JP4856560B2 - Speech synthesizer - Google Patents

Description

  The present invention relates to a speech synthesizer, and more particularly to a device capable of obtaining a desired synthesized speech without a user setting parameter values necessary for speech synthesis.

  A conventional speech synthesizer will be described. The speech synthesizer 100, which is one of the conventional speech synthesizers, can easily perform an operation of obtaining desired speech synthesis data by giving appropriate parameters to the speech synthesis engine.

  When the speech synthesizer 100 acquires a kanji kanji character string from the user via the keyboard, the speech synthesizer 100 automatically determines a parameter value necessary for speech synthesis from the acquired kanji magic character string, and based on the determined parameter value. To generate synthesized speech. Here, the parameters include the height of the accent and the accent position.

  In the speech synthesizer 100, the generated synthesized speech can be corrected. The speech synthesizer 100 visually displays the current parameter setting values on the display as shown in FIG. 16 so that the user can easily modify the parameters of the generated synthesized speech. In FIG. 16, the setting value of the accent which is one of the parameters is visually displayed on the display. The accent given to each kana character "A", "RA", "SH", "MA", "CHO", "-", "NO" that composes the kana character string "ARASHIMA CHOO" Correspondingly, each kana character is arranged in the vertical position. In this figure, a kana character other than the kana character “a” is given a relatively higher accent than the kana character “a”.

  Here, if the accent of the kana character “MA” is to be lowered, the kana character frame 72 is dragged downward by operating the mouse. In response to this, the speech synthesizer 100 changes the parameter value so as to lower the accent of the kana character “MA”. Then, as shown in FIG. 17, the speech synthesizer 100 moves the kana character “MA” kana character frame 72 downward and displays it. In this way, the user can easily edit the height of the accent, which is one of the parameters.

JP 2004-246129 A

  The speech synthesizer 100 described above has the following problems. In the speech synthesizer 100, the user can correct the synthesized speech by changing parameter values. That is, if the user gives appropriate parameters, the speech synthesizer 100 can provide the synthesized speech desired by the user. In other words, if the user cannot give an appropriate parameter value, the speech synthesizer 100 does not provide the synthesized speech desired by the user.

  In general, the user may have a specific image of the desired synthesized speech, but may not know which parameter is changed to what value to obtain the desired synthesized speech. Many. That is, there arises a problem that the user has to repeat the operation of confirming the synthesized speech after changing the parameter value setting.

  For example, when the user inputs “Nakayama” via the keyboard, the speech synthesizer 100 sets parameters for the kana character strings “NA”, “KA”, “YA”, “MA” constituting “NAKAYAMA”. As an example, the values of accents “low”, “high”, “high”, and “high” are automatically determined, and a synthesized speech of “Nakayama” is generated. In this example, the values of the accents “low”, “high”, “high”, “high” given to the kana characters “na”, “ka”, “ya”, “ma” by the speech synthesizer 100 are: Generally, it is an accent that the personal name “Nakayama” has.

  Here, it is assumed that the user inputs “Nakayama” assuming a place name in a certain region in Kinki. In this case, if the kana characters “na”, “ka”, “ya”, “ma” have no accent “low”, “high”, “low”, “low” values, Cannot obtain the desired synthesized speech. In other words, the user can set the accent value for each of the kana characters “na”, “ka”, “ya”, “ma” to “low”, “high”, “high”, “high” instead of “low” It is necessary to correct the accent value after recognizing “high”, “low”, and “low”.

  However, it is not easy for a user who is not a phonetic expert to grasp the position of accents and the size of accents of character strings such as words and phrases. Therefore, the user must change the setting of the accent position and value and repeat the operation of confirming the synthesized speech until the synthesized speech expected by the user is obtained.

  Therefore, an object of the present invention is to provide a speech synthesizer that allows a user to easily obtain a desired synthesized speech without setting / changing parameter values.

  Means for solving the problems relating to the present invention and effects of the present invention will be described below.

  In the speech synthesizer, speech synthesis program, and speech synthesis method according to the present invention, character string information representing a certain character string is acquired, a certain voice is acquired as voice information, and the voice information has the acquired voice information. Accents are extracted as accent information, and based on the accent information and the character string information, synthesized speech corresponding to the character string and having the accent is generated.

  Thereby, the synthetic | combination audio | voice corresponding to a character string can be produced | generated based on the accent which the user input with the audio | voice. Therefore, the user can easily obtain a synthesized speech having an accent assumed by the user.

  In the speech synthesizer or speech synthesis program according to the present invention, the accent information is further extracted using a fundamental frequency function representing a temporal change in the fundamental frequency of the obtained speech information. Thereby, an accent can be determined easily.

