JP2006337468A - Device and program for speech synthesis - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a device and a program for speech synthesis that synthesize speeches of a plurality of voices. <P>SOLUTION: In full-text selection mode, "all together", "repetition" or, "troll" is specified and a kind of speech is specified. Alternatively, a kind of speech is specified for each accent phrase in accent phrase mode. In the "all together" mode, a plurality of speeches are outputted simultaneously to generate such effect that a plurality of persons read a text aloud simultaneously. In the "repetition" mode, a speech of a leading speech kind is output by accent phrases, and then a speech of the speech kind of "repetition" is output to generate such effect that the output of each accent phrase with the leading speech is repeated with the speech of "repetition". In the "troll" mode, speech output is started in specified order of specified speech kinds after a first accent phrase of a last speech kind is output to generate such effect that a plurality of persons read the text aloud one after another like a troll. In the "accent phrase mode", a different number of persons read the text aloud with a different kind of speech for each accent phrase. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a speech synthesizer and a speech synthesis program, and more particularly to a speech synthesizer and a speech synthesis program that synthesize a plurality of voices.

  2. Description of the Related Art Conventionally, in a speech synthesis technique, a technique for performing various effects on output sound has been proposed so that a more effective sound can be output in accordance with the purpose of use. For example, in the game device of the invention described in Patent Document 1, a sound corresponding to the last character to be output is output again at a low volume so that the output sound has a lingering sound or the pronunciation of each character. Each time, the same volume waveform with lower volume is sequentially delayed and mixed for a certain period of time and mixed to output the sound, thereby adding echo and reverberation to the output sound. It is. Further, in the speech reading apparatus according to the invention described in Patent Document 2, when the user performs “repetitive designation”, an important part (a part including numbers) of the output speech is repeated.

As one of the effects of voice synthesis, synthesizing a plurality of voices is performed. In the singing voice synthesizer of the invention described in Patent Document 3, when singing voices divided into a plurality of parts together, the singing voice signal that is the voice synthesis waveform of each part generated by the singing voice signal generation unit is choralized. Addition is made in the signal generation unit to create a choral signal, which is converted into an analog signal and output as a singing voice. Moreover, in the singing voice synthesizing device of the invention described in Patent Document 4, as in the singing voice synthesizing device of the invention described in Patent Document 1, the chorus generating unit generates the choral voice generated by the singing voice synthesizing unit of each part. Or when the voices of different parts are output from different output units, or the singing voices of the parts previously synthesized by the singing voice synthesizing unit are temporarily stored in the external storage device, and then the singing voices of other parts are synthesized later In addition, the singing voice of the previously synthesized part is taken out and synthesized and output as a chorus voice.
Japanese Patent No. 3252896 JP-A-5-197384 Japanese Patent No. 3333022 Japanese Patent No. 3514263

  However, in the game device of the invention described in Patent Document 1, only one type of sound is used to give the sound a lingering sound or an echo effect. The voice reading device only repeats with the same voice as the voice that was output first, and does not produce the output voice using a plurality of types of voice. In addition, in the speech synthesizer described in Patent Literature 3 and Patent Literature 4, after all of the voices of a plurality of parts are generated and then synthesized and output, the larger the number of parts, Since it takes time to generate all sounds, there is a problem that it takes time from the start of sound generation to the start of output. Further, as in the speech synthesizer described in Patent Document 4, in order to output different part sounds from different output units, there is a problem in that it is necessary to provide output units corresponding to the types of sounds.

  The present invention has been made to solve the above-described problems, and an object thereof is to provide a speech synthesizer and a speech synthesis program for synthesizing a plurality of voices.

  In order to solve the above-described problem, in the speech synthesizer according to the first aspect of the present invention, a phoneme model created from phoneme data obtained by analyzing speech into an acoustic parameter sequence and a prosody model created from fundamental frequency data obtained by analyzing speech Acoustic dictionary storage means for storing an acoustic dictionary that is a set of acoustic models at least for each of a plurality of voice types, sentence input means for inputting a sentence for generating voice, and the voice input by the sentence input means Language analysis means for determining a part of speech by decomposing a sentence to be generated, determining an accent type indicating an accent position for each accent phrase, and determining a reading of a sentence for generating the speech; and the language analysis means Analysis result storage means for storing the analysis result analyzed by one or more of the acoustic dictionaries stored in the acoustic dictionary storage means Sound type designation means for designating a class, and sound for selecting the acoustic model from the acoustic dictionary based on the analysis result stored in the analysis result storage means and the voice type designated by the voice type designation means It is characterized by comprising model selection means and sound generation means for generating sound based on the acoustic model selected by the acoustic model selection means.

  Further, in the speech synthesizer according to the invention of claim 2, in addition to the configuration of the invention of claim 1, the speech generation means is configured to accentuate based on the acoustic model selected by the acoustic model selection means. A voice is generated for each phrase, word, morpheme, or character.

  Moreover, in the speech synthesizer of the invention according to claim 3, in addition to the configuration of the invention according to claim 1 or 2, the speech type designation means includes one or a plurality of speech statements for the whole sentence for generating the speech. An overall voice type designation means for designating the voice type is provided.

  Moreover, in the speech synthesizer of the invention according to claim 4, in addition to the configuration of the invention according to any one of claims 1 to 3, the speech type designation means is provided for each accent phrase in the sentence that generates the speech. An accent phrase-specific voice designation means for designating one or a plurality of the voice types is provided.

  Further, in the speech synthesizer of the invention according to claim 5, in addition to the configuration of the invention according to any of claims 1 to 4, the speech type designation means includes two parts, a first part and a second part. The voice type designation means for designating each of the voice types, wherein the voice generation means receives the voice of the voice type specified in the first part for each predetermined block of the sentence for generating the voice. A repetitive sound generating means for outputting the sound of the sound type specified in the second part after being output and synthesizing the sound so that the sound of the second part repeats the sound of the first part. It is characterized by having.

