JP2007025338A - Method and device for speech synthesis, and computer program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a device and a method for speech synthesis, and a computer program with which deterioration in voice quality can be suppressed even when a speech is synthesized so that an output speech of a keyword part and an output speech of other parts can be discriminated by changing the speaker, sound volume, tone interval, or speaking speed of the speech. <P>SOLUTION: The device (100, 900) for speech synthesis is equipped with a keyword-priority phoneme selection section (107, 109) which selects candidates for phonemes of all keywords included in a text body from a corpus in the appearance order of the keywords preferentially to candidates for phonemes of elements other than the keywords. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to speech synthesis for reading a document, and more particularly to a speech synthesis method, speech synthesis apparatus, and computer program for emphasizing keywords.

  2. Description of the Related Art Generally, a speech synthesizer is known that converts a document stored in a PC (personal computer) into a voice and reads it out based on a natural voice of a person recorded in advance. The speech synthesizer creates synthesized speech based on a corpus in which natural speech that can be divided into parts of speech is recorded.

  In the speech synthesis processing by the speech synthesizer, for example, morpheme analysis and dependency analysis are executed on the input text, and converted into phoneme symbols, accent symbols, and the like.

  Next, the phoneme duration (voice length), fundamental frequency (voice pitch), vowel-centric power (by using the part-of-speech information of the input text obtained from phoneme symbols, accent symbol strings, and morpheme analysis results ( The loudness etc. is estimated.

  Next, the synthesis unit that is closest to the estimated phoneme duration, fundamental frequency, vowel center power, etc. and that has the smallest distortion when connecting synthesis units (phonemes) stored in the waveform dictionary is the smallest. A combination is selected using dynamic programming or the like. The unit selection performed at this time uses a scale (cost value) that matches the perceptual feature.

  When the combination of the synthesis units is selected, the speech is synthesized by connecting the phonemes while changing the pitch in accordance with the selected phoneme combination. The above is the outline of the speech synthesis process.

  Among the speech synthesizers described above, an apparatus has been developed that can read out important parts in a document and emphasize parts that the document creator particularly wants to convey to the reader (for example, see Patent Document 1). .

JP-A-10-274999

  However, when emphasizing important parts in the document in the speech synthesizer, the output speech for the keyword portion and the output speech for other portions are changed by changing the speaker, volume, pitch, or speech speed of the speech. When speech synthesis is performed so that can be identified, there is a problem that the sound quality of the emphasized portion deteriorates.

  The present invention has been made in view of the above problems, and an object of the present invention is to change the voice speaker, volume, pitch, or speech speed, and output voice to the keyword part and output to other parts. To provide a new and improved speech synthesizer, speech synthesis method, and computer program capable of suppressing deterioration of sound quality even if speech synthesis is performed such that speech can be distinguished.

  In order to solve the above problems, according to a first aspect of the present invention, there is provided a speech synthesis method for performing speech synthesis by emphasizing a keyword portion. In the above speech synthesis method, a selection process is executed in which phoneme candidates for all keywords included in the text body are selected from the corpus in order of appearance of the keywords in preference to phoneme candidates other than the keywords. It is characterized by that. The corpus records, for example, natural speech that can be divided at least into parts of speech, but is not limited to this example.

  The selection process uses a combination of unit candidates that minimizes the cost value of the keyword part, and selects a unit candidate that has the minimum cost value from the start position of the text body to the first keyword start position. A combination is selected; when there are two or more keywords, the combination of unit candidates that minimizes the cost value is selected from the end position of the keyword toward the start position of the subsequent keyword. Also good.

  In the speech synthesis method, before the selection process, a keyword extraction process is performed to search from the head of the phonological symbol whether the phonological symbol of the keyword partially matches the phonological symbol of the text body. Also good.

  The keyword extraction process changes the value of the keyword position information for each phoneme described in the prosodic prediction information based on a location where the phoneme symbol of the keyword and the phoneme symbol of the text body partially match; Based on the prosodic prediction information including the changed keyword position information, the selection process may select phoneme candidates for all keywords included in the text sentence from the corpus.

  By changing the value of the keyword position information of each phoneme described in the prosodic prediction information, which is information for predicting at least one of voice pitch, voice length, or mel cepstrum, You may make it show that the phonetic symbol of a text main body corresponds partially.

  The speech synthesis method may be configured to further perform a weighting process for assigning a keyword weighting coefficient indicating a degree of emphasizing the keyword part to one or more keywords in the text body. .

  In the keyword extraction process, the keyword weighting coefficient assigned in the weighting process is acquired as keyword weighting information for one or more keywords in the text body, and the combination of the keyword weighting information and the unit candidate is acquired. A table in which a width for narrowing down is associated may be used.

  In the selection process described above, the unit candidate combinations are narrowed down based on the value that minimizes the cost value obtained by adding the target cost range, the sub cost value related to the pitch discontinuity, and the sub cost value related to the spectrum discontinuity. A combination of unit candidates having cost values that fall within the range of the width may be selected.

  In order to solve the above-described problem, according to another aspect of the present invention, a speech synthesizer for emphasizing a keyword portion and performing speech synthesis is provided. The speech synthesizer includes a keyword priority phoneme selection unit that selects phoneme candidates of all keywords included in the text body from a corpus in preference to phoneme candidates other than the keyword in the order of appearance of the keywords. It is characterized by providing. The corpus records, for example, natural speech that can be divided at least into parts of speech.

  The keyword-preferred phoneme selection unit uses a combination of unit candidates that minimizes the cost value of the keyword part, and proceeds to the keyword start position that appears first from the start position of the text body. A candidate combination is selected; when there are two or more keywords, a unit candidate combination having the minimum cost value is selected from the end position of the keyword toward the start position of the subsequent keyword. You may do it.

  The speech synthesizer may further include a keyword extraction unit that searches from the head of the phoneme symbol whether the phoneme symbol of the keyword partially matches the phoneme symbol of the text body.

  The keyword extraction unit changes the value of the keyword position information for each phoneme described in the prosodic prediction information based on a location where the phoneme symbol of the keyword and the phoneme symbol of the text body partially match; The keyword priority phoneme selection unit selects from the corpus candidate phonemes of all keywords included in the text sentence based on the prosodic prediction information including the changed keyword position information. Also good.

  The keyword extraction unit changes the value of the keyword position information of each phoneme described in the prosodic prediction information that is information for predicting at least one of voice pitch, voice length, and mel cepstrum. The value of the changed keyword position information may indicate that the phonological symbol of the keyword partially matches the phonological symbol of the text body.

  The speech synthesizer may further include a weighting unit that assigns a keyword weighting coefficient indicating a degree of emphasizing the keyword part to one or more keywords in the text body.

  The keyword extraction unit acquires, as keyword weighting information, the keyword weighting coefficient assigned by the weighting unit for one or more keywords in the text body, and combines the keyword weighting information and the unit candidate. You may comprise so that the table which matched the width | variety to narrow down may be used.

  The keyword-preferred phoneme selection unit selects the unit candidate based on a value that minimizes the cost value obtained by adding the target cost range, the sub-cost value related to the pitch discontinuity, and the sub-cost value related to the spectrum discontinuity. A combination of unit candidates having cost values that fall within the range of narrowing down the combinations may be selected.

  In order to solve the above problems, according to another aspect of the present invention, there is provided a computer program that causes a computer to function as a speech synthesizer that performs speech synthesis by emphasizing a keyword portion.

  As described above, according to the present invention, even if the voice speaker, volume, pitch, or speech speed is changed, the voice can be distinguished from the output voice for the keyword part and the output voice for the other part. Even if it is synthesized, it is possible to minimize the deterioration of the sound quality of the keyword part to be emphasized.

  DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, preferred embodiments of the invention will be described in detail with reference to the accompanying drawings. In the following description and the accompanying drawings, components having substantially the same functions and configurations are denoted by the same reference numerals, and redundant description is omitted.

(About voice synthesizer)
First, the speech synthesis apparatus 100 according to the first embodiment will be described with reference to FIG. FIG. 1 is a block diagram illustrating a schematic configuration of the speech synthesizer according to the first embodiment.

  As shown in FIG. 1, the speech synthesizer 100 includes a text analysis unit 101, a prosody prediction unit 103, a keyword extraction unit 105, a keyword priority phoneme selection unit 107, a corpus 109, and a phoneme connection unit 111. ing.