  In the speech synthesizer or the speech synthesis program according to the present invention, the character string represented by the acquired character string information and the voice information are associated with each other, and in which syllable the syllable constituting the character string has an accent. Is extracted as the accent information, and a synthesized speech corresponding to the character string in the character string information having an accent in the syllable determined to have an accent is generated. . Thereby, the syllable in which an accent exists can be specified easily.

  In the speech synthesizer or speech synthesis program according to the present invention, the feature amount of the generated synthesized speech is changed, a synthesized speech having the changed feature amount is generated, and the synthesized speech in which the feature amount is changed and before the change is made Are displayed on the display means as synthesized speech candidates, and the displayed synthesized speech candidates are selectable by the input means, and any one of the synthesized speech candidates displayed on the display means is selected. If it is determined that the synthesized speech candidate is synthesized speech, the synthesized speech candidate is determined to be synthesized speech.

  Thereby, a plurality of synthesized speech candidates can be provided. Therefore, the user can obtain synthesized speech by an easy operation of selecting from the provided synthesized speech candidates.

  In the speech synthesizer or speech synthesis program according to the present invention, the feature amount includes at least one of a pitch and a speed of sound. This makes it possible to easily obtain a synthesized speech candidate in which the pitch or / or speed of the sound is changed.

  In the speech synthesizer or the speech synthesis program according to the present invention, when the feature amount is configured by the pitch and speed of the sound, the synthesized speech candidates are arranged on a plane with the pitch and speed of the sound as two axes. .

  Thereby, the user can grasp | ascertain easily the correlation of the synthetic speech candidate provided. Therefore, the user can easily select a desired one from the synthesized speech candidates.

  In the speech synthesizer or the speech synthesis program according to the present invention, when it is determined that any of the synthesized speech candidates displayed on the display unit is selected, the synthesized speech candidate is reproduced and synthesized with the reproduced synthesized speech candidate. When the confirmed information to be confirmed as voice information is acquired, the synthesized speech candidate is determined as synthesized speech.

  Thereby, the user can select the synthesized speech candidate after confirming the reproduced sound of the synthesized speech candidate.

  Here, the correspondence relationship between the elements described in the claims and the elements in the embodiment is shown. The speech synthesizer corresponds to the speech synthesizer 21. The voice information acquisition means is the CPU 211, the sound circuit 218, the microphone 218m, the accent information extraction means is the CPU 211 and the memory 212, the voice synthesis means is the CPU 211 and the memory 212, the character string information acquisition means is the CPU 211, the memory 212, the keyboard 214, and The modified synthesized speech generation means corresponds to the mouse 215, the CPU 211 and the memory 212, the synthesized speech candidate display means corresponds to the CPU 211 and the memory 212, and the synthesized speech determination means corresponds to the CPU 211 and the memory 212, respectively.

  The voice information acquisition means performs the processes of steps S501, S503, and S513, the accent information extraction means performs the processes of steps S515, S517, S801 to S815, and S901 to S913, and the voice synthesis means performs the processes of steps S519, S1101 to S1105. The character string information acquisition means performs the processing of steps S501, S503, and S511, the modified synthesized speech generation means performs the processing of steps S1001 to S1005, S1201 to S1205, and the synthesized speech candidate display means performs the processing of step S1207. The determining means executes the processes of steps S1209 to S1215, respectively.

  Accent information corresponds to accent position information.

  The “feature amount” refers to information that characterizes the voice, and is a concept including high / low voice, fast / slow voice speed, and the like.

  “Speech” refers to a sound that can be emitted directly or indirectly, and the subject that emits a sound is not limited to humans. Moreover, what is necessary is just a thing with an accent, and it is a concept including the thing which cannot understand the meaning content of the emitted voice, for example, humming etc.

  An embodiment of a speech synthesizer according to the present invention will be described below.

1. Overview An overview of a speech synthesizer according to the present invention will be described with reference to FIG.

1. The user gives character string information which is text data of a certain character string to the speech synthesizer.

2. The user gives the speech synthesizer speech information that utters a character string with an accent assumed by the user.

3. The speech synthesizer extracts accents from the acquired speech information.

4). The speech synthesizer generates synthesized speech information corresponding to the character string based on the accent of the speech information.

  Thereby, the synthetic | combination audio | voice corresponding to a character string can be produced | generated based on the accent which the user input with the audio | voice. Therefore, the user can easily obtain a synthesized speech having an accent assumed by the user.

2. Functional Block Diagram A speech synthesizer M1 according to the present invention will be described based on the functional block diagram shown in FIG. The speech synthesizer M1 includes speech information acquisition means M11, accent information extraction means M13, speech synthesis means M15, and character string information acquisition means M17.

  The voice information acquisition unit M11 acquires a certain voice as voice information.