  In the speech synthesizer of the invention according to claim 6, in addition to the configuration of the invention according to any one of claims 1 to 5, the speech type designation means designates the speech types of a plurality of parts, respectively. A multi-part sound type designating unit; and an order designating unit for designating an order of outputting the sound among the plurality of parts, wherein the sound generating unit is first in the order designated by the order designating unit. The voice of a predetermined block of the sentence that generates the voice of the part is output, and the second and subsequent parts of the part are output when the voice of the first block of the previous part is completed is output. And a ringing voice generating means for synthesizing the voices so that the voices of the parts ring in the order specified by the order specifying means.

  Further, in the speech synthesizer of the invention according to claim 7, in addition to the configuration of the invention according to claim 5 or 6, the block is a sentence, an accent phrase, a word, or a character delimited by a predetermined symbol. It is at least one of these.

  Further, in the speech synthesizer of the invention according to claim 8, in addition to the configuration of the invention according to any of claims 3 to 7, the speech type designation means includes the accent phrase-specific speech designation means, the whole speech It is characterized by comprising a designation method selection means for selecting one of a type designation means, a part-specific voice type designation means, and a multi-part voice type designation means.

  According to a ninth aspect of the present invention, there is provided a speech synthesis program that causes a computer to function as various processing means of the speech synthesis apparatus according to any one of the first to eighth aspects.

  In the speech synthesizer of the invention according to claim 1, the acoustic dictionary storage means includes a phonological model created from phonological data obtained by analyzing speech into an acoustic parameter sequence, and a prosodic model created from fundamental frequency data analyzed by speech. An acoustic dictionary that is a set of at least an acoustic model is stored for each of a plurality of speech types, a sentence input unit inputs a sentence for generating a speech, and a language analysis unit generates a speech input by the sentence input unit. Decompose sentences into words to determine part-of-speech, determine the accent type that indicates the accent position for each accent phrase, and determine the reading of the sentence that generates speech, and the analysis result storage means is analyzed by the language analysis means And the voice type designation means designates one or a plurality of voice types from the acoustic dictionary stored in the acoustic dictionary storage means, and the acoustic model selection means. The acoustic model is selected from the acoustic dictionary based on the analysis result stored in the analysis result storage unit and the voice type designated by the voice type designation unit, and the voice generation unit selects the acoustic model selected by the acoustic model selection unit. Voice can be generated based on the above. Therefore, since the type of sound to be output can be designated, various sounds can be output by combining the sound types.

  In the speech synthesizer of the invention according to claim 2, in addition to the effect of the invention of claim 1, the speech generation means includes an accent phrase and a word based on the acoustic model selected by the acoustic model selection means. Voice can be generated for each morpheme or character. Therefore, since the voice is not output after generating the full-text voice to be output, the time from the start of voice synthesis until the voice is output is short, and there is no delay in the voice output and smooth output Can do.

  In the speech synthesizer of the invention according to claim 3, in addition to the effect of the invention according to claim 1 or 2, the overall speech type designation means of the speech type designation means is 1 for the whole sentence for generating speech. One or more audio types can be specified. Therefore, it is possible to obtain an effect that the whole sentence is uttered by a plurality of people.

  In the speech synthesizer of the invention according to claim 4, in addition to the effect of the invention according to any one of claims 1 to 3, the speech phrase-specific speech designation means of the speech type designation means is a sentence for generating speech. One or more voice types can be specified for each accent phrase in. Therefore, since it is possible to make a part of the voice uttered by a plurality of people, various effects can be performed.

  In the speech synthesizer of the invention according to claim 5, in addition to the effect of the invention according to any one of claims 1 to 4, the speech type designation means by part of the speech type designation means includes the first part and the first part. It is possible to designate two types of two-part audio types. The repetitive voice generating means of the voice generating means outputs the voice type specified in the second part after the voice of the voice type specified in the first part is output for each predetermined block of the sentence generating the voice. Since the voice can be synthesized so that the voice of the first part is repeated by the voice of the second part, the voice of the first part can be repeated with a favorite voice type.

  Further, in the speech synthesizer of the invention according to claim 6, in addition to the effect of the invention according to any one of claims 1 to 5, the multi-part speech type designating means of the speech type designating means Each type is designated, and the order designating unit can designate the order of outputting the sound among the plurality of parts. In addition, the ringing sound generating means of the sound generating means first outputs the sound of a predetermined block of the sentence that generates the sound of the first part in the order specified by the order specifying means, and the second and subsequent parts are output. Audio so that the sound of each part circulates in the order specified by the order specifying means so that the output of the sound is started when the output of the sound of the first block of the previous part is completed. Can be synthesized. Therefore, it is possible to output a sound of a favorite sound type in a favorite order and perform various effects.

  In the speech synthesizer of the invention according to claim 7, in addition to the effects of the invention of claim 5 or 6, the block is a sentence, accent phrase, word, or character delimited by a predetermined symbol. Since at least one of them can be used, audio output can be performed in a short cycle, and the audio output can be smoothly output without delay.

  In addition, in the speech synthesizer of the invention according to claim 8, in addition to the effect of the invention of any one of claims 3 to 7, the designation method selection means of the speech type designation means includes accent phrase-specific voice designation means, Since one means can be selected from the whole voice type designation means, the part-by-part voice type designation means, and the multi-part voice type designation means, a plurality of voice types can be output with various effects.

  In the speech synthesis program of the invention according to claim 9, the computer can function as various processing means of the speech synthesis apparatus according to any one of claims 1 to 8.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the speech synthesizer and the speech synthesis program of the present invention, for the sentence (text) to be output as speech, “simultaneous”, “return”, and “rotation” are designated in the “full text selection mode”, and the output speech type is set. Can be specified. Alternatively, the type of sound output for each accent phrase can be specified in the “accent phrase mode”.

