3, when the accent phrase boundary candidate is at the accent joint position and the accent position changes due to the accent joint, the accent phrase boundary candidate is editable, and the accent position change attribute is “Select”. Further, in this case, the accent phrase boundary candidate is located in the middle of the accent phrase, and thus is not set to the accent phrase boundary.

  Further, as shown in category 4, the accent phrase boundary candidate can be edited even if the accent phrase boundary candidate is not the accent coupling position and the unique accent of the prefix accent phrase is a flat plate. Since the accent position is changed by the accent combination, the accent position change attribute is “selection”. Further, in this case, since the accent phrase boundary candidate is not the accent coupling position, the accent phrase boundary candidate is the accent phrase boundary.

  In addition, as shown in category 5, if the accent phrase boundary candidate is not the accent combining position and the unique accent of the prefix accent phrase is other than a flat plate, the accent phrase boundary candidate is not editable in principle. (That is, it is not a candidate for correction). However, a candidate for an accent phrase boundary belonging to this category is at a joint position between nouns in the joint word. Therefore, as an exception, accent phrase boundary candidates can be edited. In this category, at least the accent of the prefix accent phrase changes depending on the presence or absence of accent coupling. Therefore, the accent position change attribute is “present”. By setting the accent phrase boundary candidate as the accent phrase boundary, it is necessary to change at least the position of the accent of the prefix accent phrase. Therefore, in this case, it is preferable that the accent position change determination unit 26 does not set the accent position change attribute to “selection”.

  Furthermore, as shown in category 6, if the position of the accent phrase boundary candidate is not in the middle of a compound word, the accent position does not change depending on the presence or absence of accent coupling. Therefore, the accent position change attribute is “none”. Moreover, since the unique accent of the prefix accent phrase is a flat plate, the accent phrase boundary candidates can be edited. In this case, since the accent phrase boundary candidate is not the accent coupling position, the accent phrase boundary candidate is the accent phrase boundary.

  Finally, as shown in category 7, the position of the accent phrase boundary candidate is not in the middle of the compound word, the accent phrase boundary candidate is not the position of the accent combination, and the unique accent of the prefix accent phrase is flat. Otherwise, accent phrase boundary candidates are not editable. Therefore, the accent position change attribute is “undefined”. Also in this case, since the accent phrase boundary candidate is not the position of the accent connection, the accent phrase boundary candidate is the accent phrase boundary.

FIG. 10A is a diagram illustrating an example of a display screen of the display unit 3 that displays correction target candidates according to the second embodiment.
The display screen 1000 displays the original text 1010 and the correction target candidates 1001 to 1003. Among these, the correction target candidates 1001 and 1003 are located in the middle of one accent phrase for accent combination, and thus are not accent phrase boundaries in the state before correction. Therefore, in this example, the correction target candidates 1001 and 1003 are indicated by dotted lines. On the other hand, the correction target candidate 1002 is an accent phrase boundary in the state before correction. Therefore, in this example, the correction target candidate 1002 is indicated by a solid line.

  Further, the correction target candidate 1003 is at the combined position of the compound word “speech synthesis”, and the accent position is changed by the accent combination. Therefore, with respect to the correction target candidate 1003, when changing to an accent phrase boundary and canceling accent coupling, it is possible to select whether to maintain the accent position at the time of accent coupling or to change to the unique accent of each noun. Yes. Therefore, the correction target candidate 1003 is displayed differently from the correction target candidate 1001 in order to indicate that it is possible to select whether or not the accent position has been changed. In this example, the correction target candidate 1001 is indicated by a single line, and the correction target candidate 1003 is indicated by a double line.

Also in this embodiment, the user can change whether or not the correction target candidate is an accent phrase boundary by, for example, placing the cursor on the displayed correction target candidate and clicking via the operation unit 2. For correction target candidates that can be selected to change the accent position, for example, each time the operation unit 2 is clicked, the accent position is maintained, the accent position is changed, and the accent phrase boundary is not changed. Then, the operation unit 2 outputs a signal corresponding to the operation to the processing unit 7.
When the correction unit 24 changes whether or not the correction target candidate is an accent phrase boundary, when the accent position is also changed in the preceding and following accent phrases, the correcting unit 24 instructs the language processing unit 10 to display the accent phrases before and after the accent phrase boundary. Enter the to correct the accent position.