  The accent information extraction unit M13 extracts the accent included in the audio information as accent information from the acquired audio information. Further, the accent information extracting unit M13 extracts the accent information using a fundamental frequency function that represents a temporal change in the fundamental frequency of the acquired voice information. Further, the accent information extraction unit M13 associates the character string represented by the acquired character string information with the voice information, determines which syllable of the syllable constituting the character string has an accent, A syllable that is determined to have an accent is extracted as the accent information.

  Based on the accent information and the character string information, the speech synthesizer M15 generates synthesized speech corresponding to the character string and having the accent. Further, the speech synthesizer M15 generates synthesized speech corresponding to the character string in the character string information and having an accent on a syllable that is determined to have an accent.

  The character string information acquisition unit M17 acquires character string information representing a certain character string.

  Thereby, the synthetic | combination audio | voice corresponding to a character string can be produced | generated based on the accent which the user input with the audio | voice. Therefore, the user can easily obtain a synthesized speech having an accent assumed by the user. Also, the accent can be easily determined. Furthermore, it is possible to easily identify a syllable in which an accent exists.

3. Hardware Configuration of Speech Synthesizer 21 The hardware configuration of the speech synthesizer 21 that is a speech synthesizer according to the present invention will be described with reference to FIG. The speech synthesizer 21 includes a CPU 211, a memory 212, a hard disk 213, a keyboard 214, a mouse 215, a display 216, a CD-ROM drive 217, a sound circuit 218, a speaker 218s, and a microphone 218m.

  The CPU 211 performs processing based on other applications such as an operating system (OS) and a voice synthesis program recorded on the hard disk 213. The memory 212 provides a work area for the CPU 211. The hard disk 213 records and holds other applications such as an operating system (OS) and a speech synthesis program and various data. The data recorded on the hard disk 213 will be described later.

  The keyboard 214 and the mouse 215 accept external commands. The display 216 displays an image such as a user interface. The CD-ROM drive 217 reads data from a CD-ROM, such as reading a voice synthesis program from a CD-ROM 210 in which a voice synthesis program is recorded and a program of another application from another CD-ROM. The sound circuit 218 converts the given voice synthesis data into an analog waveform and outputs the analog waveform to the speaker 218s. The sound circuit 218 also converts the analog waveform acquired via the microphone 218m into a digital waveform.

4. Data A syllable duration database (hereinafter referred to as syllable duration DB) recorded by the speech synthesizer 21 on the hard disk 213 will be described with reference to FIG. The syllable duration DB is a database that associates a syllable with a standard duration when the syllable is uttered.

  The syllable duration DB has a [syllable] column C401 and a [duration] column C405. [Syllable] column C401 describes syllable types generally used in Japanese. In the [Duration] column C405, the utterance time when the syllable described in the [Syllable] column C401 is uttered at a standard speed is described as the duration.

5. Operation of Speech Synthesizer 21 An outline of the operation of the CPU 211 of the speech synthesizer 21 will be described with reference to FIG. The CPU 211 displays a character string / voice information acquisition screen D1 for acquiring character string information or voice information (S501).

  An example of the character string / voice information acquisition screen D1 displayed on the display 216 of the voice synthesizer 21 is shown in FIG. The character string / voice information acquisition screen D1 includes a character string input area A601 and a voice information acquisition start button B601. The character string input area A601 is an area for inputting character string information representing a kana character string to be synthesized by the user using an input unit such as the keyboard 214. The voice information acquisition start button B601 is a button for selecting with the mouse 215 or the like when the user wants to input the accent given to the character string by voice.

  Returning to FIG. 5, when the CPU 211 determines that the voice information acquisition start button B601 (see FIG. 6) has been selected (S503), the character string input in the character string input area A601 of the character string / voice information acquisition screen D1. Is acquired as character string information and stored in the memory 212 (S511). In addition, the CPU 211 acquires audio information via the microphone 218m (S513). The user inputs the voice corresponding to the character string input in the character string input area A601 of the character string / voice information acquisition screen D1 toward the microphone and having an accent assumed by the user. When the CPU 211 determines that voice information has been acquired, the CPU 211 executes syllable alignment processing (S515), accent position determination processing (S517), and voice synthesis processing (S519). The CPU 211 reproduces the generated synthesized voice information via the speaker 218s (S521).

  Hereinafter, the syllable alignment process (S515), the accent position determination process (S517), and the speech synthesis process (S519) will be described.

5.1. Syllable alignment processing The syllable alignment processing (see FIG. 5: S515) executed by the CPU 211 is executed to determine at which position of the acquired speech information the syllable break of each kana character constituting the character string information is present. It is processing to do. The syllable alignment process executed by the CPU 211 will be described with reference to the flowchart shown in FIG.