  In "Batch" of "Full text selection mode", if there are multiple voices of the specified voice type, multiple voices are output at the same time, and the voice is output as if multiple people are reading the text simultaneously . If there is only one designated voice type, the voice is output as if one person is reading a text with one type of voice. In “Repeat” in “Full text selection mode”, each accent phrase is designated as the voice to be repeated after being output with the voice type designated as the leading voice (hereinafter referred to as “leading voice type”). The voice type (hereinafter referred to as “repeated voice type”) is output, and the voice is output so that the voice of the repeated voice type repeats the text output with the voice of the leading voice type for each accent phrase. In addition, when a plurality of repetitive voice types are designated, they are output simultaneously. Also, in the “full text selection mode” “ring”, voice output is started after the output of the first accent phrase of the previous voice type in the specified order for the specified voice type, In the same way as in “Ring”, voice is output as if a plurality of people are reading aloud following the previous person.

  In the “accent phrase mode”, the voice type specified for each accent phrase is output, and the voice is output as if another person is speaking for each accent phrase, or is read for each accent phrase. Sound is output so that the number of people who are changing. Note that, in the speech synthesizer and the speech synthesis program of the present embodiment, when the “full sentence selection mode” is selected, control is performed so that the speech type cannot be specified in the “accent phrase mode”.

  First, a mobile phone 1 that is a speech synthesizer according to the present embodiment will be described with reference to FIG. FIG. 1 is an external view of the mobile phone 1, and FIG. 2 is a block diagram showing an electrical configuration of the mobile phone 1. 3 is a schematic diagram illustrating the configuration of the RAM 22, FIG. 4 is a schematic diagram illustrating the configuration of the simultaneous information storage area 224 of the RAM 22, and FIG. 5 illustrates the configuration of the repetitive information storage area 225 of the RAM 22. FIG. 6 is a schematic diagram showing the configuration of the singing information storage area 226 of the RAM 22, and FIG. 7 is a schematic diagram showing the configuration of the accent phrase mode information storage area 227 of the RAM 22.

  As shown in FIG. 1, the mobile phone 1 includes a display screen 2, a numeric key input unit 3, a multi-button 4 having buttons in four directions and a determination button, a call start button 5, and a call end button. 6, a microphone 7, a speaker 8, function selection buttons 9 and 10, and an antenna 12 (see FIG. 2). The numeric key input unit 3, multi-button 4, call start button 5, call end button 6, and function selection buttons 9 and 10 constitute a key input unit 38 (see FIG. 2).

  As shown in FIG. 2, the cellular phone 1 includes an analog front end 36 that performs amplification of an audio signal from the microphone 7 and amplification of audio output from the speaker 8, and audio amplified by the analog front end 36. Digitalization of the signal and digital code received from the modem unit 34 are converted into an analog signal so that the analog front end 36 can amplify the signal, a modem unit 34 that performs modulation / demodulation, and amplification of radio waves received from the antenna 12 In addition, a transmission / reception unit 33 that performs detection and modulates and amplifies the carrier signal with the signal received from the modem unit 34 is provided.

  Further, the mobile phone 1 is provided with a control unit 20 that controls the entire mobile phone 1. The control unit 20 includes a CPU 21, a RAM 22 that temporarily stores data, and a clock function unit 23. ing. Further, the control unit 20 includes a key input unit 38 for inputting characters and the like, a display screen 2, a nonvolatile memory 30 storing a program and an acoustic dictionary of various voice types, and a melody generator 32 for generating a ring tone. It is connected. The melody generator 32 is connected to a speaker 37 that utters a ring tone generated by the melody generator 32.

  In this embodiment, five types of voices of “male”, “female”, “boy”, “girl”, and “animation” are used, with male being the first voice type and female being the second voice. Kind, boy is the third voice type, girl is the fourth voice type, and animation is the fifth voice type. The acoustic dictionary of the nonvolatile memory 30 stores five types of acoustic dictionaries for generating sounds of these five types of speech.

  Also, as shown in FIG. 3, the RAM 22 is provided with various storage areas for storing variables and generated data used when performing speech synthesis processing. For example, the text storage area 221 stores text instructed to perform speech synthesis. The analysis result storage area 222 stores the results analyzed by the morphological analysis process and the accent phrase formation process. The parameter information storage area 223 stores setting information designated in the “full sentence selection mode” and the “accent phrase mode”. Specifically, when “full text selection mode” is set, “1” is selected when “simultaneous” is selected, “2” is selected when “return” is selected, and “ring” is selected. If it is selected, “3” is set. If the voice type is set for each accent phrase in “accent phrase mode”, “9” is set. The initial value is “0”.

  The simultaneous information storage area 224 is a storage area used when “simultaneous” is selected in the “full-text selection mode”, and the repetitive information storage area 225 is “repeated” in the “full-text selection mode”. This is a storage area that is used when it is selected, and the ring information storage area 226 is a storage area that is used when “ring” is selected in the “full text selection mode”. The accent phrase mode information storage area 227 is a storage area used when “accent phrase mode” is designated. The synthesized voice storage area 228 is a storage area for storing a result of adding a plurality of voice data.

  As shown in FIG. 4, the simultaneous information storage area 224 of the RAM 22 is provided with a setting field and a sound data field for each of the first to fifth sound types. Information on whether or not each sound type is set as an output sound is stored in the setting column. In the present embodiment, “1” is stored when the output sound is set, and “0” is stored when it is not set. In the audio data column, audio data (sound source signal) generated for each audio type is stored. The voice data stored here is synthesized and a synthesized voice is output. Note that “m” is a variable used when counting voice types in the simultaneous processing (see FIG. 13) described later.

  As shown in FIG. 5, in the repetition information storage area 225 of the RAM 22, the leading voice type, the first repeated voice type, the second repeated voice type, the third repeated voice type, the fourth repeated voice A setting field and a sound data field are provided for each sound type. And, the number of repeaters column is provided. Each setting column is set with a number indicating the voice type. Also, the number of voice types designated as the read back voice type is set in the number of repeats column. For example, if the lead voice type is “male” and the repetitive voice types are “female” and “anime”, “1” is set in the lead voice type information setting field, and the first setting field in the repeat voice type field. “2”, “5” is set in the second setting field, and “2” is set in the return number field. In the audio data column, audio data (sound source signal) generated for each audio type is stored. The voice data stored here is synthesized and a synthesized voice is output. Note that “p” is a variable used when counting the type of repetitive sound in a repetitive processing (see FIG. 14) described later.

