JP2022141520A - Voice synthesis symbol editing device, method and program - Google Patents

Abstract

To provide a voice synthesis symbol editing device that can be used by reducing impact of a TTS system configuration and modifications thereof.SOLUTION: There is provided a voice synthesis symbol editing device 10 that accepts input of a pair of text and a sequence of voice synthesis symbols corresponding to the text, and outputs the sequence of voice synthesis symbols, which includes: a correspondence relation estimation unit 11 that divides the input text into predetermined character units as well as the input voice synthesis symbols into predetermined voice synthesis symbol tokens, and gives a correspondence relation between each of the divided characters and each of the zero or more voice synthesis symbol tokens, based on a predetermined criteria; and an editing interface section 123 that displays each character and each of the zero or more phonetic synthesis symbol tokens given the correspondence relation, while updating the phonetic synthesis symbol tokens by accepting user editing operations.SELECTED DRAWING: Figure 1

Description

TECHNICAL FIELD The present invention relates to a speech synthesis symbol editing apparatus, method, and program that allow a user to operate a speech synthesis symbol string by inputting a text and a speech synthesis symbol string.

Speech synthesis technology is a technique for artificially synthesizing speech. A typical usage is text-to-speech (hereafter referred to as TTS). Direct mapping of speech features to be synthesized is difficult due to the extremely complex relationship structure. Therefore, using an abstracted intermediate representation, through a two-step conversion from the text to the intermediate representation and from the intermediate representation to the speech features, we generated or prepared speech waveforms that match the speech feature information using signal processing. A synthesized speech waveform can be obtained by selecting an appropriate one from the accumulation of waveforms.

The intermediate representation is, for example, a command symbol string for speech synthesis (hereinafter referred to as a speech synthesis symbol string). symbols for text-to-speech synthesis"). IT-4006 defines two types of notation, "kana level notation" and "allophone level notation". It is not a definition, but a more abstract definition such as the types of sounds that make up speech.The same applies below.) and a combination of descriptions related to prosody control.

On the other hand, the physical features of speech are, for example, time-series data of parameters representing the acoustic features of speech, and are specifically expressed as, for example, cepstrum coefficients at 1 ms intervals and fundamental frequency values.

By the way, most reading errors in a text-to-speech synthesis system occur in conversion processing (hereinafter referred to as text analysis) from text (a sentence containing kanji and kana) to a speech synthesis symbol string. Therefore, by further correcting the synthesized speech symbol string automatically converted from the text by the creator of the synthesized speech, synthesized speech without reading errors can be created.

Japanese Unexamined Patent Publication No. 2013-134396 "Synthetic speech correction device, method, and program"

JEITA standard IT-4006 "Symbols for Japanese text-to-speech synthesis"

Definitions of many speech synthesis symbol strings are often designed based on phonetic knowledge. For example, even in a format such as JEITA IT-4006 mentioned above, which is relatively easy to describe, prior knowledge of the types of phonemes, accents related to the Tokyo dialect, and methods of describing intonation is necessary to understand the format. In the kana-level notation specified in JEITA IT-4006, phoneme symbols (defined as "pronunciation symbols" in the standard) are written in katakana (however, di, z, and w are written as di, z, and o, respectively). ``/'' ``|'' is unfamiliar to many users from the very concept of prosody.

Therefore, in order to improve convenience for general users, an editing apparatus for a speech synthesis system as shown in Patent Document 1 has been devised. However, the method shown here is equivalent to the process of text analysis processing based on morphological analysis processing, which is also used in ordinary TTS systems, and includes correction processing by the user. The string editing system includes the text analysis process of the TTS system.

By the way, if a user who has no knowledge or experience with speech synthesis symbol strings edits speech synthesis symbols for speech synthesis, the system will consider each character of the text, or each morpheme, and the symbolic elements that make up the speech synthesis symbol. (hereinafter referred to as speech synthesis symbol tokens).