  The CPU 211 obtains the duration corresponding to the kana characters constituting the character string information obtained in step S511 (see FIG. 5) from the [duration] column C405 of the syllable duration DB (see FIG. 4) (S801). The CPU 211 calculates a total duration obtained by summing up the values of the acquired [duration] column C405 (S803). The CPU 211 calculates the ratio between the calculated total duration and the duration of each kana character for the kana characters constituting the character string information (S805).

  Further, the CPU 211 measures the utterance time of the audio information acquired in step S513 (see FIG. 5) (S811). Based on the ratio between the total duration calculated in step S805 and the duration of each kana character, and the utterance time measured in step S811, the CPU 211 correlates the kana characters constituting the character string information with the speech information. Is stored in the memory 212 as a character-speech correspondence table (S815).

  Here, the character-speech correspondence table will be described with reference to FIG. The character-speech correspondence table has a [character] column C1401 and a [corresponding time] column C1403. [Character] column C1401 describes the kana characters constituting the character string information. The [corresponding time] column C1403 describes the time corresponding to the kana character described in the [character] column C1401 in the voice information.

  For example, when the kana character “NA” constituting the character string information “NAKAYAMA” corresponds to “0” to “0” 30 of the voice information, [corresponding time] corresponding to “NA” in the [character] column C1401 In a column C1403, a value “0′00” −0′30 ″ ”is described.

  Thus, the correspondence between the voice information acquired from the microphone 218m and the time from which time corresponds to each kana character constituting the character string information and each kana character constituting the character string information. It becomes possible to grasp.

5.2. Accent Position Determination Processing The accent position determination processing (see FIG. 5: S517) executed by the CPU 211 will be described with reference to the flowchart shown in FIG. The CPU 211 calculates a fundamental frequency function for the audio information acquired in step S513 (see FIG. 5) (S901). The calculation of the fundamental frequency in the fundamental frequency function is performed by calculating an autocorrelation function of the acquired voice information and obtaining a period in which the correlation value is equal to or greater than a certain threshold value.

  The CPU 211 calculates a first derivative function of the calculated fundamental frequency function (S903). Then, the CPU 211 determines that there is an accent position at a position where the calculated first derivative function value of the fundamental frequency function changes from positive to negative (S905), and temporarily stores the time corresponding to the accent position as accent position information to the memory 212. (S907). The CPU 211 obtains a character-speech correspondence table from the memory 212 (S909), and determines which kana character corresponds to the time of the accent position information among the kana characters constituting the character string information (S911). The CPU 211 temporarily stores the kana character having the accent position in the memory 212 as accent character information (S913).

  The syllable alignment process has finished associating the speech information with the kana characters constituting the character string information, and can grasp from which time to which time the speech information corresponds to which kana character. It is possible. Therefore, it can be determined which kana character corresponds to the time corresponding to the accent position information.

5.3. Speech Synthesis Processing The speech synthesis processing (see FIG. 5: S519) executed by the CPU 211 will be described using the flowchart shown in FIG. The CPU 211 acquires the character string information acquired in step S511 from the memory 212 (S1101). Further, the CPU 211 acquires the accent character information obtained by the accent position determination process (see S517 in FIG. 5) from the memory 212 (S1103).

  The CPU 211 generates synthesized speech information based on the character string information and the accent character information (S1105), and stores the synthesized speech information in the memory 212 (see S1107). Note that, for the generation of synthesized speech, a speech synthesis technique that has been generally used is used.

  As described above, the speech synthesizer 21 acquires the accent given to the character string for which the synthesized speech is to be generated from the speech information uttered by the user. That is, the user can provide the speech synthesizer 21 with the voice having the accent given to the kana character string by himself / herself. Therefore, the user can easily obtain a synthesized voice of a character string having an accent uttered by the user.

6. Specific Example The operation of the CPU 211 of the speech synthesizer 21 described so far will be described with a specific example.

  The user displays a character string / voice information acquisition screen D1 for acquiring character string information or voice information (see FIG. 5: S501). The kana character string “Nakayama” is input to the character string input area A601 of the character string / voice information acquisition screen D1 (see FIG. 6) displayed on the display 216 of the voice synthesizer 21 by using the keyboard 214, and the voice information is input. It is assumed that the acquisition start button B601 is selected with the mouse 215. It is assumed that the user aims to generate a synthesized voice for the place name “Nakayama” in a certain region in Kinki.

  When the CPU 211 determines that the voice information acquisition start button B601 has been selected (see FIG. 5: S503), the CPU 211 acquires the kana character string “Nakayama” input in the character string input area A601 as character string information and stores it in the memory 212. (See FIG. 5: S511).