  As shown in FIG. 6, in the singing information storage area 226 of the RAM 22, an audio type column and an audio data column for each morpheme are provided for each order of the audio types to be sung. A number of singers column and a morpheme number column of the first accent phrase are provided. In the voice type column, a number indicating the voice type output in each order is set, and in the number of ringers column, the number of voice types specified as the voice type to be rotated is set, and the first accent phrase In the morpheme number column, the number of morphemes of the first accent phrase is calculated and stored in the singing process described later (see FIG. 15). For example, when “Women” is No. 1, “Girls” is No. 2, and “Animation” is No. 3, and the order of singing is specified, “2” and the second voice are entered in the first voice type column. “4” is set in the type column, “5” is set in the third audio type column, and “3” is set in the number of singers. In the voice data column for each morpheme, voice data (sound source signal) for each morpheme of each voice type is set. In the present embodiment, since five types of speech are used, five morpheme storage areas are also provided. If the speech data column is shown as (order, morpheme), one morpheme is generated. The voice data is stored in (q, q + morphic prime number of the first accent phrase × (q−1)). Note that “q” is a variable for counting the order of singing in the singing process (see FIG. 15) described later. When outputting the sound, the sound data stored in (1,1), (2,1), (3,1), (4,1), (5,1) is output. .

  As shown in FIG. 7, the accent phrase mode information storage area 227 of the RAM 22 is provided with a setting field for each accent phrase and a sound data field for each accent phrase for each sound type. The setting field for each accent phrase can store as many settings as the number of accent phrases in the text to be synthesized for each speech type, and whether or not to output the speech of that speech type as an accent phrase. Information is set. In this embodiment, “1” is stored when output is set, and “0” is stored when it is not set. In the voice data field for each accent phrase, voice data (sound source signal) can be stored as many as the number of accent phrases for each voice type, and the voice data generated for the voice type set to be output is stored. Remembered. Note that “m” is a variable used when counting the voice type in the accent phrase processing (see FIG. 16) described later, and “n” is a variable used when counting the accent phrase.

  Next, with reference to FIG. 8 to FIG. 11, a screen for inputting text for voice output and a screen displayed when selecting “full sentence selection mode” and “accent phrase mode” will be described. 8 is an image diagram of the main screen 290, FIG. 9 is an image diagram of the audio output screen 200, FIG. 10 is an image diagram of the full sentence selection mode screen 210, and FIG. 11 is an image diagram of the accent phrase mode screen 230. It is.

  A main screen 290 shown in FIG. 8 is a screen that is displayed when the mobile phone 1 is operated to start a speech synthesis program. A text input field 291 for inputting text for voice output, a reference button 292, and an OK button. A button and a cancel button are provided. When the text input field 291 is selected, the text can be input by operating the key input unit 38. When the reference button 292 is selected, the text to be input to the text input field 291 is selected from the mail text stored in the nonvolatile memory 30 of the mobile phone 1 or the text displayed on the screen connected to the Internet. can do. When the OK button is selected, the text input in the text input field 291 is stored in the text storage area 221 of the RAM 22 as text for voice output. When the cancel button is selected, the speech synthesis process ends.

  The audio output screen 200 shown in FIG. 9 is a screen that is displayed when the OK button is selected on the main screen 290, and the “full-text selection mode” and the “accent phrase mode” can be selected. The voice output screen 200 is provided with a text display column 201, an output button, and a cancel button. The text display column 201 stores text as text to be input to the text input column 291 of the main screen 290 and synthesized. The mode selection tags “◇” and “▽” are inserted in the text stored in the area 221 and displayed. When the text display field 201 is selected, the cursor 202 is displayed. When the four buttons of the multi-button 4 are operated to move the cursor over the tag and the selection button is used to select each tag. The mode screen corresponding to is displayed. The ◇ tag is a setting tag for “full text selection mode”, and the ▽ tag is a setting tag for “accent phrase mode”.

  In the example shown in FIG. 9, the text display column 201 displays “◇ ▽ Athletic meet ▽ Memories.

  When the tag is selected, a full text selection mode screen 210 shown in FIG. 10 is displayed. The full text selection mode screen 210 is provided with a mode selection field 211, a voice type selection field 212, an order designation field 213, an OK button, and a cancel button. The mode selection field 211 is a radio button, and one of “simultaneous”, “return”, and “rotation” can be selected. The voice type selection column 212 is a check box, and one or more voice types can be selected from “male”, “female”, “boy”, “girl”, and “anime”. In the order designation field 213, a numerical value can be input. In “repeating”, the voice of the voice type input as “1” is set as the leading voice type. In “ring”, singing is performed in the order of the numbers input here. Note that “simultaneous” is not used even if a numerical value is input in the order designation column 213, and “repeated” and “spinning” are not used even if a numerical value is input in the audio type selection column 212. The value is not used. When the OK button is selected, the setting contents are stored in a predetermined storage area of the RAM 22 according to the input mode, and the sound output screen 200 is displayed again. If the cancel button is selected, the input content is not stored and the process returns to the audio output screen 200.

  When the ▽ tag is selected on the audio output screen 200, an accent phrase mode screen 230 shown in FIG. 11 is displayed. In the accent phrase mode screen 230, a voice type selection field 231, an OK button, and a cancel button are provided. The voice type selection column 231 is a check box, and one or more voice types can be selected from “male”, “female”, “boy”, “girl”, and “anime”. When the OK button is selected, the setting contents are stored in the accent phrase mode information storage area 227 of the RAM 22 according to the input contents, and the sound output screen 200 is displayed again. If the cancel button is selected, the input content is not stored and the process returns to the audio output screen 200.

  Next, the speech synthesis process will be described with reference to the flowcharts of FIGS. FIG. 12 is a flowchart of the main process of the speech synthesis program in the present embodiment, FIG. 13 is a flowchart of the simultaneous process performed in the main process, and FIG. 14 is performed in the main process. FIG. 15 is a flowchart of a repetitive process performed in the main process, and FIG. 16 is a flowchart of an accent phrase process performed in the main process.