In addition, since it is not easy for users with no knowledge or experience to read synthesized speech symbol strings, it is considered that the decision as to whether or not synthetic symbol string editing is necessary is made mainly based on listening to synthesized speech. At this time, if the reading by the target TTS system can be accurately reproduced on the editing system, it is convenient to use the editing device as a system for checking synthesized speech. For example, in an information distribution system that uses speech synthesis to send additional information, such as a speech synthesis symbol string, only for sentences that are misread by automatic processing, for data compression, etc., whether or not there is a need to send additional information depends on the editing device. can be determined by Therefore, it is desirable that the editing system can also present the result of automatic processing in the target TTS system and reproduce the synthesized speech.

However, in conversion from text to a speech synthesis symbol string, for example, an end-to-end estimation method using deep learning, etc., the correspondence relationship between the characters that make up the text and each symbol that makes up the output speech synthesis symbol string There may be configurations of TTS systems that use conversion processes in which is not directly available. When a target TTS system uses such a conversion technique, conventional speech synthesis symbol string editing devices for such a TTS system have conventionally (1) used text characters or morphemes in the system and corresponding (2) use a different text analysis process in the editing system than that used in the TTS system, such that the correspondence between the two can be obtained; There was a problem that one of them must be selected.

In addition, since there are relatively many text analysis errors as described above, the actual TTS system frequently changes the internal processing of the text analysis process. Considering the convenience of editors, it is desirable that the text analysis results of the actual TTS system can always be accurately reproduced on the editing device. I had a problem with the need.

SUMMARY OF THE INVENTION In view of the problems of the prior art described above, it is an object of the present invention to provide a text-to-speech symbol editing apparatus, method, and program that can be used while reducing the effects of the configuration and modification of a TTS system.

To achieve the above object, the present invention provides a text-to-speech symbol editing device for inputting a text and a set of text-to-speech synthesis symbol strings corresponding to the text, and for outputting a speech synthesis symbol string, comprising: is divided into predetermined character units, and the input speech synthesis symbol is divided into predetermined speech synthesis symbol tokens. and an editing unit that displays each character and zero or more phoneme synthesis symbol tokens to which the correspondence relationship is given while updating the phoneme synthesis symbol tokens by accepting user editing operations. and an interface unit. Further, the present invention is characterized by a method and a program corresponding to the speech synthesis symbol editing device.

According to the present invention, after giving the correspondence between each character in the text and each of zero or more speech synthesis symbol tokens, the correspondence is displayed to the user and the user's editing operation is accepted. , it becomes possible to use the speech synthesis symbol editing device while reducing the influence of the structure of the TTS system and its modification.

1 is a functional block diagram of a speech synthesis symbol editing device according to an embodiment; FIG. 1 is a functional block diagram of a synthesized speech editing device according to one embodiment; FIG. It is a figure which shows the example of a process in a correspondence estimation part. It is a figure which shows the hardware constitutions in a common computer.

FIG. 1 is a functional block diagram of a speech synthesis symbol editing device according to one embodiment. The speech synthesis symbol editing device 10 includes a correspondence estimation unit 11 , an editing result display unit 12 , a speech synthesis symbol editing unit 13 , a speech synthesis symbol correction unit 14 , an editing record database 15 and an editing reproduction unit 16 . FIG. 2 is a functional block diagram of a synthetic speech editing device 20 according to one embodiment. Equipped with 22.

As shown in FIG. 2, the speech synthesis symbol editing device 10 converts the speech synthesis symbol string (original is output so that the user can refer to it together with the Japanese text of the text), and provides an interface that accepts an operation for editing by the user for correction. The speech synthesis symbol string edited by the user's operation is further output to the speech synthesis unit 22 and reproduced as synthesized speech, so that the user can listen to and confirm the synthesized speech based on the edited speech synthesis symbol string. be. The user may sequentially edit the speech synthesis symbol string with the speech synthesis symbol editing device 10 while listening to and confirming the synthesized speech. (On the other hand, a skilled user may be able to omit the work of listening to and confirming the synthesized speech by the speech synthesizing unit 22 by grasping the state of the synthesized speech only from the speech synthesis symbol string.)

First, an outline of processing of each unit of the speech synthesis symbol editing device 10 will be described below.

The speech synthesis symbol editing apparatus 10 can accept editing work by the user in the following flow (1) to (4) and output a speech synthesis symbol string edited by the user.