  As shown in FIG. 1, the text analysis unit 101 inputs a text body expressed in Kanji kana characters and a keyword expressed in Kanji kana characters to be emphasized in the text body. Convert keywords to phonetic symbols. Note that the phoneme is a unit that can be segmented as represented by, for example, a phoneme symbol, but is not limited to such an example.

  In addition, the text analysis unit 101 performs morpheme analysis and dependency analysis on the text body expressed by the kanji characters after conversion into the above phoneme symbols, and the morpheme analysis result indicating the accent symbol string and the part of speech information of the text body Is output.

The prosody prediction unit 103 is based on the part of speech information of the text body obtained from the phoneme symbol of the text body converted by the text analysis unit 101, the accent symbol string output from the text analysis unit 101, and the morphological analysis result. Thus, a pitch (voice pitch: fundamental frequency F ₀ ), a phoneme duration (voice length), and a mel cepstrum representing a waveform component are predicted. The predicted result is prosodic prediction information. Details of the mel cepstrum and the like are described in JP-A-2003-208188.

  As shown in FIG. 1, the keyword extraction unit 105 searches in order from the head of the phoneme symbol of the text body whether or not the phoneme symbol of the keyword partially matches in the phoneme symbol of the text body.

  In addition, the keyword extraction unit 105 determines the keyword for each phoneme described in the prosodic prediction information received from the prosody prediction unit 103 based on the location where the phonological symbol of the keyword partially matches the phonological symbol in the text body. The value of the position information is changed for all phonemes that partially match among the phoneme symbols in the text body.

  The keyword priority phoneme selection unit 107 selects a phoneme from the corpus 109 using the pitch predicted by the prosody prediction unit 103, the phoneme duration, and the mel cepstrum as parameters of the phoneme selection process. The corpus 109 is stored in storage means such as a hard disk drive.

  In the process of selecting a phoneme, the keyword priority phoneme selection unit 107 uses a scale (cost) that matches the perceptual characteristic. In addition, the cost function mapped from the observable feature quantity to the psychological quantity is the sub-cost related to prosody, the sub-cost related to pitch discontinuity, the sub-cost related to phonological environment substitution, the sub-cost related to spectrum discontinuity, and the phonetic It is configured as a function obtained by adding five weighted sub-cost functions with the sub-cost related to suitability (for example, see Japanese Patent Application Laid-Open No. 2003-208188).

  Note that the speech synthesizer 100 is a device capable of outputting a synthesized speech based on a text body and a keyword, and can output the text body as speech by outputting the synthesized speech. Device. More specifically, the speech synthesizer 100 exemplifies, for example, a PC having an input unit corresponding to a CPU, memory, HDD (hard disk drive), mouse, etc., a display unit equivalent to a liquid crystal display, etc. However, it is not limited to such an example.

  In addition, the display part with which the speech synthesizer 100 concerning this Embodiment is provided outputs the display screen data and audio | voice data processed so that display was possible by CPU. In addition, the display unit is exemplified by a TV or a liquid crystal display device, for example, and both of them are provided with a speaker, and can output a sound or a moving image in addition to a still image.

  The input unit is, for example, a pointing device such as a mouse, a trackball, a trackpad, a stylus pen, or a joystick that can receive an operation instruction from a user, an operation means such as a keyboard, a button, a switch, and a lever, and an input It comprises an input control circuit that generates a signal and outputs it to the CPU. The user of the speech synthesizer 100 can input various data and instruct a processing operation to the speech synthesizer 100 by operating this input unit.

(Speech synthesis method)
Next, the speech synthesis method according to the first embodiment will be described with reference to FIG. FIG. 2 is a flowchart showing an outline of the speech synthesis method according to the first embodiment.

  As shown in FIG. 2, first, a text body including one or more keywords and one or more keywords to be emphasized are input to the text analysis unit 101 (S201). The text body and the keyword will be described by taking an example where the text body and the keyword are expressed in kanji characters, but are not limited to such examples.

  Next, the text analysis unit 101 converts the input text body and keywords into phonological symbols (S203).

  The text analysis unit 101 performs morphological analysis and dependency analysis on the text body expressed by kanji characters, and outputs an accent symbol string and a morphological analysis result representing the part-of-speech information of the text body (S203).

  As shown in FIG. 2, the output result having information on the phoneme symbol converted to the text body and the morphological analysis result is defined as, for example, an intermediate language of the text body, but is limited to such an example. Not.

  Here, the phoneme symbol 501 in which the text body is converted (“text body phoneme symbol” shown in FIG. 5A) is, for example, “hajime... Aoki” as shown in FIG. 5A.

  An output result having a phoneme symbol converted for a keyword is used as a keyword intermediate language. However, the present invention is not limited to such an example.

  Further, as shown in FIG. 5A, the keyword phonological symbol 502 (“keyword phonological symbol” shown in FIG. 5A) is, for example, “aoki”.

  Next, as shown in FIG. 2, the text analysis unit 101 determines whether the text body is a text body in order to input the text body to the prosody prediction unit 103 (S205).

  As a result of the determination (S205), if it is a text body, the text analysis unit 101 outputs the text body intermediate language and transmits it to the prosody prediction unit 103.

  Furthermore, if the result of determination (S205) is a text body, the text analysis unit 101 outputs the text body intermediate language to the keyword extraction unit 105.

  On the other hand, if it is not a text body, that is, if it is a keyword, the text analysis unit 101 transmits the keyword intermediate language to the keyword extraction unit 105.

Next, the prosody prediction unit 103 selects at least one or all of pitch (voice pitch: fundamental frequency F ₀ ), phoneme duration (voice length), or mel cepstrum representing a waveform component. Prediction is performed (S207).

  The prosody prediction unit 103 outputs prosodic prediction information to the keyword extraction unit 105 as predicted information.

  Here, the prosody prediction information will be described. As shown in FIG. 5A, the prosody prediction information 503 is a phoneme (prosody prediction information 503) of a phoneme symbol 501 (“hajime ... aoki ...” in the example shown in FIG. 5A). "Hajime ... aoki ..." described in the vertical direction in the figure), "start" indicating the phoneme start time, "duration" indicating the duration of the phoneme, and one or more pitches of the phoneme "Pitch" representing the melody, and "Mel cep" representing one or more mel cepstrum of the phoneme.

  Next, as shown in FIG. 2, the keyword extraction unit 105 performs each phoneme described in the prosodic prediction information based on the location where the keyword phoneme symbol partially matches the phoneme symbol in the text body. The value of each keyword position information is changed for all phoneme symbols that partially match the phoneme symbols in the text body, and the changed prosodic prediction information with keyword position information is output to the keyword priority phoneme selection unit 107 To do.

(About processing by the keyword extraction unit)
Next, as shown in FIG. 2, the keyword extraction unit 105 describes information (keyword position information) for indicating the position where the keyword exists in the text body, so that each phoneme in the prosodic prediction information is recorded. An area is secured for and initialized (S209). In the initialization, the flag value of the keyword position information is set to false, but the present invention is not limited to this example.

  Next, the keyword extraction unit 105 converts the phoneme position indicator that indicates which phoneme is currently pointed to in the phoneme symbol between the text body and the keyword to the text body and the first phoneme in one or more keywords. Set (S211).

  Next, the keyword extraction unit 105 obtains the number of phonemes for the text body and a plurality of keywords, and sets the number of keyword phonemes for each keyword phoneme number and the number of text body phonemes for the text phoneme number (S213). ).

  Next, the keyword extraction unit 105 determines whether the end of the text body is reached (S215).

  As a result of determining whether or not it is the end of the text body (S215), if it is not the end of the text body, the keyword extracting unit 105 extracts a keyword portion in the text body (S217). The keyword extraction process (S217) will be described later.

  In the case of the end of the text body, the keyword extraction unit 105 assigns keyword priority information to prosodic prediction information having prosodic prediction information and keyword position information (accompanying) for each phoneme in the prosodic prediction information. Output to the phoneme selection unit 107.

  As shown in FIG. 4B, the criterion of whether or not the end of the text body is determined is based on whether or not the phoneme position indicator of the text body is smaller than the number of text phonemes. If the phoneme position indicator of the text body is smaller than the number of text phonemes, it is determined that the end of the text body is reached.

  In step S219 shown in FIG. 2, although the details will be described later, the keyword-preferred phoneme selection unit 107 selects an optimal phoneme from the keyword portion in the text body using the corpus 109 shown in FIG.

  When the phoneme selection for the keyword part is completed, the optimal phoneme for the part other than the keyword part is selected.