  The main process shown in FIG. 12 is started when an instruction to perform a voice synthesis process is given in the mobile phone 1. First, various storage areas are cleared as an initial process (S1). Then, a main screen 290 for speech synthesis processing is displayed, and when a sentence to be output is acquired (S2), morphological analysis processing is performed (S3), and accent phrase formation processing is performed (S4).

  In S2, the main screen 290 is displayed, and a character is input to the text input field 291 by operating the key input unit 38, or is displayed by connecting to a mail text or the Internet stored in the nonvolatile memory 30 of the mobile phone 1. When a text for voice output is input by inserting a sentence described on the screen and the OK button is selected on the main screen 290, the text is stored in the text storage area 221 of the RAM 22 so that the output sentence is stored. To be acquired.

  In the morphological analysis process, a language dictionary (not shown) having part-of-speech information, reading information, connection information, accent information, and the like is referred to, morphological analysis is performed by a known longest match method, and stored in the text storage area 221. Text is parsed into morphemes (parts of speech). In the accent phrase forming process, the connection information of the language dictionary is referred to and the morphemes are collected into accent phrases. Further, in the accent phrase forming process, the accent position is also determined from the accent information in the language dictionary. When words are moved to a compound word, accent position change processing is also performed for words with accent position movement. Finally, the reading information of the language information is referred to, and the text is replaced with the katakana text. If the text reads “I covered New York for a week,” “Issou Kambakari ( 6) | New York wo (3) Shuzaishita (0) "is output. Here, “|” indicates an exhalation paragraph delimiter, () indicates an accent phrase delimiter, and numbers in () indicate the accent position of the accent phrase. The results of the morphological analysis process (S3) and the accent phrase formation process (S4) are stored in the analysis result storage area 222. Further, here, the number of accent phrases of the text to be output as speech is calculated and stored in the analysis result storage area 222.

  Then, parameter selection processing is performed (S5). In this parameter selection process, the audio output screen 200 (see FIG. 9) is displayed on the display screen 2, and by selecting the ◇ tag and the ▽ tag, the full text selection mode screen 210 (see FIG. 10), the accent phrase mode screen 230 ( 11) is displayed, and various settings for audio output are performed. When the output button is selected on the voice output screen 200, if there is a selection in the mode selection field 211 of “full text selection mode”, “1” to “3” are selected in the mode selection field 211 in the RAM 22 parameter information storage area 223. If there is no selection in the “full text selection mode” mode selection field 211 and “accent phrase mode” is set, “9” is stored. If neither mode is set, the initial value remains “0”. If any mode is set, the setting content is stored in each storage area of the RAM 22, and the parameter selection process ends.

  Then, the value stored in the RAM 22 parameter information storage area 223 is read (S6). If the read values are “1”, “2”, “3”, “9” and the mode is set (S7: YES), there is a possibility that there is a voice output by a plurality of people. , Go to S8. Also, if the mode is not set other than these values (S7: NO), since the sound is output by one type of speech, the cepstrum analysis is performed on the whole sentence based on the results of the morphological analysis processing and the accent phrase formation processing. Processing is performed to generate a sound source signal (S11), and the sound source signal is output as sound through the MLSA filter (S12). Then, the process ends.

  In this cepstrum analysis processing, an acoustic dictionary stored in the nonvolatile memory 30 (in this embodiment, the acoustic dictionary of the first speech type “male” is used) based on the result of morphological analysis. A stored phoneme model is selected to generate a phoneme sequence, and a phoneme model of each phoneme is combined to generate a mel cepstrum sequence and a voiced / unvoiced information sequence (hereinafter referred to as a mcep sequence). The acoustic dictionary includes “a, b, by, ch, cl, d, dy, e, f, fy, g, gy, h, hy, i, j, k, ky, m, my, n, N, ny, o, p, pau, py, r, ry, s, sh, t, ts, ty, u, w, y, z ”are stored in a list of 38 phoneme models. In addition, there are cases where the phoneme environment and the prosodic environment before and after are considered. This phonological model is obtained by performing mel cepstrum analysis on natural speech. Each phonological model is divided into frames (one frame is 10 ms), and a mel cepstrum coefficient is stored for each frame. ing. In addition, voiced or unvoiced information is stored for each frame. Note that “pau” indicates a pause.

  Also, the prosodic model sequence corresponding to the accent delimiter and accent type analyzed by the accent phrase forming process of S4 is selected from the prosodic model of the acoustic dictionary, and the prosodic model sequence is generated. In the example of “I covered New York for only a week”, a prosodic model sequence of “(9, 6), pau, (6, 3), (5, 0)” is generated. This indicates that there is a pose after the prosody model of the accent type 6 with 9 mora (beats), and then the accent type 3 with 6 mora and the accent type 0 with 5 mora. Next, the generated prosodic model sequence is connected to generate a pitch sequence. However, the mora length is expanded and contracted to synchronize with the phoneme model according to the length of each phoneme in the phoneme model sequence at the time of connection.

  Then, a sound source signal is generated based on the voiced / unvoiced information of the generated msep sequence and the generated pitch sequence. As for the sound source signal, a pulse train signal is generated in the voiced portion and a noise signal is generated in the unvoiced portion based on the pitch sequence.

  If the values read from the RAM 22 parameter information storage area 223 are “1”, “2”, “3”, “9”, and the mode is set (S7: YES), the “full text selection mode” Is determined (S8), and if "full text selection mode" is not specified (S8: NO) instead of "1", "2", "3", "accent phrase" Since “mode” is designated, accent phrase processing is performed (S13, see FIG. 16), and the processing ends. If “full text selection mode” is specified (S8: YES), the read value is “1”, and “simultaneous” is selected (S9: YES), simultaneous processing is performed. (S14, see FIG. 13). When the read value is “2” and “return” is selected (S9: NO, S10: YES), a repeat process is performed (S15, see FIG. 14). When the read value is “3” and “ring” is selected (S9: NO, S10: NO), the ringing process is performed (S16, see FIG. 15). Then, the process ends.