(1) The speech synthesis symbol editing device 10 converts a sentence mixed with kanji and kana, which is a text, and a speech synthesis symbol string, which is the text analysis result (result obtained by analyzing the text in the text analysis unit 21 in FIG. 2). It is received by the correspondence estimation unit 11 as an input.

(2) The correspondence estimation unit 11 divides the speech synthesis symbol string into strings of speech synthesis symbol tokens, associates zero or more speech synthesis symbol tokens with each character of the text, and uses them as correspondence information. Output to the editing result display unit 12 as indicated by line L3. (Alternatively, as an alternative example to be described later, the output of the correspondence estimation unit 11 may be output to the editing result display unit 12 via the editing reproduction unit 16, as indicated by line L4.)

(3) The editing result display unit 12 presents the correspondence, which is the estimation result of the correspondence estimation unit 11, to the user. The editing result display unit 12 is implemented by a display as hardware, and can present to the user by displaying each speech synthesis symbol token and each text in the corresponding relationship together with the corresponding relationship. The editing result display unit 12 also displays an instruction input from the user who has confirmed the displayed estimation result (an instruction input for correcting a part of the speech synthesis symbol token that needs to be corrected and is the next speech synthesis symbol editing unit 13), the speech synthesis symbol tokens are concatenated to generate a speech synthesis symbol string (corrected by user editing).

The text-to-speech symbol string displayed and generated (held) by the editing result display unit 12 is shared with the text-to-speech symbol editing unit 13 during editing by the user operation, as indicated by the double-headed arrow L0. When the user determines that the editing has been completed (or determines that the user wants to obtain an intermediate confirmation result) and performs an input to confirm the editing, as shown by the line L1, the edited speech synthesis is confirmed. It is output from the speech synthesis symbol editing device 10 as a symbol string. (Alternatively, instead of directly outputting as shown by line L1, as shown by line L2, as an alternative example described later, automatic correction processing by the speech synthesis symbol correction unit 14 may be further applied and then output. good.)

That is, the editing result display unit 12 serves as a display interface to display to the user the correspondence between each character of the text being edited and zero or more speech synthesis symbol tokens. Next, when the speech synthesis symbol token is corrected by the user's operation in the speech synthesis symbol editing unit 13, the result of the correction is successively reflected and displayed to the user via the editing result display unit 13. FIG. In this way, both the editing result display unit 12 and the speech synthesis symbol editing unit 13 form an editing interface unit 123 for editing the speech synthesis symbols for realizing a display function reflecting the sequential editing results. of editing buffers to the user, and such editing buffers can be provided on a sentence-by-sentence basis, or on any basis the user wishes to edit.

(4) The text-to-speech symbol editing unit 13 replaces or deletes the text-to-speech symbol token associated with a predetermined character based on the judgment of the user who visually confirms the correspondence relationship, according to the determined user operation, Alternatively, a speech synthesis symbol token linked to the character is added, and the result is sent back to the editing result display unit 12 (as indicated by the two-way arrow L0, the editing result display unit 12 and the speech synthesis symbol (shared by the editing unit 13), the correction of the correspondence relationship by the user's operation is sequentially accepted and displayed to the user.

(5) The speech synthesis symbol correction section 14 is used when outputting the speech synthesis symbol string as indicated by the line L2 instead of directly outputting the speech synthesis symbol string as indicated by the line L1. is rewritten based on a predetermined rule so as to meet the required specifications of the speech synthesis symbol string and output.

The flow of (1) to (5) above was based on the assumption that the user would edit everything each time. , and as a configuration for reducing the user's editing work load, the processing of the flow of (6) to (7) may be performed by the following mechanism. As an additional configuration for this purpose, the text-to-speech symbol editing apparatus 10 may include an edit record database 15 and an edit reproduction unit 16. FIG. (If this additional configuration is applied, the output from the correspondence unit 11 is processed according to the flow indicated by line L4, and if not applied, the output is processed according to the flow indicated by line L3.)