  All the phoneme selection processes are performed by the keyword priority phoneme selection unit 107, so that a waveform segment can be output (S219).

  Next, the phoneme connection unit 111 determines whether the phoneme of the waveform segment currently handled is the end of the text body (S221).

  As a result of the above determination (S221), if it is not the end of the text body, the phoneme connection unit 111 connects the current waveform segment to the next waveform segment (S223).

  On the other hand, if it is the end of the text body (S221), the phoneme connection unit 111 outputs synthesized speech generated by connecting the waveform segments (S225). With the output of the synthesized speech (S225), the speech synthesizer 100 can read the text body while emphasizing the keyword.

(Keyword location extraction process)
Next, the keyword location extraction process (S217) shown in FIG. 2 will be described in more detail with reference to FIGS. 3, 4A, and 4B. 3 and 4A are flowcharts showing an outline of the keyword location extraction processing according to the first embodiment, and FIG. 4B shows whether or not it is the ending of the keyword according to the first embodiment. It is explanatory drawing which shows the outline of this judgment standard.

  As shown in FIG. 3, if it is not the end of the text body (S215), the keyword extraction unit 105 handles the keywords in order from one or more keywords, so It is determined whether or not it falls within the range of the number of keywords (S240).

  The order of keywords currently handled is the keyword input order, but is not limited to this example.

  Next, the keyword extraction unit 105 determines whether the phonological position indicator of the currently handled keyword (this keyword) indicates the ending of the keyword (S241).

  When it is the ending of this keyword (S241), the keyword position information for each phoneme in the prosodic prediction information in the prosodic prediction information of the phonemes of the number of the keyword phonemes before the position indicated by the phoneme position indicator in the text body is true. (Step S253 shown in FIG. 4A).

  As shown in FIG. 4B, the criterion for determining whether or not this keyword is ending is determined by whether or not the phoneme position indicator of the currently handled keyword (this keyword) is smaller than the number of phonemes of this keyword. .

  As shown in FIG. 3, the keyword extraction unit 105 determines whether or not the text body, the phoneme of the text body indicated by the phoneme position indicator of the keyword, and the phoneme of the keyword are the same phoneme. Determination is made (S243).

  When the phonemes match (S243), the phoneme position indicator of the keyword is set as the next phoneme (S247). When the phonemes do not match (S243), the phoneme position indicator of the keyword is set as the first phoneme. (S245).

  Next, the keyword extraction unit 105 sets the position pointed to by the phoneme position indicator of the currently handled keyword (this keyword) as the first phoneme of the keyword (S245).

  Next, the keyword extraction unit 105 sets the position pointed to by the phoneme position indicator of the currently handled keyword (this keyword) to the phoneme next to the phoneme currently pointed by the keyword position indicator (S247). ).

  Next, the keyword extraction unit 105 changes the currently handled keyword to the next keyword in the keyword input order (S249).

  If it is the end of the text body (S215), the keyword extraction unit 105 outputs prosodic prediction information with keyword position information (S251).

  Further, as shown in FIG. 4A, the keyword extraction unit 105 counts the number of the keyword phonemes from the phoneme pointed by the phoneme position indicator in the text body toward the head phoneme. The keyword position information for each phoneme in the prosodic prediction information corresponding to all the phonemes is changed (S253). The keyword position information is changed from false to true, but is not limited to this example.

  Next, as shown in FIG. 4A, the keyword extraction unit 105 executes the same processing as step S245 shown in FIG. 3 (S255).

  If the result of determination shown in FIG. 3 is not within the number of keywords (S240), the keyword extraction unit 105 changes the keyword to the first keyword as shown in FIG. 4A (S257).

  Next, as shown in FIG. 4A, the keyword extraction unit 105 sets the position pointed to by the phoneme position indicator in the text body to the phoneme next to the phoneme currently pointed by the phoneme position indicator in the text body (S259). ).

  FIG. 5B shows the flow of data processed by each unit of the text analysis unit 101 to the phoneme connection unit 111 in the speech synthesizer 100 according to the first embodiment shown in FIG.

  Thus, a series of processing of the keyword location extraction processing (S217) by the keyword extraction unit 105 shown in FIGS. 3 and 4 is completed.

(Keyword priority phoneme selection processing by the keyword priority phoneme selection unit 107)
Next, as shown in FIG. 2, when a keyword part is extracted (S217), keyword priority phoneme selection processing by the keyword priority phoneme selection unit 107 is executed (S219).

  In the keyword priority phoneme selection process (S219), a sub cost value related to phoneme compatibility, a sub cost value related to phonological environment substitution, and a sub cost value related to prosody are obtained by using the corpus 109 for the phoneme of the text body. Then, phonemes included in a range of values having a certain range are selected (target selection) from the minimum sub cost value (target cost value) obtained by adding the three sub cost values.

  Here, the keyword priority phoneme selection process (S219) according to the first embodiment will be described in detail with reference to FIGS. FIG. 6 is an explanatory diagram showing an outline of the keyword priority phoneme selection process according to the first embodiment. FIGS. 7 to 10 show an outline of the keyword priority phoneme selection process according to the first embodiment. It is a flowchart which shows.

  As shown in FIG. 6, the keyword priority phoneme selection unit 107, based on the phoneme candidate obtained by the target selection, the keyword start position where the flag value of the keyword position information given to each phoneme of the text body is true. To the keyword end position, the target cost value, the sub-cost value related to the pitch discontinuity, and the sub-cost value related to the spectrum discontinuity are combined to generate a unit candidate that minimizes the cost value. Use to select.

  The keyword priority phoneme selection unit 107 uses the combination of unit candidates that minimizes the cost value of the keyword part to select the combination of unit candidates that minimizes the cost value from the start position of the text body toward the keyword start position. , Select using dynamic programming.

  The keyword priority phoneme selection unit 107 uses the combination of unit candidates that minimizes the cost value of the keyword part to select the combination of unit candidates that minimizes the cost value from the keyword end position toward the end position of the text body. , Select using dynamic programming.

  As shown in FIG. 6, when there are two or more keywords, the cost value for the keyword portion is calculated in the order of appearance of the keywords from the start to the end of the text body.

  Also, as shown in FIG. 6, when there are two or more keywords, the cost value calculation method other than the keyword part is calculated from the start position of the text body toward the first keyword start position. The smallest unit candidate combination is selected using dynamic programming.

  As shown in FIG. 6, when there are two or more keywords, the cost value calculation method other than the keyword part uses the combination of unit candidates that minimizes the cost value of the keyword part. ) From the end position to the start position of the next appearing keyword (subsequent keyword), the combination of unit candidates that minimizes the cost value is selected using dynamic programming.

  As shown in FIG. 6, when there are two or more keywords, the cost value calculation method other than the keyword part is the minimum cost value from the end position of the keyword that appears last to the end position of the text body. A unit candidate combination is selected using dynamic programming.

  As shown in FIG. 7, the keyword priority phoneme selection unit 107 first selects a target for the phoneme of the text body (S271).

  When the target is selected (S271), the keyword priority phoneme selection unit 107 sets the search variable to the first phoneme of the text body and sets the number of keywords to 0 (S273), thereby determining which phoneme is currently indicated in the text body. The expressed search variable points to the phoneme at the beginning of the text body.

  Next, the keyword priority phoneme selection unit 107 determines whether or not the position of the phoneme pointed to by the search variable is within the range of the total number of phonemes in the text body (S275). That is, as shown in FIG. 7, it is determined whether or not the relation of search variable <number of phonemes of text is satisfied.

  If the result of determination in step S275 is that the search variable is smaller than the number of phonemes in the text, the keyword priority phoneme selection unit 107 determines whether the phoneme pointed to by the search variable is a phoneme in the keyword (S277).

  Normally, when the phoneme pointed to by the search variable is a phoneme in a keyword, the keyword position information is true. If it is not a phoneme within the keyword, the keyword position information is false.

  On the other hand, if the search variable is larger than the number of phonemes of the text (S275), the subsequent processing shown in FIG. 8 (set the search variable to the first phoneme of the text body) is executed.

  If the keyword position information pointed to by the search variable is false (S277), the search variable is set to the next phoneme (S287).

  If the keyword position information pointed to by the search variable is true, a process of checking whether the keyword position information of the phoneme next to the search variable is true or whether there is the next phoneme is executed (S279).

  As shown in FIG. 7, in step S279, it is determined whether the phoneme next to the phoneme indicated by the search variable is a phoneme in the keyword.