  Further, the CPU 211 acquires audio information via the microphone 218m (see FIG. 5: S513). The user enters the character string “Nakayama” as a place name toward the microphone (accents for the kana characters “na”, “ka”, “ya”, “ma” are “low”, “high”, “low”, “ Pronounce "low") and input it as voice information. When the CPU 211 determines that the voice information has been acquired, the CPU 211 sets the duration corresponding to each kana character “NA”, “KA”, “YA”, “MA” constituting the acquired character string information “NAKAYAMA” as the syllable duration. Obtained from the [duration] column C405 of the DB (see FIG. 7: S801). The CPU 211 calculates a total duration obtained by summing up the values of the acquired [duration] column C405 (see FIG. 7: S803). The CPU 211 calculates a ratio between the calculated total duration and the duration of each kana character for the kana characters constituting the character string information (see FIG. 7: S805).

  Further, the CPU 211 measures the sound generation time of the sound information acquired from the microphone 218m (see FIG. 7: S811). The CPU 211 determines the correspondence between each kana character constituting the character string information and the speech information based on the ratio between the total duration and the duration of each kana character and the measured pronunciation time (FIG. 7 :). A character-speech correspondence table (see FIG. 8) is generated and stored in the memory 212 (see FIG. 7: S815).

  The CPU 211 calculates a fundamental frequency function for the acquired audio information (see FIG. 9: S901). The CPU 211 calculates a first derivative function of the calculated fundamental frequency function (see FIG. 9: S903), and determines that there is an accent position at a position where the value of the first derivative function changes from positive to negative (see FIG. 9: S905). . The CPU 211 temporarily stores the time corresponding to the accent position as accent position information in the memory 212 (see FIG. 9: S907). The CPU 211 obtains the character-speech correspondence table (see FIG. 9: S909), and the time of the accent position information includes the kana characters “NA”, “KA”, “YA”, “MA” that constitute the character string information. It is determined which kana character corresponds (see FIG. 9: S911). If the CPU 211 determines that the kana character having the accent position is the kana character “ka”, the CPU 211 temporarily stores it in the memory 212 as accent character information (see FIG. 9: S913).

  Then, the CPU 211 acquires character string information “Nakayama” from the memory 212 (see FIG. 10: S1101). The CPU 211 acquires the kana character “ka”, which is accent character information, from the memory 212 (see FIG. 10: S1103).

  The CPU 211 generates synthesized speech information based on the character string information and accented character information (see FIG. 10: S1105), and stores the synthesized speech information in the memory 212 (see FIG. 10: S1107).

  The CPU 211 reproduces the generated synthesized voice information via the speaker 218s (see S521 in FIG. 5).

Thereby, the user can easily obtain a synthesized voice of the character string “Nakayama” having an accent uttered by the user.

1. Overview An overview of a second embodiment of the speech synthesizer according to the present invention will be described. In the first embodiment described above, the speech synthesizer 21 extracts synthesized speech of a character string having the accent by extracting the accent given to the character string from the speech information that the user has uttered. Generated. As a result, the user can easily obtain a synthesized voice of a character string having an accent uttered by the user.

  On the other hand, the speech synthesizer 21 generates synthesized speech using syllable speech data having the standard pitch and speed of each syllable. Depending on the user, there may be a request such as “with a slightly higher tone” or “with a slightly slower speed” instead of the standard pitch or speed.

  The speech synthesizer in this embodiment provides the user with some synthesized speech in which predetermined parameters are changed for the synthesized speech in order to satisfy the user's request to easily change the synthesized speech. However, it is possible to easily obtain the synthesized speech desired by the user.

  In the present embodiment, the same configurations and operations as those in the first embodiment are given the same numbers as those in the first embodiment.

2. Functional Block Diagram A speech synthesizer M51 according to the present invention will be described based on the functional block diagram shown in FIG. The speech synthesizer M51 includes speech information acquisition means M11, accent information extraction means M13, speech synthesis means M15, character string information acquisition means M17, modified synthesized speech generation means M21, synthesized speech candidate display means M23, and synthesized speech determination means M25. have. Since the voice information acquisition unit M11, the accent information extraction unit M13, the voice synthesis unit M15, and the character string information acquisition unit M17 have the same configuration as that of the first embodiment, the description below is omitted.

  The modified synthesized speech generation means M21 changes the feature amount of the generated synthesized speech and generates synthesized speech having the changed feature amount.

  The synthesized speech candidate display means M23 displays the synthesized speech in which the feature amount is changed and the synthesized speech before the change is made as a synthesized speech candidate on the display means. The synthesized speech candidate display means M23 is configured so that the displayed synthesized speech candidates can be selected by the input means. The synthesized speech candidate display means M23 arranges the synthesized speech candidate on a plane with the pitch and speed of the sound as two axes when the feature amount is constituted by the pitch and speed of the sound.