(6) The editing record database 15 performs editing (replacement, deletion, addition) operations on speech synthesis symbol tokens linked to characters of the text by user operations in the phonetic symbol editing unit 13. Store in the edit record database 15 in the form of a set of tokens. At this time, one or more characters before and after the text and the token speech synthesis symbol before editing may be stored together.

(7) The edit reproduction unit 16 refers to the edit record database 15 for the string of characters that make up the text and the speech synthesis symbol tokens associated with each character, which are output from the correspondence estimation unit 11. , performs an editing operation on the speech synthesis symbol token associated with each character according to a predetermined standard. The user may be allowed to select whether or not to allow the edit reflecting section 16 to function.

Details of each unit of the speech synthesis symbol editing device 10 will be further described below.

<Correspondence estimation unit 11>
The processing in the correspondence estimation unit 11 is, for example, as follows. As exemplified in FIG. 3, for the text "It's nice weather today." become that way. The input of the speech synthesis symbol editing device 10 is a pair of both, which becomes the input of the correspondence estimation unit 11 as it is.

The correspondence estimator divides the text into characters and the speech synthesis symbols into tokens. However, hiragana and katakana are divided into mora units, and a character string of one mora sentence is regarded as one character. For example, "kya" is regarded as one character.

In the case of the example in Figure 3, as shown in the figure as well, the units enclosed by squares (□) are divided into ``now'', ``day'', ``ha'', ``good'', ``i'', ``heaven'', ``ki'', and `` is not it""."
``kyo'''``ー''``wa''``|''``i'''``ー''``/''``te'''``n''``ki''``de''``su''``ne''``.''
is divided into (Note that "i'" and "-" are the result of further application of the processing described later.)

Note that the accent core symbol "'" in IT-4006 is not regarded as an independent token here, but is considered as one token together with the preceding reading symbol.

Then, for each character of the text, zero or more speech synthesis symbol tokens are associated. The correspondence between the kana characters in the text and the reading symbols in the text-to-speech symbol tokens is clear, and kanji characters usually have some reading, and prosody boundary marks may be associated with punctuation marks and parentheses. By applying the empirical rule (rule base) of
``Now'' = ``Kyo'''
"day" = "-"
"Ha" = "Wa"
"" = "|"
``Good'' = ``Yo'''
"I" = "I"
"" = "/"
"Heaven" = "Te'""N"
"ki" = "ki"
``de'' = ``de''
"su" = "su"
"ne" = "ne"
"." = "."
A corresponding relationship is obtained. (In FIG. 3, the unit of the speech synthesis symbol string corresponding to the character unit of the text is surrounded by squares (□) and arranged vertically. It can be displayed graphically and shown to the user.)

In this example, since there is no text character corresponding to the prosodic boundary symbols "|" and "/" in the speech synthesis symbol string, the correspondence is described by inserting an empty character "" into the text characters.

The process of automatically obtaining this correspondence relationship may be performed by, for example, defining costs as shown below and using a method based on the minimum cost method. Such processing can be easily realized using existing optimization methods such as dynamic programming.

The hiragana/katakana characters match the reading symbol katakana: cost 0
Hiragana "wa" is read as "wa": cost 1
Hiragana ``he'' is read as ``e'': cost 1
Hiragana "A" ~ "O" with reading "-" (long vowel symbol): Cost 1
A terminal symbol "." corresponds to the period "." of the text: cost 1
Hiragana/katakana characters do not satisfy the above four rules: Cost 100
Any reading symbol except ``n'' of 1 mora corresponds to 1 kanji character: cost 20
The reading symbol "n" corresponds to one kanji character: cost 70
Any reading symbol that does not start with "N" of N mora (N≧2) corresponds to one kanji character: Cost 10 × (N-1)
Corresponds to reading symbols starting with "N" of N mora (N≧2) for one kanji character: Cost 50 + 10 × (N - 1)
There is no reading symbol corresponding to one kanji character: cost 100 (that is, even if there is no possible reading for kanji, it is allowed.)

In addition, for readings of kanji, a single kanji dictionary (dictionary in which the reading is assigned to each kanji character) is prepared in advance, and the cost is defined so that the cost is small when the reading matches the reading. By doing so, the correspondence between characters and speech synthesis symbol tokens can be obtained with higher accuracy. For example, in the above example, since the pronunciation symbol tokens are evenly assigned to the locations where the kanji are lined up, errors tend to occur in locations consisting of long kanji character strings, but using the single kanji dictionary Such errors can be reduced by more accurately handling the correspondence between kanji and reading symbols.