  If the keyword position information of the phoneme next to the search variable is true, the cost value of the phoneme pointed to by the search variable and the next phoneme is obtained (S281).

  If the keyword position information of the phoneme next to the search variable is false, the number of keywords is incremented by 1 (S283).

  As shown in FIG. 7, in step S281, between the candidate obtained by the target selection of the phoneme indicated by the search variable and the candidate obtained by the target selection of the phoneme next to the phoneme indicated by the search variable. The cost value obtained by adding the target cost value, the sub cost value related to the pitch discontinuity, and the sub cost value related to the spectrum discontinuity is obtained.

  When the cost value is obtained (S281), the search variable is set to the next phoneme (S287).

  Further, when the number of keywords expressing how many keywords exist in the text body is incremented by 1 (S283), the keyword priority phoneme selection unit 107 sets a path with the minimum cost value (S285).

  In step S285 for setting the path, between the candidate obtained by target selection of the phoneme indicated by the search variable and the candidate obtained by target selection of the phoneme next to the phoneme indicated by the search variable, A combination (path) of unit candidates that obtains the minimum cost value by adding the target cost value, the sub cost value related to the pitch discontinuity, and the sub cost value related to the spectrum discontinuity is obtained.

  When the path with the minimum cost value is set (S285), the search variable is set to the next phoneme (S287).

  As shown in FIG. 8, the keyword priority phoneme selection unit 107 sets the search variable to the head phoneme of the text body (S290). Note that step S290 is substantially the same as step S273 in FIG.

  When the search variable is set to the head phoneme of the text body (S290), it is confirmed whether or not the keyword position information pointed to by the search variable is false (S291). Note that step S291 is substantially the same as step S277 in FIG.

  When the keyword position information indicated by the search variable is false (S291), it is confirmed whether the keyword position information of the phoneme next to the search variable is false (S293).

  On the other hand, if the keyword position information pointed to by the search variable is true (S291), it is confirmed whether the keyword position information pointed to by the search variable is true (S298).

  As shown in FIG. 8, in step S293, it is determined whether or not the next phoneme of the phoneme indicated by the search variable is a phoneme in the keyword.

  Next, when the keyword position information of the phoneme next to the search variable is false (S293), the cost value of the phoneme pointed to by the search variable and the next phoneme is obtained (S295). Note that step S295 and the process of obtaining the cost value (S281) in FIG. 7 are substantially the same.

  If the keyword position information of the phoneme next to the search variable is true (S293), a path with the minimum cost value is set (S296). Note that step S296 is substantially the same as the process (S285) for setting the minimum path in FIG.

  After obtaining the cost value of the phoneme indicated by the search variable and the next phoneme, the search variable is set to the next phoneme (S297). Note that step S297 and the processing for setting the next phoneme in FIG. 7 (S287) are substantially the same.

  Next, as shown in FIG. 8, the keyword priority phoneme selection unit 107 checks whether or not the keyword position information indicated by the search variable is false (S298). Note that step S298 is substantially the same as the process (S277) for confirming whether the keyword position information indicated by the search variable in FIG. 7 is false.

  As a result of the confirmation (S298), if the keyword position information indicated by the search variable is true, it is confirmed whether or not the result of decrementing (subtracting) the number of keywords by 1 is 1 or more (S299).

  On the other hand, if the keyword position information indicated by the search variable is false as a result of the confirmation (S298), the subsequent processing (S301) shown in FIG. 9 is executed.

  As shown in FIG. 8, in step S299, if there are a plurality of keywords in the text body, and there is one or more phonemes between the keywords, the search variable is set to the next phoneme (S300). . Step S300 is substantially the same as step S287 in FIG.

  Next, as shown in FIG. 9, the keyword priority phoneme selection unit 107 first checks whether or not the keyword position information pointed to by the search variable is false (S301), and then the keyword position information pointed to by the search variable is If it is false, it is confirmed whether or not the keyword position information of the phoneme next to the search variable is false (S303). Step S303 is substantially the same as step S293 in FIG.

  If the keyword position information of the phoneme next to the search variable is false (S303), the cost value of the phoneme pointed to by the search variable and the next phoneme is obtained (S305).

  On the other hand, if the keyword position information of the phoneme next to the search variable is true, a path with the minimum cost value is set (S307).

  Next, as shown in FIG. 9, the above-described step S307 by the keyword priority phoneme selection unit 107 is substantially the same as the step S285 for setting the path with the minimum cost value in FIG.

  When the keyword priority phoneme selection unit 107 sets a path that minimizes the cost value (S307), the keyword priority phoneme selection unit 107 decrements the number of keywords by 1 (S309).

  As shown in FIG. 9, in step S309 by the keyword priority phoneme selection unit 107, when there are a plurality of keywords in the text body, the phoneme selection processing between the keywords is completed by referring to the number of keywords. Used to express what has been done. For example, when the number of keywords is 1, it indicates that the phoneme selection processing between the keywords is completed, but the present invention is not limited to this example.

  When the number of keywords is decremented by 1 in step S309, the keyword priority phoneme selection unit 107 sets the search variable to the next phoneme (S311). Note that step S311 is substantially the same as step S287 of FIG.

  As shown in FIG. 9, after executing step S299 in FIG. 8, the keyword priority phoneme selection unit 107 checks whether or not the keyword position information indicated by the search variable is true (S313).

  If the keyword position information pointed to by the search variable is true (S313), it is confirmed whether the keyword position information of the phoneme next to the search variable is the end of the sentence (S315).

  On the other hand, if the keyword position information pointed to by the search variable is false (S313), a process is performed to check whether the keyword position information of the phoneme next to the search variable shown in FIG. 10 is the end of the sentence (S320). .

  As described above, in step S315 shown in FIG. 9, it is determined whether or not the next phoneme of the phoneme indicated by the search variable is the end of the text body.

  When the keyword position information of the phoneme next to the search variable is the end of the sentence (S315), a waveform segment is output as shown in FIG. 10 (S329).

  On the other hand, if the keyword position information of the next phoneme after the search variable is not the end of the sentence (S315), the search variable is set to the next phoneme (S317). Note that step S317 is substantially the same as step S287 of FIG.

  When the search variable is set to the next phoneme (S317), it is confirmed again whether or not the keyword position information indicated by the search variable is true (S313). Subsequent processing is as described above.

  The process (S320) for confirming whether or not the keyword position information of the phoneme next to the search variable performed by the keyword priority phoneme selection unit 107 shown in FIG. 10 is the end of the sentence is substantially the same as step S315 of FIG. .

  As a result of the confirmation (S320), if the keyword position information of the phoneme next to the search variable is the end of the sentence, next, a path with the minimum cost value is set (S327). The process for setting the path (S327) is substantially the same as step S285 shown in FIG.

  When the path with the minimum cost value is set (S327), a waveform segment is output (S329).

  On the other hand, if the keyword position information of the phoneme next to the search variable is not the end of the sentence (S320), the keyword priority phoneme selection unit 107 obtains the cost value of the phoneme pointed to by the search variable and the next phoneme (S323). Note that step S323 is substantially the same as step S281 in FIG.

  When the search variable is set to the next phoneme (S325), a process is performed again to check whether the keyword position information of the next phoneme after the search variable is the end of the sentence (S320). The subsequent processing is as described above.

  The process (S329) for outputting the waveform segment shown in FIG. 10 outputs the waveform segment obtained by selecting the phoneme for the text body.

  When a waveform segment is output by the keyword priority phoneme selection unit 107 (S329), the phoneme connection unit 111 connects the waveform segments selected by the phoneme and outputs them as synthesized speech. When the synthesized speech is output from an output unit such as a speaker, the speech synthesizer 100 can read out the text while emphasizing the keyword.

Although the description of the speech synthesizer 100 according to the first embodiment is finished as described above, the speech synthesizer 100 has the following excellent effects.
(1) The keywords included in the text body can be read out more smoothly than the parts other than the keyword, the sound quality is better than the parts other than the keyword, the reading out of the keyword part is more emphasized, and can be made more prominent. It is possible to convey the keyword part more clearly to the viewer.

(About voice synthesizer)
Next, a speech synthesizer 900 according to the second embodiment will be described with reference to FIG. FIG. 11 is a block diagram illustrating a schematic configuration of the speech synthesizer according to the second embodiment. Hereinafter, differences from the first embodiment will be described in detail, but the other points are substantially the same, and detailed description thereof will be omitted.