  When the synthesized speech determining unit M25 determines that any of the synthesized speech candidates displayed on the display unit is selected, the synthesized speech determining unit M25 determines the synthesized speech candidate as a synthesized speech. When the synthesized speech determination unit M25 further determines that any of the synthesized speech candidates displayed on the display unit is selected, the synthesized speech candidate is reproduced and confirmed as synthesized speech information for the reproduced synthesized speech candidate. When the confirmation information to be acquired is acquired, the synthesized speech candidate is determined as a synthesized speech.

  Thereby, a plurality of synthesized speech candidates can be provided. Therefore, the user can obtain synthesized speech by an easy operation of selecting from the provided synthesized speech candidates. In addition, synthesized speech candidates can be easily obtained in which the pitch or / and speed of the sound is changed. Furthermore, the user can easily grasp the correlation between the provided synthesized speech candidates. Therefore, the user can easily select a desired one from the synthesized speech candidates. Furthermore, the user can select a synthesized speech candidate after confirming the reproduced sound of the synthesized speech candidate.

3. Hardware Configuration of Speech Synthesizer 51 The hardware configuration of the speech synthesizer 51, which is a speech synthesizer according to the present invention, is the same as the hardware configuration (see FIG. 3) in the first embodiment.

4. Data The data of the speech synthesizer 51 which is the speech synthesizer according to the present invention is the same as the data in the first embodiment (see FIG. 4).

5. Operation of Speech Synthesizer 51 An overview of the operation of the CPU 211 of the speech synthesizer 51 will be described with reference to FIG. The processing in steps S501 to S519 and S521 in FIG. 12 is the same as the operation of the CPU 211 of the speech synthesizer 21 of the first embodiment. Therefore, in the following, the processing of steps S551 and S553 will be described.

  In addition, when the speech synthesis process in step S519 ends, the CPU 211 executes a height / speed extraction process (S551). When the height / speed extraction process (S551) ends, the CPU 211 executes a synthesized speech candidate provision process (S553).

  Hereinafter, the height / speed extraction process (S551) and the synthesized speech candidate providing process (S553) will be described.

5.1. Height / Speed Extraction Processing The height / speed extraction processing (see FIG. 12: S551) executed by the CPU 211 will be described with reference to the flowchart shown in FIG. The CPU 211 acquires, as height information, the pitch of a sound that is one of the syllable feature amounts in the audio information acquired in step S513 (S1001). Further, the CPU 211 calculates the speed of the voice information, which is one of the syllable feature amounts in the voice information acquired in step S513, as speed information (S1003). The CPU 211 temporarily stores and holds the calculated height information and speed information in the memory 212 (S1005). In addition, when acquiring height information and speed information, information relating to sound (acoustic information), information relating to prosody (prosodic information), and the like in frequency components of speech information may be used.

5.2. Synthetic Speech Candidate Providing Processing The synthetic speech candidate providing processing (see FIG. 12: S553) executed by the CPU 211 will be described using the flowchart shown in FIG. The CPU 211 acquires the speed information and the height information extracted from the height / speed extraction process S551 (see FIG. 12) from the memory 212 (S1201). The CPU 211 changes the values of the speed information and the height information of the synthesized voice information, and generates synthesized voice information having the changed feature amount (S1203). Specifically, a parameter characterizing the synthesized speech is set in advance, and a cost value based on the value of the parameter is calculated as a reference cost value. On the other hand, a cost value based on a value corresponding to the parameter of the synthesized speech information synthesized in step S1203 is calculated as a synthesized cost value. Then, the reference cost value and the synthesis cost value are compared, and synthesized speech information having a synthesis cost value that meets a predetermined condition is generated. Note that the CPU 211 generates a plurality of synthesized speech information as much as possible.

  The CPU 211 adds the synthesized speech information generated in step S1105 (see FIG. 10) to the synthesized speech information generated in step S1203 to obtain synthesized speech candidate information (S1204). Then, the CPU 211 arranges each piece of synthesized speech candidate information in a synthesized speech selection diagram with two axes of speed and pitch based on the values of the speed information and the height information of the synthesized speech candidate information. (S1205). The CPU 211 displays the generated synthesized speech selection diagram on the display 216 (S1207).

  A synthesized speech selection diagram displayed on the display 216 is shown in FIG. In the synthesized speech selection diagram, a plan view constituted by a vertical axis representing the pitch of the sound and a horizontal axis representing the speed of the sound is displayed. Based on the speed information value and the height information value, the synthesized speech candidate information is specified and displayed on the plan view of the synthesized speech selection diagram. Note that the synthesized speech information synthesized in step S1105 (see FIG. 10) is arranged at the intersection (origin) of the vertical axis and the horizontal axis. The synthesized speech candidate information displayed on the synthesized speech selection diagram is configured so that the user can select it using the mouse 215 or the like.