  FIG. 10 (b) shows the reading and accent of the original text shown in FIG. 10 (a). A polygonal line 1101 displayed superimposed on the original reading “Konoji Dino On Sego Seino Sympowa” written in katakana represents an accent when the accent phrase boundary is not corrected. A broken line 1102 represents the accent when the correction target candidates 1001 and 1002 are changed as to whether or not the boundary is an accent phrase boundary. A broken line 1103 represents an accent when the correction target candidate 1003 is changed to an accent phrase boundary while maintaining the accent position. On the other hand, the broken line 1104 represents the accent when the correction target candidate 1003 is changed to an accent phrase boundary while the accent positions of the nouns before and after the correction target candidate 1003 are changed to positions corresponding to the unique accents.

  FIG. 11 is an operation flowchart of the synthesized speech editing process executed by the synthesized speech editing unit 12 according to the second embodiment.

The accent phrase boundary candidate extraction unit 21 refers to the part of speech of each word included in the original text, and sets the boundary between the independent word and the immediately preceding word as an accent phrase boundary candidate (step S201).
The correction target candidate determination unit 22 refers to the intermediate notation, and sets, among the accent phrase boundary candidates, an accent phrase boundary candidate that satisfies one of the above conditions (1) and (2) as a correction target candidate. (Step S202).

  The compound word determination unit 25 determines, for each accent phrase boundary candidate, whether or not a plurality of nouns are in the combined position of the nouns in the continuous compound word, and identifies the candidate accent phrase boundary at the combined position. (Step S203). The accent position change determination unit 26 determines whether or not the accent position changes when the accent phrase boundary candidate at the combined position is changed according to the accent combination rule of the compound word. According to the result, the accent position change determination unit 26 sets an accent position change attribute to each accent phrase boundary candidate (step S204).

The display control unit 23 displays the correction target candidate on the display unit 3 so that it can be seen that the correction target candidate can be edited. At that time, the display control unit 23 displays a display in which the correction target candidate whose accent position changes by changing whether or not the accent phrase boundary is different from the correction target candidate and the correction target candidate whose accent position does not change. Each correction target candidate is displayed so that it becomes (step S205).
The correction unit 24 corrects the notation of the accent phrase boundary at the position corresponding to the changed correction target candidate in the intermediate notation in accordance with the operation via the operation unit 2 (step S206). Then, the correction unit 24 stores the corrected intermediate notation in the storage unit 6. Then, the synthesized speech editing unit 12 ends the synthesized speech editing process.

  As described above, the speech synthesizer according to the second embodiment determines a correction target candidate whose accent position changes depending on whether or not it is an accent phrase boundary at a combined position in a compound word. It is displayed on the display unit separately from the correction target candidates whose positions do not change. The speech synthesizer also determines whether to maintain or change the accent position when a correction target candidate whose accent position changes depending on whether the accent phrase boundary is used or not is changed according to a user operation. Can be selected. Therefore, this speech synthesizer can obtain a more natural synthesized speech by changing the accent phrase boundary.

  Furthermore, a computer program that causes a computer to realize each function of the processing unit of the speech synthesizer according to each of the above embodiments is recorded on a computer-readable medium, for example, a magnetic recording medium, an optical recording medium, or a semiconductor memory. It may be provided in the form.

  All examples and specific terms listed herein are intended for instructional purposes to help the reader understand the concepts contributed by the inventor to the present invention and the promotion of the technology. It should be construed that it is not limited to the construction of any example herein, such specific examples and conditions, with respect to showing the superiority and inferiority of the present invention. Although embodiments of the present invention have been described in detail, it should be understood that various changes, substitutions and modifications can be made thereto without departing from the spirit and scope of the present invention.

DESCRIPTION OF SYMBOLS 1 Speech synthesizer 2 Operation part 3 Display part 4 Communication interface part 5 Output part 6 Storage part 7 Processing part 8 Speaker 10 Language processing part 11 Speech synthesizer 12 Synthetic speech editing part 21 Accent phrase boundary candidate extraction part 22 Correction object candidate determination 23 Display control unit 24 Correction unit 25 Compound word determination unit 26 Accent position change determination unit

Claims (10)