  Here, the simultaneous processing will be described with reference to the flowchart of FIG. First, an initial value “1” is set in a variable n for counting accent phrases (S21). Then, the analysis result of the nth accent phrase is read from the analysis result storage area 222 (S22). Then, an initial value “1” is set to the variable m for counting the voice type (S23), and it is determined whether or not the mth voice type is set as the output voice (S24). . This is determined by whether or not “1” is stored in the m-th audio type setting field of the simultaneous information storage area 224.

  If “1” is stored in the setting column and it is set as an output sound (S24: YES), cepstrum processing is performed for the nth accent phrase in the mth sound type, and a sound source signal is generated. Then, it is stored in the audio data column of the mth audio type in the simultaneous information storage area 224 (S25). Then, “1” is added to the voice type count variable m (S26), and the process proceeds to S27. Further, if “1” is not stored in the setting column and it is not set as the output sound (S24: NO), it is not necessary to generate the sound data (sound source signal), so nothing is done in S27. Proceed to

  In S27, it is determined whether or not the processing has been performed for all voice types depending on whether or not the value of the variable m is larger than “5 (the number of voice types in the present embodiment)”. If it is not greater than “5”, the processing for all the voice types has not been completed yet (S27: NO), so the process returns to S24, and the processing for the next voice type is performed. When the processing of S24 to S27 is repeated and the value of the variable m becomes larger than “5” (S27: YES), all the audio data (sound source signal) stored in the audio data column of the simultaneous information storage area 224 is stored. The waveform is processed by level correction and stored in the synthesized voice storage area 228 (S28), and the voice is output (S29).

  Then, “1” is added to the variable n for counting the accent phrases (S30), and the value of the variable n is not larger than the number of accent phrases of the output text stored in the analysis result storage area 222 ( (S31: NO), the process returns to S22, and the next accent phrase is processed (S22 to S31). Then, when the processes of S22 to S31 are repeated and all the accent phrases have been processed (S31: YES), the simultaneous process ends, the process returns to the main process, and the main process also ends.

  In this way, if “simultaneous” is selected in the “full-text selection mode” and there are a plurality of voices of the designated voice type, the effect is obtained that a plurality of people read the same sentence. In addition, since cepstrum processing is performed for each accent phrase and the sound is output, even when multiple sound types are output at the same time, the user outputs instructions until the sound is output. Smooth audio output is possible without taking time.

  Next, the repetition process will be described with reference to the flowchart of FIG. First, an initial value “1” is set in a variable n for counting accent phrases (S42). Then, the analysis result of the nth accent phrase is read from the analysis result storage area 222 (S43), and the sound type acoustic dictionary set as the lead sound type in the repetitive information storage area 225 is referred to, so that the nth A cepstrum process is performed on the accent phrase, a sound source signal is generated, and stored in the voice data column of the lead voice type in the repetition information storage area 225 (S44). Then, sound is output (S45).

  Next, an initial value “1” is set in the variable p for counting the type of repetitive voice (S46), and the voice type indicated by the value stored in the p-th setting field is used for the nth accent phrase. A cepstrum process is performed, a sound source signal is generated, and stored in the p-th audio data column of the repetitive audio type in the repetitive information storage area 225 (S47). Then, “1” is added to the variable p (S48). It is determined whether or not the variable p is larger than that of the repeater and processing has been performed for all the types of readback voice (S49). If processing has not been performed for all the repetitive voice types (S49: NO), the process returns to S47, and cepstrum analysis processing is performed for the next repetitive voice type (S47). Then, when the processing of S47 to S49 is repeated and the processing is completed for all the repetitive voices (S49: YES), the voice data (sound source signal) stored in the voice data column of the repetitive voice type information is added. Then, waveform processing is performed by level correction, and it is stored in the synthesized voice storage area 228 (S50), and voice is output (S51).

  Then, “1” is added to the variable n for counting the accent phrase (S52), and if the value of the variable n is not larger than the number of accent phrases (S53: NO), the process returns to S43 and the next accent phrase is determined. Processing is performed (S43 to S53). When the processes of S43 to S53 are repeated and all the accent phrases have been processed (S53: YES), the repeat process is ended, the process returns to the main process, and the main process is also ended.

  In this way, when “repeat” is selected in “full-text selection mode”, the sound of one accent phrase is output by the sound designated as the lead speech type, and then designated as the type of repetitive speech The voice of the same accent phrase is continuously output by the generated voice. Again, because cepstrum processing is performed for each accent phrase and audio is output, even if multiple audio types are output simultaneously, until the user outputs audio after the user issues an audio output instruction Smooth audio output is possible without taking time.

  Next, the singing process will be described with reference to the flowchart of FIG. First, the morpheme number of the first accent phrase is calculated and set in the morpheme number field of the first accent phrase in the ring information storage area 226 (S61). Then, an initial value “1” is set to the variable s for counting morphemes (S62), the analysis result of the sth morpheme is read from the analysis result storage area 222 (S63), and the order of singing is counted. The initial value “1” is set in the variable q (S64). Then, cepstrum processing is performed on the s-th morpheme with the voice type indicated by the value set in the q-th voice type field of the ring information storage area 226, and a sound source signal is generated (S65). 226 is stored in the (q + first accent phrase morpheme number × (q−1))-th column in the voice data column for each q-th morpheme (S66).