  Returning to FIG. 14, the CPU 211 determines whether or not the determination button B1301 (see FIG. 15) has been selected (S1209). If the CPU 211 determines that the decision button B1301 has not been selected and one synthesized speech candidate information has been selected (S1211), the CPU 211 outputs the sound corresponding to the selected synthesized speech candidate information from the speaker 218s via the sound circuit 218. (S1213).

When determining that the decision button B1301 has been selected, the CPU 211 determines that the synthesized speech candidate information selected at that time is synthesized speech information, and records it in the hard disk 213 (S1215).

[Other Examples]
(1) Syllable alignment processing In the first embodiment described above, in the syllable alignment processing for determining at which position of the acquired speech information the syllable alignment is present, the standard duration of syllables constituting the syllable array is added. Then, the accent position was determined from the correlation with the duration of the voice information. However, the present invention is not limited to the illustrated example as long as it can be determined at which position in the voice information the syllable break is present. For example, the position of the syllable break may be determined at a position in the speech information by using a speech recognition technique using a hidden Markov model.

(2) Syllable duration DB
In the above-described first embodiment, the duration corresponding to the kana characters constituting the character string information is acquired using the simple syllable duration DB including the [syllable] column and the [duration] column. However, even for the same syllable, the duration changes depending on the preceding and succeeding syllables, the position of the syllable in the utterance phrase, and the strength of the utterance. Therefore, a more complicated database including those conditions may be used. Further, instead of using a simple table search to obtain the duration of the syllable from the database, a probability model, a decision tree, or the like may be used.

(3) Audio information In the first embodiment, the audio information is acquired by the sound circuit 218 and the microphone 218m. However, the audio information is not limited to the example as long as the audio information can be obtained. For example, audio information may be acquired via a recording medium such as a CD-ROM. Moreover, you may make it acquire audio | voice information via a communication line.

(4) Character String Information In the first embodiment, Japanese kana characters are exemplified as the character string information. However, the character string information is not limited to the illustrated example as long as it is a character string. For example, a foreign language such as English may be used.

(5) Feature Amount In the above-described second embodiment, the pitch and speed of the sound are used as the feature amount when generating the synthesized speech candidate information. However, as long as the synthesized speech candidate information can be generated, it is illustrative. It is not limited to. For example, it may be information that expresses subjective features of voice such as bright and dark.

(6) Synthetic Speech Candidate Information In the above-described second embodiment, when generating a synthesis as the synthesized speech candidate information, the reference cost calculated using the sound pitch and speed values of the speech information and the synthesized speech information as parameter values Although the value and the synthesis cost value are used, the present invention is not limited to the illustrated method as long as the synthesized speech candidate information can be generated.

  Moreover, in calculating the reference cost value, the pitch and speed of the sound acquired from the voice information are used, but the reference cost value is not limited to the example. For example, the reference cost value may be calculated using a preset parameter value. In this case, the pitch and speed of the sound need not be extracted from the sound information.

(7) Candidate display In the above-described second embodiment, the synthesized speech selection diagram having the pitch and the pitch of the sound as the feature amount and the sound pitch and the velocity as the two axes is generated. It is not limited to. For example, three types and four types of feature amounts may be selected.

  Further, the synthesized speech selection diagram is not limited to the example of biaxial display as long as the synthesized speech candidate information can be selected. For example, 3-axis, 4-axis display, etc. may be used.

  Furthermore, the synthesized speech information synthesized in step S1105 (see FIG. 10) is arranged at the intersection (origin) between the vertical axis and the horizontal axis of the synthesized speech selection diagram, but it was acquired from the user in step S513. Audio information may be displayed at the origin.

(8) Order of processing in flowchart In the first and second embodiments, each processing is realized based on each flowchart shown in the figure. However, as long as each process can be realized, the order of the processes in each flowchart is not limited to the example.

It is the figure which showed the outline | summary of the speech synthesizer in this invention. It is the figure which showed the functional block diagram of the speech synthesizer in Example 1. FIG. 2 is a diagram illustrating a hardware configuration of a speech synthesizer 21. FIG. It is the figure which showed the data structure of syllable duration DB. 3 is a flowchart showing the operation of the speech synthesizer 21. It is the figure which showed the character string and audio | voice information acquisition screen D1. It is a flowchart showing a syllable alignment process. It is the figure which showed the character-speech correspondence table. It is the flowchart which showed the accent position judgment process. It is the flowchart which showed the speech synthesis process. It is the figure which showed the functional block diagram of the speech synthesizer in Example 2. FIG. 5 is a flowchart showing the operation of the speech synthesizer 51. It is the flowchart which showed the height and speed extraction process. It is the flowchart which showed the synthetic speech candidate provision process. It is the figure which showed the synthetic | combination audio | voice selection figure. It is a figure for demonstrating the conventional speech synthesizer. It is a figure for demonstrating the conventional speech synthesizer.