<Speech Synthesis Symbol Editor 13>
The text-to-speech symbol editing unit 13 corrects, deletes, and adds tokens corresponding to characters in accordance with user input. At this time, if an operation to insert a prosody mark is performed, and if there is no corresponding character, including an empty character "", the rule-based may be automatically performed by In addition, if there is no speech synthesis symbol token corresponding to the empty character "" in the speech synthesis symbol token due to deletion, the process of deleting the empty character "" from the character string is automatically performed by the rule base without user operation. You can do it with

<Speech synthesis symbol correction section 14>
The speech synthesis symbol correction unit 14 is a mechanism for preventing the speech synthesis symbol editing device 10 from outputting a formally incorrect speech synthesis symbol string. (An invalid speech synthesis symbol string cannot be speech-synthesized by the speech synthesis unit 22 in FIG. 2.)

In the edit interface unit 123, while the user edit operation is sequentially reflected, among the speech synthesis symbol token strings held (as an edit buffer) in association with the corresponding relationship with each character, the fact that the user has confirmed is displayed. The partial string (or the entire editing buffer) input to the edit result display unit 12 may be output to the speech synthesis symbol correction unit 14 . The user listens to and confirms the synthesized speech corresponding to the subsequence by outputting only the portion of this subsequence as synthesized speech in, for example, the speech synthesizing unit 22 in FIG. Such work is possible. In addition to the case where the user has determined that the user's editing operation has been completely completed, this confirmation input is made when the user wishes to confirm by listening to the synthesized speech through an intermediate confirmation ( (to continue editing if the synthesized speech is unnatural).

For example, if the user deletes the "/" in "Kyo'wa | Yo'i/Te'nkidesune." However, within an accent phrase (a section delimited by either an accent phrase boundary ``/'', a phrase boundary ``|'', or a pause ``_''), there is an accent core It is a form that does not satisfy the JEITA IT-4006 rule that there must be at most one "'".

On the other hand, the speech synthesis symbol correction unit 14 rewrites the speech synthesis symbol string so as to have an appropriate form in terms of format based on a predetermined rule. For example, if a rule is set such that the first accent kernel within an accent phrase is left and the other accent kernel symbols are deleted, the speech synthesis symbol modification unit 14 has the proper format, ``Kyo'wa|Yo''. Itenkidesune." is output as a speech synthesis symbol string. (That is, the accent core of "te'" is deleted and the accent core "'" in the accent phrase is rewritten to at most one and output.) Thus, the output result of the speech synthesis symbol editing device 10 is converted into speech synthesis. The symbol string can be directly input to a speech synthesizer (speech synthesizer 22 in FIG. 2). For example, the user can confirm the display contents of the editing result display 12 by synthesized speech.

Here, the processing of this speech synthesis symbol correction unit 14 (as an alternative example of line L1) is the processing immediately before output as indicated by line L2, so that the original information other than the edited part is transferred to the system ( (as shared data in the editing result display unit 12 and the speech synthesis symbol editing unit 13 indicated by the line L0), for example, after that, the speech synthesis symbol editing unit 13 reinserts "/" at the original position In this case, the original speech synthesis symbols are reproduced, in which the reading symbols for ``天'' are ``te''' and ``n'' (te has an accent core).

<Speech synthesis symbol editing unit 13, editing record database 15, editing reproduction unit 16>
The speech synthesis symbol editing unit 13 may record the editing operation in the editing record database 15 in a predetermined format. For example, when the reading symbol for ``good'' is rewritten from ``yo''' to ``i'', the prescribed format is, for example, ``rewrite the reading symbol ``yo''' of the character ``ryo'' to ``i''''. Record the action in the editorial database. Here, as another example of a predetermined format, ``Rewrite the reading symbol ``Yo'' of the character ``Yo'' with the preceding character '' and the following character ``I'' as ``I''''. , a form that describes more detailed conditions regarding characters, or a more simplified form such as ``Rewrite the reading symbol of the character ``good'' to ``Yo'''' may also be used.