  Then, “1” is added to the variable q that counts the order of singing (S67), and whether or not the processing has been completed up to the last voice type depending on whether or not the value of the variable q is larger than the number of singers. A determination is made (S68). If the process has not been completed to the end (S68: NO), the process returns to S65 and the process for the next type of sound is performed (S65 to S68). Then, the processes of S65 to S68 are repeated, and when the process is completed up to the last voice type (S68: YES), (1, 1), (2, 1), (3, 1) in the voice data column for each morpheme. , (4,1), (5,1) are added to the sound data (sound source signal), the waveform is processed by level correction, and stored in the synthesized sound storage area 228 (S69), Audio is output (S70). Then, the information in the voice data column for each morpheme is shifted (S71). Specifically, the audio data (1, 1), (2, 1), (3, 1), (4, 1), and (5, 1) output as audio are deleted. The audio data (1, 2) is stored in (1, 1), the audio data (2, 2) is stored in (2, 1), and the audio data (3, 2) is (3, 3). 1), (4,2) audio data is stored in (4,1), and (5,2) audio data is stored in (5,1). Then, (1,3) audio data is stored in (1,2), (2,3) audio data is stored in (2,2), and (3,3) audio data is (3,3). 2), audio data of (4, 3) is stored in (4, 2), audio data of (5, 3) is stored in (5, 2), etc. The data in the current column is shifted to the previous column.

  Then, “1” is added to the variable s for counting morphemes (S72). Then, depending on whether or not the value of the variable s is larger than all the morpheme numbers, it is determined whether or not the processing of all the morphemes has been completed (S73). If the value of the variable s is not larger than the number of morphemes (S73: NO), the process returns to S63, and the next morpheme is processed (S63 to S73). And if the process of S63-S73 is repeated and the process of all the morphemes is complete | finished (S73: YES), a ring process will be complete | finished, it will return to a main process and a main process will also be complete | finished.

  In this way, in the designated order, the audio output is started after the output of the first accent phrase of the immediately preceding audio type is completed, and a plurality of people are assigned to the previous person in the same way as in “chorus” in chorus. The sound is then output as if reading out. Here, since cepstrum processing is performed for each morpheme and the sound is output, even when a plurality of sound types are output at the same time, the sound is output after the user instructs the sound output. Smooth audio output is possible without taking time.

  Next, accent phrase processing will be described with reference to the flowchart of FIG. First, an initial value “1” is set in a variable n for counting accent phrases (S81). Then, the analysis result of the nth accent phrase is read from the analysis result storage area 222 (S82). Then, an initial value “1” is set to the variable m for counting the voice type (S83), and it is determined whether or not the mth voice type is set as the output voice (S84). . This is determined by whether or not “1” is stored in the setting field for each nth accent phrase of the mth speech type in the accent phrase mode information storage area 227.

  If “1” is stored in the setting column and it is set as an output sound (S84: YES), cepstrum processing is performed for the nth accent phrase in the mth sound type, and a sound source signal is generated. (S85) and stored in the voice data column for each nth accent phrase of the mth voice type in the accent phrase mode information storage area 227 (S86). Then, “1” is added to the voice type count variable m (S87), and the process proceeds to S88. Further, if “1” is not stored in the setting column and it is not set as the output sound (S84: NO), it is not necessary to generate the sound data (sound source signal), so nothing is done in S87. Then, “1” is added to the voice type count variable m (S87), and the process proceeds to S88.

  In S88, it is determined whether or not the processing has been performed for all voice types depending on whether or not the value of the variable m is larger than “5 (number of voice types in the present embodiment)”. If it is not greater than “5”, the processing for all voice types has not been completed yet (S88: NO), the process returns to S84, and the processing for the next voice type is performed. When the processing of S84 to S88 is repeated and the value of the variable m becomes larger than “5” (S88: YES), all the data stored in the voice data column of the nth accent phrase in the accent phrase mode information storage area 227 are stored. The sound data (sound source signal) is added, waveform processing is performed by level correction, and the result is stored in the synthesized sound storage area 228 (S89), and the sound is output (S90).

  Then, “1” is added to the variable n for counting the accent phrase (S91), and if the value of the variable n is not larger than the number of accent phrases (S92: NO), the process returns to S82 and the next accent phrase is determined. Processing is performed (S82 to S92). Then, when the processes of S82 to S92 are repeated and the processing of all accent phrases is completed (S92: YES), the accent phrase process is terminated, the process returns to the main process, and the main process is also terminated.

  In this way, voice data of a voice type designated for each accent phrase can be created, and voices can be output with different voice types and different voice types for each accent phrase. Again, because cepstrum processing is performed for each accent phrase and audio is output, even if multiple audio types are output simultaneously, until the user outputs audio after the user issues an audio output instruction Smooth audio output is possible without taking time.

  In addition, the storage area storing the acoustic dictionary of the nonvolatile memory 30 in the above embodiment corresponds to “acoustic dictionary storage means”, and the analysis result storage area 222 of the RAM 22 corresponds to “analysis result storage means”. In addition, the text input field 291 of the main screen 290 corresponds to “sentence input means”, the voice output screen 200, the full sentence selection mode screen 210, and the accent phrase mode screen 230 correspond to “voice type designation means”, and the voice output screen. 200 and the full sentence selection mode screen 210 correspond to “whole voice type designation means”, the voice output screen 200 and the accent phrase mode screen 230 correspond to “speech designation means by accent phrase”, and the voice output screen 200 and the full sentence selection mode. The screen 210 corresponds to “part-specific voice type designation means”, the voice output screen 200 and the full-text selection mode screen 210 correspond to “multi-part voice type designation means”, and the voice output screen 200 and the full-text selection mode screen 210 correspond to “ Corresponds to “order designation means”.

  The CPU 21 that performs the processes of S3 and S4 of the main process shown in FIG. 12 corresponds to “language analysis means”. S11 of the main process shown in FIG. 12, S25 of the simultaneous process shown in FIG. 13, and the repetition shown in FIG. The CPU 21 that performs the processes S44 and S47, the ring process S65 shown in FIG. 15, and the accent phrase process S85 shown in FIG. 16 corresponds to the “acoustic model selection means”. Then, S11 of the main process shown in FIG. 12, S25 of the simultaneous process shown in FIG. 13, S44 and S47 of the repeat process shown in FIG. 14, S65 of the ring process shown in FIG. 15, and S85 of the accent phrase process shown in FIG. The CPU 21 that performs the processing corresponds to “sound generation means”, the CPU 21 that performs the repetitive processing illustrated in FIG. 14 corresponds to “repeated sound generation means”, and the CPU 21 that performs the revolving processing illustrated in FIG. It corresponds to. The tag in the text display field 201 of the voice output screen 200 and the mode selection field 211 in the full text selection mode screen 210 correspond to “designation method selection means”.