An input unit that obtains the original text of the synthesized speech and data representing the reading of the original text;
A storage unit for storing a word dictionary in which part-of-speech and accent positions for each word are registered;
By performing linguistic processing for reading the original text and the original text with reference to the word dictionary, the part of speech of each word included in the original text is specified, and the original text is divided in units of accent phrases, A language processing unit for generating an intermediate notation representing a position where the accent of the phrase and the accent coupling occur;
An accent phrase that identifies a word that is an independent word with reference to the part of speech of each word included in the original text, and that uses a boundary between the independent word and the word immediately before the independent word as a candidate for an accent phrase boundary, respectively A boundary candidate extraction unit;
Of the accent phrase boundary candidates, depending on whether the accent phrase precedes the candidate and whether the candidate is at the position where the accent is combined, whether or not the accent phrase boundary, A correction target candidate determining unit that identifies candidates whose accent phrases before and after the candidate do not become false accents as correction target candidates;
A display control unit for displaying the correction target candidates on a display unit;
A speech synthesizer.   The correction target candidate determination unit selects a candidate whose accent phrase position before and after the candidate does not change even if the accent phrase boundary is changed among the accent phrase boundary candidates. The speech synthesizer according to claim 1, which is a candidate.   The speech synthesis device according to claim 2, wherein the correction target candidate determination unit sets, as the correction target candidate, a candidate at an accent-combined position among the accent phrase boundary candidates.   The correction target candidate determination unit sets a candidate in the accent phrase boundary candidate at a position where no accents are combined and the accent phrase immediately before the candidate is a flat plate as the correction target candidate. The speech synthesizer according to claim 2 or 3.   5. The apparatus according to claim 1, further comprising a correction unit that corrects whether or not the selected correction target candidate is a boundary of an accent phrase in the intermediate notation in accordance with an operation through an operation unit. The speech synthesizer described. For each of the correction target candidates, a compound word determination unit that determines whether or not a plurality of nouns are in the combined position of nouns in a compound word that is continuous,
Based on the compound word accent combining rule, if the boundary of the accent phrase among the correction target candidates at the combined position is changed, the position of at least one of the accent phrases before and after the correction target candidate The first attribute indicating that the accent position changes is assigned to the correction target candidate that changes, and on the other hand, even if the boundary of the accent phrase is changed, the accent phrase of the accent phrase before and after the correction target candidate is changed. An accent position change determination unit that assigns a second attribute indicating that the accent position does not change to a correction target candidate whose position does not change,
The display control unit makes the display on the display unit of the correction target candidate having the first attribute different from the display on the display unit of the correction target candidate having the second attribute. 5. The speech synthesizer according to any one of 4 above. The combining rule is that the noun accent immediately before the combining position is a flat plate, and the noun accent immediately after the combining position is not a flat plate when the compound word is not changed.
The accent position change determination unit is configured such that an accent of a noun included in an accent phrase immediately before the correction target candidate at the combined position is other than a flat plate, or an accent immediately after the correction target candidate at the combined position The speech synthesizer according to claim 5, wherein when the noun accent included in the phrase is a flat plate, the first attribute is assigned to the correction target candidate.   When the correction target candidate having the first attribute selected in the intermediate notation data is corrected to be a boundary of an accent phrase in response to an operation through the operation unit, the first attribute The speech synthesizer according to claim 6, further comprising a correcting unit that determines whether or not to correct the positions of the accent phrases before and after the correction target candidate having the above. Referring to the word dictionary in which the part of speech and accent position for each word are registered, by performing linguistic processing on the synthesized speech and the reading of the original, the part of speech of each word included in the original is identified, And, the original text is divided in units of accent phrases to generate an intermediate notation representing the position of each accent phrase and the position where the accent combination occurs,
Identify words that are independent words with reference to the part of speech of each word included in the original text, and each boundary between the independent word and the word immediately before the independent word is a candidate for an accent phrase boundary,
Of the accent phrase boundary candidates, depending on whether the accent phrase precedes the candidate and whether the candidate is at the position where the accent is combined, whether or not the accent phrase boundary, Identify candidates whose accent phrases before and after the candidate are not incorrect accents,
Displaying the candidate for correction on a display unit;
A synthesized speech editing method. Referring to the word dictionary in which the part of speech and accent position for each word are registered, by performing linguistic processing on the synthesized speech and the reading of the original, the part of speech of each word included in the original is identified, And, the original text is divided in units of accent phrases to generate an intermediate notation representing the position of each accent phrase and the position where the accent combination occurs,
Identify words that are independent words with reference to the part of speech of each word included in the original text, and each boundary between the independent word and the word immediately before the independent word is a candidate for an accent phrase boundary,
Of the accent phrase boundary candidates, depending on whether the accent phrase precedes the candidate and whether the candidate is at the position where the accent is combined, whether or not the accent phrase boundary, Identify candidates whose accent phrases before and after the candidate are not incorrect accents,
Displaying the candidate for correction on a display unit;
A computer program for synthetic speech editing for causing a computer to execute this.

Images