Also, if there is an edit record database 15, an edit reproduction unit 16 can be provided. The editing reproduction unit 16 refers to the character and the speech synthesis symbol token for the output of the correspondence estimation unit 11 for the new input to the speech synthesis symbol editing device 10, queries the edit record database, and responds to the query. If the edit information is in the edit record database 15 , this edit operation is performed on the input of the edit reproduction unit 16 and output to the edit result display unit 12 .

That is, whether the rewrite rules for each of the editing information recorded in the editing record database 15 are applicable to the string of characters and speech synthesis symbol tokens to which the correspondence is given, which is obtained by the correspondence estimation unit 11. are collated, and if there is an applicable one, it is applied and then output to the editing result display unit 12.

If the previous example were in the edit record database, the input "Today is a good day." ” can be output to the text-to-speech symbol editing unit 13, thereby reducing the user's time and effort for similar correction operations.

Note that the user may be allowed to select whether or not to apply this process. That is, when there is a place to which the rewrite rule of the editing information is applicable, it is not immediately applied and output to the edit result display unit 12, but as information to the effect that each rewrite rule is applicable to the relevant place. The editing results are displayed to the user in the editing result display unit 12, and the user inputs an instruction as to whether or not to actually apply to the speech synthesis symbol editing unit 13. The edited result may be updated and displayed by the editing result display unit 12. FIG.

As described above, according to the speech synthesis symbol editing device 10 of each embodiment, the correspondence between each character of the text and the speech synthesis symbol token is obtained regardless of the text analysis method of the target TTS system (the text analysis unit 21 in FIG. 2). Relationships can be presented to the user. As a result, the user can refer to the text to search for corrections in the synthesized speech symbol string, and can easily correct misreading of the synthesized speech. In addition, the TTS system and the text-to-speech symbol editing device 10 become more independent, and user convenience can be maintained without modifying the text-to-speech symbol editing device 10 each time the TTS system is modified.

That is, by uniquely generating the correspondence relationship between the characters constituting the text and the reading symbol of the speech synthesis symbol string by the correspondence estimation unit 11 in the speech synthesis symbol editing device 10, the user's convenience can be maintained and the TTS Modifications to the text analysis unit 21 within the system can be prevented from affecting the text-to-speech symbol editing system.

Various supplementary examples, additional examples, alternative examples, etc. will be described below.

<1> Exceptional Processing in Correspondence Estimating Unit 11 In the correspondence estimating unit 11, each character of the above-described text may exceptionally be treated as one character when a plurality of characters are consecutive. For example, for ``kyou''(kyo'-), the above method can be processed without failure by conveniently associating ``now'' = ``kyo''' and day = ``-''. However, the display of such a correspondence gives the user a sense of discomfort. On the other hand, this problem can be avoided by exceptionally regarding "today" as one character.

Also, when the correspondence estimation unit 11 sets a large penalty for assigning no reading symbols to a certain character in the text as described above, In such a case, the estimation of the correspondence relationship fails, and at least the surrounding characters of the text are also affected. However, since the number of words that require such processing is limited, it is possible to determine in advance a rule based on experience regarding what kind of character string is to be treated as a single character.

Note that the speech synthesis symbol editing unit 13 does not assume modification of text characters, except for manipulation of empty characters "" accompanying insertion/deletion of prosodic boundary symbols, so this method does not cause any problems.

Further, as a contrivance in terms of implementation, when the correspondence estimation unit 11 divides the text into character units, an empty character "" may always be inserted between the original characters. In an implementation that always inserts a blank character " " between characters of the original text, the text-to-speech symbol editing unit 13 does not need to insert/delete the blank character " " accompanying addition/deletion of prosodic boundary symbols.

<2> With respect to the speech synthesis symbol correction unit 14, although it was described above that the correction result is not reflected in the editing buffer held by the editing interface unit 123, it may be reflected. That is, the result of correction similar to the output result of the speech synthesis symbol correction unit 14 may be displayed on the editing result display unit 12 and reflected as the automatic correction result.