  As shown in FIG. 11, the speech synthesizer 900 includes a text analysis unit 101, a prosody prediction unit 103, a keyword weighting unit 901, a keyword extraction unit 905, a keyword priority phoneme selection unit 907, a corpus 109, and a phoneme. And a connecting portion 111.

  Note that the speech synthesizer 900 according to the second embodiment is an apparatus capable of outputting synthesized speech based on the text body and keywords, as with the speech synthesizer 100 according to the first embodiment. However, it is a device that can read out the text body as speech by outputting the synthesized speech.

  More specifically, the speech synthesizer 900 exemplifies, for example, a PC having an input unit corresponding to a CPU, memory, HDD (hard disk drive), mouse, etc., a display unit equivalent to a liquid crystal display, etc. However, it is not limited to such an example.

  Note that the display unit provided in the speech synthesizer 900 according to the present embodiment outputs display screen data and audio data processed so as to be displayed by the CPU. In addition, the display unit is exemplified by a TV or a liquid crystal display device, for example, and both of them are provided with a speaker, and can output a sound or a moving image in addition to a still image.

  The input unit is, for example, a pointing device such as a mouse, a trackball, a trackpad, a stylus pen, or a joystick that can receive an operation instruction from a user, an operation means such as a keyboard, a button, a switch, and a lever, and an input It comprises an input control circuit that generates a signal and outputs it to the CPU. The user of the speech synthesizer 900 can input various data and instruct processing operations to the speech synthesizer 900 by operating this input unit.

  The keyword weighting unit 901 shown in FIG. 11 determines the degree of emphasis for reading out a plurality of keyword parts in the text body, that is, the weight coefficient for narrowing down the combination of unit candidates by the keyword priority phoneme selection unit 907, Stored as keyword weighting information.

  The keyword extraction unit 905 determines the keyword weighting information obtained by the keyword weighting unit 901 based on the part where the phoneme symbol of the keyword and the phoneme symbol in the text body partially match. The combination of unit candidates corresponding to is changed to a width value for narrowing down. The changed keyword strength information is included in the head phoneme of the keyword in the prosodic prediction information.

  The keyword extraction unit 905 uses a table in which keyword weighting information indicating the degree of emphasis for reading out the keyword portion and a width for narrowing down the combination of unit candidates are associated with each other. The table is stored in a storage unit such as an HDD provided in the speech synthesizer 900, for example.

  The keyword priority phoneme selection unit 907 is based on the phoneme candidate obtained by the target selection, from the keyword start position to the keyword end position where the flag value of the keyword position information given to each phoneme of the text body is true, The cost value that falls within the range of narrowing the combination of unit candidates from the value that minimizes the cost value that is the sum of the target cost value, the sub cost value related to pitch discontinuity, and the sub cost value related to spectrum discontinuity. The combination of unit candidates is selected using dynamic programming.

(Speech synthesis method)
Next, a speech synthesis method according to the second embodiment will be described with reference to FIGS. FIGS. 12 and 13 are flowcharts showing an outline of the speech synthesis method according to the second embodiment.

  As shown in FIGS. 12 and 13, the speech synthesis method according to the second embodiment is different from the speech synthesis method according to the first embodiment in keyword weighting processing (S1201) and keyword strength information. The processing is different in that it includes initialization processing (S1203), and the keyword location extraction processing (S1205) and the keyword priority phoneme selection processing (S1207) are the processing and processing according to the first embodiment. The contents are different, but details will be described later.

  First, as shown in FIG. 12, a text body including one or more keywords and one or more keywords to be emphasized are input to the text analysis unit 101 (S201). The text body and the keyword will be described by taking an example where the text body and the keyword are expressed in kanji characters, but are not limited to such examples.

  Next, the text analysis unit 101 converts the input text body and keywords into phonological symbols (S203).

  The text analysis unit 101 performs morphological analysis and dependency analysis on the text body expressed by kanji characters, and outputs an accent symbol string and a morphological analysis result representing the part-of-speech information of the text body (S203).

  As shown in FIG. 12, an output result having information on phonological symbols converted to a text body and a morphological analysis result is defined as, for example, a text body intermediate language. Not.

  Next, as shown in FIG. 12, the text analysis unit 101 determines whether the text body is a text body in order to input the text body to the prosody prediction unit 103 (S205).

  As a result of the determination (S205), if it is a text body, the text analysis unit 101 outputs the text body intermediate language and transmits it to the keyword extraction unit 905 together with the prosodic prediction unit 103.

  On the other hand, when it is not a text body, that is, when it is a keyword, the text analysis unit 101 transmits the keyword intermediate language to the keyword weighting unit 901.

Next, the prosody prediction unit 103 selects at least one or all of pitch (voice pitch: fundamental frequency F ₀ ), phoneme duration (voice length), or mel cepstrum representing a waveform component. Prediction is performed (S207).

  The prosody prediction unit 103 outputs prosodic prediction information to the keyword extraction unit 901 as predicted information.

  Also, as shown in FIG. 12, when the keyword weighting unit 901 receives the keyword intermediate language from the text analysis unit 101 (S205), the degree of emphasis when reading a plurality of keyword parts in the text body, that is, The keyword priority phoneme selection unit 907 determines a weighting factor for narrowing down the combination of unit candidates and holds it as keyword weighting information (S1201).

  Next, the keyword weighting unit 901 determines a weighting factor and generates keyword weighting information (S1201), and outputs the keyword intermediate language and the keyword weighting information to the keyword extraction unit 905.

(About processing by the keyword extraction unit)
Next, as shown in FIG. 12, the keyword extraction unit 905 describes information (keyword position information) for indicating where the keyword exists in the text body, so that each phoneme in the prosodic prediction information is recorded. An area is secured for and initialized (S209). In the initialization, the flag value of the keyword position information is set to false, but the present invention is not limited to this example.

  Next, the keyword extraction unit 905 initializes keyword strength information expressing a width for narrowing down the combination of unit candidates (S1203).

  In step S1203, keyword strength information is initialized for each phoneme in the prosodic prediction information, which expresses the range for narrowing the combination of unit candidates. The initialization will be described by taking, for example, a case where the value of keyword strength information is set to 0, but is not limited to such an example.

  Next, the keyword extraction unit 905 converts a phoneme position indicator representing which phoneme is currently pointed to in the phoneme symbol between the text body and the keyword to the text body and the first phoneme in one or more keywords. Set (S211).

  Next, the keyword extraction unit 905 obtains the number of phonemes for the text body and a plurality of keywords, and sets the number of keyword phonemes for each keyword phoneme number and the number of text body phonemes for the text phoneme number (S213). ).

  Next, the keyword extraction unit 905 determines whether the end of the text body is reached (S215).

  As a result of determining whether or not it is the end of the text body (S215), if it is not the end of the text body, the keyword extracting unit 905 extracts the location of the keyword in the text body (S217).

  After the keyword location extraction process (S217) is executed, the first phoneme of the keyword in the prosodic prediction information is based on the location where the keyword phoneme symbol partially matches the phoneme symbol in the text body. In order to obtain the keyword strength information provided in the keyword, the keyword weighting information indicating the degree of strength at which the keyword portion is read out and the width for narrowing down the combination of unit candidates are used in the keyword strength information by using a table in which the combinations are narrowed down. Is changed to a width value for narrowing down the combination of unit candidates corresponding to the keyword weighting information obtained by the keyword weighting unit 901 (S2105).

  The keyword extraction unit 905 outputs to the keyword priority phoneme selection unit 907 “prosody prediction information with keyword position information and strength information (keyword strength information)” composed of prosodic prediction information with keyword position information and keyword strength information. To do.

  Next, when it is confirmed that it is the end of the text body (S215), keyword priority phoneme selection processing is executed as shown in FIG. 13 (S1219).

  In the keyword priority phoneme selection process (S1219), based on the phoneme candidate obtained by the target selection, the keyword end position is changed from the keyword start position where the flag value of the keyword position information assigned to each phoneme of the text body is true. Until the target cost value, the sub cost value related to the pitch discontinuity, and the sub cost value related to the spectrum discontinuity are combined, the unit between the widths for narrowing the combination of unit candidates from the value that minimizes the cost value. Candidate combinations are selected using dynamic programming.

  In step S1219, when there are two or more keywords, the cost value for the keyword portion may be calculated in the order in which the keywords appear from the start to the end of the text body.

  If there are two or more keywords in step S1219, the cost value calculation method other than the keyword portion is such that the cost value is the minimum from the start position of the text body to the first keyword start position. A combination of unit candidates may be selected using dynamic programming.