Explanation of symbols

21... Synthesizer 51... Speech synthesizer M 11... Speech information acquisition means M 13... Accent information extraction means M 15. ... Character string information acquisition means M21... Changed synthesized speech generation means M23... Synthetic speech candidate display means M25.

Claims (6)

A speech synthesizer that generates synthesized speech corresponding to a character string,
The speech synthesizer
Character string information acquisition means for acquiring character string information representing a certain character string;
Voice information acquisition means for acquiring a certain voice as voice information;
From the acquired voice information, an accent information extracting means for extracting the accent of the voice information as accent information,
Based on the accent information and the character string information, synthesized speech corresponding to the character string, the voice synthesizing means for generating the accented sound,
Changed synthesized speech generation means for changing the feature amount of the synthesized speech composed of sound level information and speed information and generating at least one synthesized speech having the changed feature amount;
The synthesized speech in which the feature amount is changed and the synthesized speech before the change are arranged on a plane with two pitches of sound pitch and speed as a synthesized speech candidate based on the feature amount and displayed on the display unit. Synthetic speech candidate display means for displaying,
When it is determined that any of the synthesized speech candidates displayed on the display unit is selected, a synthesized speech determination unit that determines the synthesized speech candidate as a synthesized speech;
A speech synthesizer. A speech synthesis program that generates synthesized speech corresponding to a character string,
The speech synthesis program is
Computer
Character string information acquisition means for acquiring character string information representing a certain character string;
Voice information acquisition means for acquiring a certain voice as voice information;
From the acquired voice information, an accent information extracting means for extracting the accent of the voice information as accent information,
Based on the accent information and the character string information, synthesized speech corresponding to the character string, the voice synthesizing means for generating the accented sound,
Changed synthesized speech generation means for changing the feature amount of the synthesized speech composed of sound level information and speed information and generating at least one synthesized speech having the changed feature amount;
The synthesized speech in which the feature amount is changed and the synthesized speech before the change are arranged on a plane with two pitches of sound pitch and speed as a synthesized speech candidate based on the feature amount and displayed on the display unit. Synthetic speech candidate display means for displaying,
When it is determined that any of the synthesized speech candidates displayed on the display unit is selected, a synthesized speech determination unit that determines the synthesized speech candidate as a synthesized speech;
Speech synthesis program to function as. In the speech synthesis apparatus according to claim 1 or the speech synthesis program according to claim 2,
The accent information extracting means further includes:
Extracting the accent information using a fundamental frequency function representing a temporal change in the fundamental frequency of the acquired voice information;
It is characterized by. In any of the speech synthesizer or the speech synthesis program according to claim 3,
The accent information extracting means further includes:
By associating the character string represented by the acquired character string information with the speech information, it is determined which syllable has an accent among the syllables constituting the character string, and the syllable that the accent has been determined is Extract as accent information,
The speech synthesis means further includes:
Generating synthesized speech corresponding to a character string in the character string information, wherein the syllable is determined to have an accent,
It is characterized by. In any one of the speech synthesizer or the speech synthesis program according to claims 1 to 4 ,
The synthesized speech determination means further includes:
When it is determined that one of the synthesized speech candidates displayed on the display means is selected, the synthesized speech candidate is reproduced, and when the confirmed information to be confirmed as synthesized speech information for the reproduced synthesized speech candidate is acquired, the synthesis is performed. Determining speech candidates as synthesized speech;
It is characterized by. A synthetic speech generation method for generating synthetic speech corresponding to a character string using a computer,
The computer obtains character string information representing a certain character string,
The computer acquires a certain voice as voice information,
The computer extracts the accent that the audio information has from the acquired audio information as accent information,
Based on the accent information and the character string information, the computer generates a synthesized speech corresponding to the character string and having the accent,
A computer changes a feature amount of the synthesized speech constituted by sound level information and speed information, and generates at least one synthesized speech having the changed feature amount;
The computer arranges the synthesized speech in which the feature amount is changed and the synthesized speech before the change is made on a plane having two axes of the pitch and speed of the sound as a synthesized speech candidate based on the feature amount. Display
When the computer determines that one of the displayed synthesized speech candidates is selected, the synthesized speech generation method determines the synthesized speech candidate as a synthesized speech .

Images