As mentioned above, if no correction is made, the malformed text-to-speech symbol string will be displayed in the editing result display section 12. Displaying a speech synthesis symbol string that cannot be synthesized into speech as it is may impair the user's convenience. Therefore, as a modification, this problem can be avoided by displaying the corrected result.

<3> Instead of displaying the speech synthesis symbol token as character information, or in addition to displaying the speech synthesis symbol token as character information, the editing result display unit 12 graphically displays, for example, the pitch pattern corresponding to the speech synthesis symbol token. You may

<4> In addition to using the speech synthesis symbol editing device 10 as a single configuration as shown in FIG. and/or the speech synthesizing unit 22 may be arranged on the rear stage side of the speech synthesis symbol editing device 10 and used. In FIG. 2 , the synthetic speech editing device 20 is used for convenience, but the text analysis unit 21 and/or the speech synthesis unit 22 may be included in the speech synthesis symbol editing device 10 .

<5> FIG. 4 is a diagram showing an example of a hardware configuration in a general computer device 70. As shown in FIG. The speech synthesis symbol editing device 10 (or the synthetic speech editing device 20, hereinafter the same) can be implemented as one or more computer devices 70 having such a configuration. When the speech synthesis symbol editing device 10 is realized by two or more computer devices 70, information required for processing may be transmitted and received via a network. The computer device 70 includes a CPU (central processing unit) 71 that executes predetermined instructions, and a GPU (graphics processing unit) as a dedicated processor that executes part or all of the execution instructions of the CPU 71 instead of the CPU 71 or in cooperation with the CPU 71. ) 72, RAM 73 as a main storage device that provides a work area to CPU 71 (and GPU 72), ROM 74 as an auxiliary storage device, communication interface 75, display 76, input interface 77 that accepts user input by mouse, keyboard, touch panel, etc. A microphone 78 and a bus BS for exchanging data therebetween are provided.

Each functional unit of the speech synthesis symbol editing device 10 can be realized by the CPU 71 and/or the GPU 72 that reads and executes a predetermined program corresponding to the function of each unit from the ROM 74 . Both the CPU 71 and the GPU 72 are a kind of arithmetic unit (processor). Here, when display-related processing is performed, the display 76 further operates in conjunction, and when communication-related processing relating to data transmission/reception is performed, the communication interface 75 further operates in conjunction. The result of processing by the speech synthesis symbol editing device 10 (display by the editing result display unit 12, etc.) may be displayed on the display 76 and output. The synthesized speech obtained by the speech synthesizing unit 22 may be reproduced by the microphone 78 and output to the user.

<6> According to the text-to-speech symbol editing device 10 according to each embodiment of the present invention, it is possible to improve the efficiency of editing work for obtaining high-quality synthetic speech (consisting of natural utterances) from text. It becomes possible. This provides high-quality synthesized speech for people with disabilities who cannot access information by reading it visually as text, or in environments where such constraints are imposed. By doing so, it will be possible to provide easy-to-understand and smooth information access at a lower cost. It will be possible to contribute to "correcting inequality between nations."

10...speech synthesis symbol editing device, 11...correspondence estimation unit, 12...editing result display unit, 13...speech synthesis symbol editing unit, 123...editing interface unit, 14...speech synthesis symbol correcting unit, 15...editing record database, 16…Editing Reproduction Section

Claims (14)