  In the above step S1219, when there are two or more keywords, the cost value calculation method other than the keyword part uses the combination of unit candidates that minimizes the cost value of the keyword part (the relevant keyword ) From the end position to the start position of the next appearing keyword (subsequent keyword), a combination of unit candidates that minimizes the cost value may be selected using dynamic programming.

  In the above step S1219, when there are two or more keywords, the cost value calculation method other than the keyword portion is such that the cost value is minimized from the end position of the keyword that appears last to the end position of the text body. A combination of unit candidates may be selected using dynamic programming.

  As described above, as shown in FIG. 13, all the phoneme selection processes are performed by the keyword priority phoneme selection unit 107, so that a waveform segment can be output (S1219).

  Next, the phoneme connection unit 111 determines whether the phoneme of the waveform segment currently handled is the end of the text body (S221).

  As a result of the above determination (S221), if it is not the end of the text body, the phoneme connection unit 111 connects the current waveform segment to the next waveform segment (S223).

  On the other hand, if it is the end of the text body (S221), the phoneme connection unit 111 outputs synthesized speech generated by connecting the waveform segments (S225). With the output of the synthesized speech (S225), the speech synthesizer 900 can read the text body while emphasizing the keyword.

(Keyword location extraction process)
Next, the keyword location extraction processing (S215, S217, S2105) according to the second embodiment shown in FIG. 12 will be described in more detail with reference to FIGS. FIGS. 14 and 15 are flowcharts showing an outline of the keyword location extraction processing according to the second embodiment.

  In the keyword location extraction process (S217), as described above, the keyword extraction unit 905 previously represents a phoneme position indicator that indicates which phoneme is currently pointed out in the phoneme symbol between the text body and the keyword. Is set as the head phoneme for the text body and multiple keywords.

  First, as shown in FIG. 14, if it is not the end of the text body (S215), the keyword extraction unit 905 handles keywords in order from one or more keywords. , It is determined whether or not the total number of keywords falls within the range (S240).

  The order of keywords currently handled is the keyword input order, but is not limited to this example.

  Next, the keyword extraction unit 905 determines whether or not the phonological position indicator of the currently handled keyword (this keyword) indicates the ending of the keyword (S241).

  If it is the ending of the keyword (S241), a process for obtaining a value for narrowing down the combination of unit candidates corresponding to the weight of the keyword is performed (S1229 shown in FIG. 15).

  If it is not the end of the keyword (S241), it is determined whether the text body indicated by the phoneme position indicator matches the phoneme of the keyword (S243).

  Note that the criterion for determining whether or not the ending of the keyword according to the second embodiment is the same as the criterion according to the first embodiment, and the phoneme of the currently handled keyword (this keyword). Judgment is made based on whether the position indicator is smaller than the number of phonemes of the keyword.

  When the phonemes match (S243), the phoneme position indicator of the keyword is set as the next phoneme (S247). When the phonemes do not match (S243), the phoneme position indicator of the keyword is set as the first phoneme. (S245).

  Next, the keyword extraction unit 905 sets the position pointed to by the phoneme position indicator of the currently handled keyword (this keyword) as the first phoneme of the keyword (S245).

  Next, the keyword extraction unit 905 sets the position pointed to by the phoneme position indicator of the currently handled keyword (this keyword) as the phoneme next to the phoneme currently pointed by the keyword position indicator (S247). ).

  Next, the keyword extraction unit 905 changes the currently handled keyword to the next keyword in the keyword input order (S249).

  Note that if it is the end of the text body (S215), the keyword extraction unit 905 outputs keyword position information and strong and weak prosodic prediction information (S251).

  Next, as shown in FIG. 15, in step S <b> 1229 by the keyword extraction unit 905, it corresponds to the keyword weighting information indicating the degree of emphasis for reading the keyword portion held by the currently handled keyword (this keyword). Then, a process of acquiring values for narrowing down the combination of unit candidates from a table in which keyword weighting information indicating the degree of emphasis for reading out the keyword portion and a width for narrowing down the combination of unit candidates is associated.

  Next, the keyword extraction unit 905 corresponds to all the phonemes for the number of the keyword phonemes from the phoneme pointed by the phoneme position indicator in the text body toward the head phoneme. The keyword position information for each phoneme in the prosodic prediction information is changed (S253). The keyword position information is changed from false to true, but is not limited to this example.

  After executing the processing of step S253, the keyword extraction unit 905 next changes the value of the keyword strength information included in the head phoneme of the keyword in the text body to a value that narrows down the combination of unit candidates (S1300).

  In step S1300, a process of changing the value of the keyword strength information included in the first phoneme of the keyword in the text body to a width value for narrowing down the combination of unit candidates acquired in step S1229 is performed.

  Next, as shown in FIG. 15, the keyword extraction unit 905 executes the same processing as step S245 shown in FIG. 14 (S255).

  If the determination result shown in FIG. 14 is not within the number of keywords (S240), the keyword extraction unit 905 changes the keyword to the first keyword as shown in FIG. 15 (S257).

  Next, the keyword extraction unit 905 sets the position indicated by the phoneme position indicator in the text body to the phoneme next to the phoneme currently indicated by the phoneme position indicator in the text body (S259).

  Thus, a series of processing of keyword part extraction processing (S215, S217, S2105) by the keyword extraction unit 905 according to the second embodiment shown in FIGS. 14 and 15 is completed.

(Keyword priority phoneme selection processing by the keyword priority phoneme selection unit 907)
Next, the keyword priority phoneme selection process (S1219) according to the second embodiment will be described in detail with reference to FIGS. FIGS. 16 to 20 are flowcharts showing an outline of the keyword priority phoneme selection process according to the second embodiment.

  First, as shown in FIG. 16, the keyword priority phoneme selection unit 907 selects a target for the phoneme of the text body (S271).

  When the target is selected (S271), the keyword priority phoneme selection unit 907 sets the search variable to the first phoneme of the text body and sets the number of keywords to 0 (S273), thereby determining which phoneme is currently indicated in the text body. The expressed search variable points to the phoneme at the beginning of the text body.

  Next, the keyword priority phoneme selection unit 907 determines whether or not the position of the phoneme pointed to by the search variable is within the range of the total number of phonemes in the text body (S275). That is, as shown in step S275 in FIG. 16, it is determined whether or not the search variable <the number of phonemes of the text.

  If the result of determination in step S275 is that the search variable is smaller than the number of phonemes in the text, the keyword priority phoneme selection unit 907 determines whether the phoneme pointed to by the search variable is a phoneme in the keyword (S277).

  Normally, when the phoneme pointed to by the search variable is a phoneme in a keyword, the keyword position information is true. If it is not a phoneme within the keyword, the keyword position information is false.

  If the search variable is larger than the number of phonemes of the text (S275), the subsequent processing shown in FIG. 18 (set the search variable to the first phoneme of the text body (S290)) is executed.

  If the keyword position information pointed to by the search variable is false (S277), the search variable is set to the next phoneme as shown in FIG. 17 (S287).

  If the keyword position information pointed to by the search variable is true, as shown in FIG. 17, it is determined whether or not the keyword position information pointed to by the search variable is true and the keyword position information of the phoneme preceding the search variable is false. Processing to confirm is executed (S 1333).

  In step S1333 shown in FIG. 17, it is determined whether or not it is the first phoneme of the keyword in the text body. When the keyword position information indicated by the search variable is true and the keyword position information of the phoneme preceding the search variable is false, the search variable indicates the head phoneme of the keyword.

  Next, when the keyword position information pointed to by the search variable is true and the keyword position information of the phoneme preceding the search variable is false (S 1333), the keyword strength information included in the head phoneme of the keyword pointed to by the search variable is obtained. Obtain (S1334).

  On the other hand, if the keyword position information pointed to by the search variable is true and the keyword position information of the phoneme preceding the search variable is not false (S 1333), the keyword position information of the phoneme next to the search variable is true or It is confirmed whether there is a phoneme (S279).

  In step S1334, the value of keyword strength information expressing the range for narrowing down the combination of unit candidates included in the head phoneme of the keyword indicated by the search variable is acquired. If the value of the keyword strength information is acquired (S1334), next, step S279 is executed.

  As shown in FIG. 17, in step S279, it is determined whether the phoneme next to the phoneme indicated by the search variable is a phoneme in the keyword.

  If the keyword position information of the phoneme next to the search variable is true, the cost value of the phoneme pointed to by the search variable and the next phoneme is obtained (S281).

  If the keyword position information of the phoneme next to the search variable is false, the number of keywords is incremented by 1 (S283).