A speech synthesis symbol editing device for inputting a set of text and a speech synthesis symbol string corresponding to the text and outputting a speech synthesis symbol string,
dividing the input text into predetermined character units, dividing the input speech synthesis symbol into predetermined speech synthesis symbol tokens, and combining each of the divided characters with zero or more speech synthesis symbol tokens; a correspondence estimating unit that gives the correspondence of based on a predetermined criterion;
an editing interface unit that displays each character given the corresponding relationship and each of zero or more phoneme synthesis symbol tokens while updating the phoneme synthesis symbol tokens by accepting a user editing operation. Text-to-speech symbol editor. 2. The text-to-speech symbol editing apparatus according to claim 1, wherein the user editing operation is modification, addition or deletion of a text-to-speech symbol token. Of the string of phoneme synthesis symbol tokens updated and displayed on the editing interface unit, if there is a part that violates a predetermined rule for speech synthesis with respect to a substring that has been specified by the user, it conforms to the predetermined rule. 3. The speech synthesis symbol editing apparatus according to claim 1, further comprising a speech synthesis symbol correction unit for outputting a string of phoneme synthesis symbol tokens modified as follows. 4. The speech synthesis symbol editing apparatus according to claim 3, wherein said editing interface unit holds a string of phoneme synthesis symbol tokens without reflecting the correction by said speech synthesis symbol correction unit. 4. The speech synthesis symbol editing apparatus according to claim 3, wherein said editing interface unit holds a string of phoneme synthesis symbol tokens reflecting the correction by said speech synthesis symbol correcting unit. 2. The apparatus further comprises an editing record database for storing, as editing information, user editing operations accepted by the editing interface section in the form of rewriting rules for phoneme synthesis symbol tokens for sets of characters and phoneme synthesis symbol tokens. 6. The speech synthesis symbol editing device according to any one of 1 to 5. The editing record database is referred to for each character and each string of zero or more phoneme synthesis symbol tokens to which the correspondence relationship is given in the correspondence relationship estimation unit, and the rewriting rule is determined for the locations where the rewriting rule is applicable. 7. The text-to-speech symbol editing device according to claim 6, further comprising: an editing reproduction unit that outputs to the editing interface unit by applying the rewriting rule or as a state in which the rewriting rule is applicable. . The correspondence estimation unit has a mechanism for generating possible readings for each character of a predetermined character set, and the predetermined criterion in the correspondence estimation unit is 8. A criterion for associating characters with one or more speech synthesis symbol tokens in said speech synthesis symbol string corresponding to the generated reading of each character. 1. The speech synthesis symbol editing device according to claim 1. 9. The method of claim 8, wherein the predetermined character set includes a set of hiragana and katakana, and the mechanism for generating possible readings allows for cases where there are no possible readings for kanji characters. Text-to-speech symbol editor. 10. The correspondence estimating unit according to claim 8, wherein the correspondence estimating unit has a mechanism for treating a predetermined character string in the text as one character to correspond to one or more speech synthesis symbol tokens. text-to-speech symbol editor. further comprising a text analysis unit that analyzes the text and outputs a corresponding speech synthesis symbol string;
11. The speech synthesis symbol editing apparatus according to claim 1, wherein the correspondence estimation unit receives the speech synthesis symbol string output from the text analysis unit as an input corresponding to the text. Further comprising a speech synthesizing unit that performs speech synthesis on a substring that has been specified by a user, of the string of phoneme synthesis symbol tokens that are updated and displayed by the editing interface unit, and outputs a corresponding synthesized speech. 12. The speech synthesis symbol editing apparatus according to any one of claims 1 to 11, characterized by: A speech synthesis symbol editing method for inputting a set of text and a speech synthesis symbol string corresponding to the text and outputting a speech synthesis symbol string,
dividing the input text into predetermined character units, dividing the input speech synthesis symbol into predetermined speech synthesis symbol tokens, and combining each of the divided characters with zero or more speech synthesis symbol tokens; a correspondence estimation step of giving the correspondence of based on a predetermined criterion;
and an editing interface stage for displaying each character given the correspondence and zero or more phoneme synthesis symbol tokens while updating the phoneme synthesis symbol tokens by accepting a user editing operation. How to edit text-to-speech symbols. A speech synthesis symbol editing device for inputting a set of text and a speech synthesis symbol string corresponding to the text and outputting a speech synthesis symbol string,
dividing the input text into predetermined character units, dividing the input speech synthesis symbol into predetermined speech synthesis symbol tokens, and combining each of the divided characters with zero or more speech synthesis symbol tokens; a correspondence estimating unit that gives the correspondence of based on a predetermined criterion;
a speech synthesis symbol editing device, comprising: an editing interface unit that displays each character given the corresponding relationship and each of zero or more phoneme synthesis symbol tokens while updating the phoneme synthesis symbol tokens by accepting a user editing operation. As
A text-to-speech symbol editing program that causes a computer to function.

Images