  As shown in FIG. 17, in step S281, between the candidate obtained by the target selection of the phoneme indicated by the search variable and the candidate obtained by the target selection of the phoneme next to the phoneme indicated by the search variable. The cost value obtained by adding the target cost value, the sub cost value related to the pitch discontinuity, and the sub cost value related to the spectrum discontinuity is obtained.

  When the cost value is obtained (S281), the search variable is set to the next phoneme (S287).

  When the number of keywords expressing how many keywords are present in the text body is incremented by 1 (S283), the keyword priority phoneme selection unit 907 falls within the range of the value of (cost value + keyword strength information). A path having a value is set (S1335).

  In step S1335, the target cost value between the candidate obtained by target selection of the phoneme indicated by the search variable and the candidate obtained by target selection of the phoneme next to the phoneme indicated by the search variable is , A cost that falls within a range of values obtained by adding the keyword cost information acquired in step S1334 to the minimum cost value obtained by adding the sub cost value related to the pitch discontinuity and the sub cost value related to the spectrum discontinuity. The unit candidate combination (path) having a value is obtained.

  When the above path is set (S1335), the search variable is set to the next phoneme (S287).

  As shown in FIG. 18, the keyword priority phoneme selection unit 907 sets the search variable to the first phoneme of the text body (S290). Note that step S290 is substantially the same as step S273 in FIG.

  When the search variable is set to the head phoneme of the text body (S290), it is confirmed whether or not the keyword position information pointed to by the search variable is false (S291). Note that step S291 is substantially the same as step S277 in FIG.

  When the keyword position information indicated by the search variable is false (S291), it is confirmed whether the keyword position information of the phoneme next to the search variable is false (S293).

  On the other hand, if the keyword position information pointed to by the search variable is true (S291), it is confirmed whether the keyword position information pointed to by the search variable is true (S298).

  As shown in FIG. 18, in step S293, it is determined whether or not the next phoneme of the phoneme indicated by the search variable is a phoneme in the keyword.

  Next, when the keyword position information of the phoneme next to the search variable is false (S293), the cost value of the phoneme pointed to by the search variable and the next phoneme is obtained (S295). Note that step S295 is substantially the same as the process (S281) for obtaining the cost value in FIG.

  If the keyword position information of the phoneme next to the search variable is true (S293), a path with the minimum cost value is set (S296). Note that step S296 is substantially the same as the process (S285) for setting the minimum path in FIG.

  After obtaining the cost value of the phoneme indicated by the search variable and the next phoneme, the search variable is set to the next phoneme (S297). Note that step S297 and the processing for setting the next phoneme in FIG. 7 (S287) are substantially the same.

  Next, as shown in FIG. 18, the keyword priority phoneme selection unit 907 checks whether or not the keyword position information indicated by the search variable is false (S298). Note that step S298 is substantially the same as the process (S277) for confirming whether or not the keyword position information pointed to by the search variable in FIG. 16 is false.

  As a result of the confirmation (S298), if the keyword position information indicated by the search variable is true, it is confirmed whether or not the result of decrementing (subtracting) the number of keywords by 1 is 1 or more (S299).

  On the other hand, if the keyword position information indicated by the search variable is false as a result of the confirmation (S298), the subsequent processing (S301) shown in FIG. 19 is executed.

  As shown in FIG. 18, in step S299, if there are a plurality of keywords in the text body, and there is one or more phonemes between the keywords, the search variable is set to the next phoneme (S300). ). Step S300 is substantially the same as step S287 in FIG.

  Next, as shown in FIG. 19, the keyword priority phoneme selection unit 907 first checks whether or not the keyword position information indicated by the search variable is false (S301), and then the keyword position information indicated by the search variable is If it is false, it is confirmed whether or not the keyword position information of the phoneme next to the search variable is false (S303). Step S303 is substantially the same as step S293 in FIG.

  If the keyword position information of the phoneme next to the search variable is false (S303), the cost value of the phoneme pointed to by the search variable and the next phoneme is obtained (S305).

  On the other hand, if the keyword position information of the phoneme next to the search variable is true, a path with the minimum cost value is set (S307).

  Next, as shown in FIG. 19, the above-described step S307 by the keyword priority phoneme selection unit 907 is substantially the same as step S285 of setting the path with the minimum cost value in FIG.

  When the keyword priority phoneme selection unit 907 sets a path that minimizes the cost value (S307), the keyword priority phoneme selection unit 907 decrements the number of keywords by 1 (S309).

  As shown in FIG. 19, in step S309 by the keyword priority phoneme selection unit 907, when there are a plurality of keywords in the text body, the phoneme selection process between the keywords is completed by referring to the number of keywords. Used to express what has been done. For example, when the number of keywords is 1, it indicates that the phoneme selection processing between the keywords is completed, but the present invention is not limited to this example.

  When the number of keywords is decremented by 1 in step S309, the keyword priority phoneme selection unit 907 sets the search variable to the next phoneme (S311). Note that step S311 is substantially the same as step S287 of FIG.

  As shown in FIG. 19, after executing step S299 in FIG. 18, the keyword priority phoneme selection unit 907 checks whether or not the keyword position information indicated by the search variable is true (S313).

  If the keyword position information pointed to by the search variable is true (S313), it is confirmed whether the keyword position information of the phoneme next to the search variable is the end of the sentence (S315).

  On the other hand, if the keyword position information pointed to by the search variable is false (S313), a process is performed to check whether the keyword position information of the phoneme next to the search variable shown in FIG. 20 is the end of the sentence (S320). .

  As described above, in step S315 shown in FIG. 19, it is determined whether or not the next phoneme of the phoneme indicated by the search variable is the end of the text body.

  If the keyword position information of the phoneme next to the search variable is the end of the sentence (S315), a waveform segment is output as shown in FIG. 20 (S329).

  On the other hand, if the keyword position information of the next phoneme after the search variable is not the end of the sentence (S315), the search variable is set to the next phoneme (S317). Note that step S317 is substantially the same as step S287 of FIG.

  When the search variable is set to the next phoneme (S317), it is confirmed again whether or not the keyword position information indicated by the search variable is true (S313). Subsequent processing is as described above.

  The process (S320) for confirming whether or not the keyword position information of the phoneme next to the search variable performed by the keyword priority phoneme selection unit 907 shown in FIG. 20 is the end of the sentence is substantially the same as step S315 of FIG. .

  As a result of the confirmation (S320), if the keyword position information of the phoneme next to the search variable is the end of the sentence, next, a path with the minimum cost value is set (S327). The process for setting the path (S327) is substantially the same as step S285 shown in FIG.

  When the path with the minimum cost value is set (S327), a waveform segment is output (S329).

  On the other hand, if the keyword position information of the phoneme next to the search variable is not the end of the sentence (S320), the keyword priority phoneme selection unit 907 obtains the cost value of the phoneme pointed to by the search variable and the next phoneme (S323). Note that step S323 is substantially the same as step S281 in FIG.

  When the search variable is set to the next phoneme (S325), a process is performed again to check whether the keyword position information of the next phoneme after the search variable is the end of the sentence (S320). The subsequent processing is as described above.

  The process of outputting the waveform segment shown in FIG. 20 (S329) outputs the waveform segment obtained by selecting the phoneme for the text body.

  When a waveform segment is output by the keyword priority phoneme selection unit 907 (S329), the phoneme connection unit 111 connects the waveform segments selected by the phoneme and outputs them as synthesized speech. When the synthesized speech is output from an output unit such as a speaker, the speech synthesizer 100 can read out the text while emphasizing the keyword.

Although the description of the speech synthesizer 900 according to the second embodiment is completed as described above, the speech synthesizer 900 has the following excellent effects.
(1) The degree to which the reading of a keyword is emphasized, that is, when the keyword part is emphasized, the connection with the front and back is smoothed, and by adjusting the degree of emphasis of the keyword part, Can read aloud.

  The series of processes described above can be performed by dedicated hardware or software. When a series of processing is performed by software, a program constituting the software is installed in an information processing apparatus such as a general-purpose computer or a microcomputer, and functions as the speech synthesizer 100 and the speech synthesizer 900.

  The program can be recorded in advance on a hard disk or ROM as a recording medium built in the computer so as to be executable.

  Alternatively, the program is not limited to a hard disk drive, but a removable recording medium such as a flexible disk, CD-ROM (Compact Disc Read Only Memory), MO (Magneto Optical) disk, DVD (Digital Versatile Disc), magnetic disk, and semiconductor memory. In addition, it can be stored (recorded) temporarily or permanently. Such a removable recording medium can be provided as so-called package software.

  The program is installed on the computer from the removable recording medium as described above, and is transferred from the download site to the computer wirelessly via a digital satellite broadcasting artificial satellite, or a LAN (Local Area Network) or the Internet. Such a program can be transferred to a computer via a network, and the computer can install the program transferred in this way in a storage means such as a built-in hard disk.

  Here, in this specification, the processing steps for describing a program for causing a computer to perform various processes do not necessarily have to be processed in time series in the order described in the flowchart, but in parallel or individually. This includes processing to be executed (for example, parallel processing or processing by an object).

  The program may be processed by one computer, or may be distributedly processed by a plurality of computers.

  As mentioned above, although preferred embodiment of this invention was described referring an accompanying drawing, this invention is not limited to this example. It is obvious for a person skilled in the art that various changes or modifications can be envisaged within the scope of the technical idea described in the claims, and these are naturally within the technical scope of the present invention. It is understood that it belongs.

  In the above embodiment, each unit (text analysis unit 101 to keyword priority phoneme selection unit 907) included in the speech synthesizer 100 and the speech synthesizer 900 has been described as an example of hardware, but the present invention is applied. It is not limited to examples. For example, at least one of the above units may be a program composed of one or two or more modules or components.

1 is a block diagram showing a schematic configuration of a speech synthesizer according to a first embodiment. It is a flowchart which shows the outline of the speech synthesis method concerning 1st Embodiment. It is a flowchart which shows the outline of the extraction process of the keyword location concerning 1st Embodiment. It is a flowchart which shows the outline of the extraction process of the keyword location concerning 1st Embodiment. It is explanatory drawing which shows the outline of the criteria of judgment whether it is the ending of the keyword concerning 1st Embodiment. It is explanatory drawing which shows schematic structure of the various information concerning this Embodiment. It is explanatory drawing which shows roughly the flow of the data in the speech synthesizer concerning 1st Embodiment. It is explanatory drawing which shows the outline of the keyword priority phoneme selection process concerning 1st Embodiment. It is a flowchart which shows the outline of the keyword priority phoneme selection process concerning 1st Embodiment. It is a flowchart which shows the outline of the keyword priority phoneme selection process concerning 1st Embodiment. It is a flowchart which shows the outline of the keyword priority phoneme selection process concerning 1st Embodiment. It is a flowchart which shows the outline of the keyword priority phoneme selection process concerning 1st Embodiment. It is a block diagram which shows schematic structure of the speech synthesizer concerning 2nd Embodiment. It is a flowchart which shows the outline of the speech synthesis method concerning 2nd Embodiment. It is a flowchart which shows the outline of the speech synthesis method concerning 2nd Embodiment. It is a flowchart which shows the outline of the extraction process of the keyword location concerning 2nd Embodiment. It is a flowchart which shows the outline of the extraction process of the keyword location concerning 2nd Embodiment. It is a flowchart which shows the outline of the keyword priority phoneme selection process concerning 2nd Embodiment. It is a flowchart which shows the outline of the keyword priority phoneme selection process concerning 2nd Embodiment. It is a flowchart which shows the outline of the keyword priority phoneme selection process concerning 2nd Embodiment. It is a flowchart which shows the outline of the keyword priority phoneme selection process concerning 2nd Embodiment. It is a flowchart which shows the outline of the keyword priority phoneme selection process concerning 2nd Embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 Speech synthesizer 101 Text analysis part 103 Prosody prediction part 105,905 Keyword extraction part 107,907 Keyword priority phoneme selection part 109 Corpus 111 Phoneme connection part 901 Keyword weighting part

Claims (17)

A speech synthesis method that synthesizes speech by emphasizing a keyword part:
A selection process is performed in which candidate phonemes of all keywords included in the text body are selected from a corpus in order of appearance of the keywords in preference to phoneme candidates other than the keywords. Speech synthesis method. The selection process uses a combination of unit candidates that minimizes the cost value of the keyword part, and selects a unit candidate that has a minimum cost value from the start position of the text body to the first keyword start position. Select a combination;
The combination of unit candidates that minimizes the cost value is selected from the end position of the keyword toward the start position of the subsequent keyword when there are two or more keywords. The speech synthesis method described. In the speech synthesis method, before the selection process, a keyword extraction process is performed to search from the head of the phoneme symbol if the phoneme symbol of the keyword partially matches the phoneme symbol of the text body. The speech synthesis method according to claim 1 or 2. The keyword extraction process changes a value of keyword position information for each phoneme described in the prosodic prediction information based on a part where the phoneme symbol of the keyword and the phoneme symbol of the text body partially match;
Based on the prosodic prediction information including the keyword position information after the change, the selection process selects phoneme candidates of all keywords included in the text sentence from the corpus, The speech synthesis method according to claim 1. By changing the value of the keyword position information of each phoneme described in the prosodic prediction information, which is information for predicting at least one of voice pitch, voice length, and mel cepstrum, the phonological symbol of the keyword The speech synthesis method according to claim 1, wherein the phoneme symbol of the text body partially matches. In the speech synthesis method, a weighting process for adding a keyword weighting coefficient indicating a degree of emphasizing the keyword part to one or more keywords in the text body is further performed. Item 6. The speech synthesis method according to any one of Items 1 to 5. In the keyword extraction process, the keyword weighting coefficient assigned in the weighting process is acquired as keyword weighting information for one or more keywords in the text body, and the combination of the keyword weighting information and the unit candidate is acquired. The speech synthesis method according to claim 4, wherein a table in which a width for narrowing down is associated is used. In the selection process, the unit candidate combinations are narrowed down based on a value that minimizes a cost value obtained by adding the target cost width, the sub cost value related to the pitch discontinuity, and the sub cost value related to the spectrum discontinuity. 8. The speech synthesis method according to claim 7, wherein a combination of unit candidates having cost values that fall within the range of width is selected. A speech synthesizer that emphasizes keyword parts and synthesizes speech:
A keyword-preferred phoneme selection unit that selects phoneme candidates of all keywords included in a text body from a corpus in order of appearance of the keywords in preference to phoneme candidates other than the keyword. , Speech synthesizer. The keyword priority phoneme selection unit uses a combination of unit candidates that minimizes the cost value of the keyword part, and the cost value is minimized from the start position of the text body to the first keyword start position. Select a unit candidate combination;
The combination of unit candidates that minimizes the cost value is selected from the end position of the keyword toward the start position of the subsequent keyword when there are two or more keywords. The speech synthesizer described. The speech synthesizer further includes a keyword extracting unit that searches from the head of the phoneme symbol whether the phoneme symbol of the keyword partially matches the phoneme symbol of the text body. The speech synthesizer according to 9 or 10. The keyword extraction unit changes the value of the keyword position information for each phoneme described in the prosodic prediction information based on a location where the phoneme symbol of the keyword and the phoneme symbol of the text body partially match;
The keyword priority phoneme selection unit selects, from the corpus, phoneme candidates of all keywords included in the text sentence based on the prosodic prediction information including the changed keyword position information. The speech synthesizer according to any one of claims 9 to 11. The keyword extraction unit changes the value of the keyword position information of each phoneme described in the prosody prediction information, which is information for predicting at least one of voice pitch, voice length, and mel cepstrum. 13. The speech according to claim 9, wherein the changed value of the keyword position information indicates that the phonological symbol of the keyword and the phonological symbol of the text body partially match. Synthesizer. 10. The speech synthesizer further includes a weighting unit that assigns a keyword weighting coefficient indicating a degree of emphasizing the keyword part to one or more keywords in the text body. The speech synthesizer in any one of -13. The keyword extraction unit acquires, as keyword weighting information, a keyword weighting coefficient assigned by the weighting unit for one or more keywords in the text body, and combines the keyword weighting information and the unit candidate. The speech synthesizer according to claim 12, wherein a table in which a narrowing width is associated is used. The keyword-preferred phoneme selection unit selects the unit candidate based on a value that minimizes a cost value obtained by adding a target cost range, a sub-cost value related to pitch discontinuity, and a sub-cost value related to spectrum discontinuity. 16. The speech synthesizer according to claim 15, wherein a combination of unit candidates having a cost value that falls within a range for narrowing down combinations is selected. The computer program according to any one of claims 9 to 16, wherein the computer is caused to function as a speech synthesizer for emphasizing a keyword portion and performing speech synthesis.

Images