  Note that the speech synthesizer and speech synthesis program of the present invention are not limited to the above-described embodiments, and various changes can be made without departing from the scope of the present invention. In the above embodiment, the voice synthesizer is the mobile phone 1. However, it goes without saying that the voice synthesizer is not limited to the mobile phone, and may be another device such as a dedicated terminal for voice synthesis or a personal computer. . In the above-described embodiment, all processing is completed in the mobile phone 1, but an apparatus provided with “sentence input means” and “speech type designation means” for a user to instruct execution of voice output processing. And the device provided with “language analysis means”, “analysis result storage means”, “acoustic model selection means”, and “speech generation means” for generating the output voice data do not need to be the same device, such as the Internet or The speech synthesizer of the present invention may be configured by connecting these devices via a network such as a LAN to enable data transmission / reception.

  Control of simultaneous processing, repetitive processing, ring processing, and accent phrase processing is not limited to the above-described embodiment, and may be any control that can obtain the same effect. In the above embodiment, the voice type is selected for each accent phrase. However, this block is not limited to the accent phrase, and a predetermined symbol (for example, #, *, /, ○, ●, ◎, etc.) is used. It may be a sentence, word, character, morpheme (part of speech), or the like. In the simultaneous processing and the repetitive processing, cepstrum processing is performed for each accent phrase and voice output is performed. However, the cycle of this processing is not limited to the accent phrase, and a predetermined symbol (for example, #, *, /, It may be a sentence, word, character, morpheme (part of speech), etc. The user inputs a predetermined symbol when a sentence (text) to be output is input.

  In the above embodiment, only one type of voice is used as the lead voice type in “repeating”, but it is needless to say that a plurality of voices may be designated as the lead voice type.

  Further, in the above embodiment, as the “rotation process”, as in the case of “rotation” in music, when the audio output of the first accent phrase of the previous audio type is completed, the audio of the next audio type is output. The output is started and the voice is continuously output until the end of the sentence. However, the method of outputting a plurality of voices in order is not limited to this. For example, after the voice output of the first accent phrase of the first voice type is finished, the second accent phrase of the first voice type and the first voice type of the second voice type are changed to the start position of the accent phrase. The short accent phrase is compensated for by the pause, and after this voice output is completed, the third accent phrase of the first voice type and 2 of the second voice type are output. The first voice type and the first voice type of the third voice type are output with the start position of the accent phrase aligned, and the shorter accent phrase is supplemented with the pause time for the shortest accent phrase, and so on. In addition, the voice may be output with a shift for each accent phrase. In this case, it is efficient to create a sound source signal for each accent phrase and output the sound instead of creating a full-text sound source signal and outputting the sound, and real-time sound output can be performed.

  The speech synthesizer and speech synthesis program of the present invention can be applied to a speech synthesizer and speech synthesis program that synthesize a plurality of speech.

1 is an external view of a mobile phone 1. FIG. 2 is a block diagram showing an electrical configuration of the mobile phone 1. FIG. 3 is a schematic diagram showing a configuration of a RAM 22. FIG. 3 is a schematic diagram showing a configuration of a simultaneous information storage area 224 of a RAM 22. FIG. 3 is a schematic diagram showing a configuration of a repetitive information storage area 225 of a RAM 22. FIG. It is a schematic diagram which shows the structure of the rotation information storage area 226 of RAM22. 3 is a schematic diagram showing a configuration of an accent phrase mode information storage area 227 of a RAM 22. FIG. It is an image figure of the main screen 290. It is an image figure of the audio | voice output screen 200. FIG. It is an image figure of the full text selection mode screen. It is an image figure of the accent phrase mode screen. It is a flowchart of a main process. It is a flowchart of the simultaneous process performed in the main process. It is a flowchart of the repeat process performed in the main process. It is a flowchart of the singing process performed in the main process. It is a flowchart of the accent phrase process performed in the main process.

Explanation of symbols

1 Mobile phone 2 Display screen 21 CPU
22 RAM
30 Non-volatile memory 38 Key input unit 200 Voice output screen 210 Full-text selection mode screen 221 Text storage area 222 Analysis result storage area 223 Parameter information storage area 224 Simultaneous information storage area 225 Repetition information storage area 226 Ring information storage area 227 Accent phrase mode information Storage area 230 Accent phrase mode screen 290 Main screen m Voice type count variable n Accent phrase count variable p Repeated voice type count variable q Rotation order count variable s Morphological count variable

Claims (9)

An acoustic dictionary, which is a set of acoustic models including at least a phonological model created from phoneme data analyzed from speech into acoustic parameter sequences and a prosodic model created from fundamental frequency data analyzed from speech, is stored for each type of speech. Acoustic dictionary storage means for
A sentence input means for inputting a sentence for generating speech;
The sentence that generates the speech input by the sentence input unit is decomposed into words to determine the part of speech, the accent type indicating the accent position is determined for each accent phrase, and the sentence that generates the speech is read Language analysis means to determine;
Analysis result storage means for storing the analysis result analyzed by the language analysis means;
Voice type designation means for designating one or a plurality of the voice types of the acoustic dictionary stored in the acoustic dictionary storage means;
Acoustic model selection means for selecting the acoustic model from the acoustic dictionary based on the analysis result stored in the analysis result storage means and the voice type designated by the voice type designation means;
A speech synthesizer comprising: speech generation means for generating speech based on the acoustic model selected by the acoustic model selection means.   The voice according to claim 1, wherein the voice generation unit generates a voice for each accent phrase, word, morpheme, or character based on the acoustic model selected by the acoustic model selection unit. Synthesizer.   The voice according to claim 1 or 2, wherein the voice type designation means includes a whole voice type designation means for designating one or a plurality of the voice types for the whole sentence generating the voice. Synthesizer.   4. The voice phrase designation means includes accent phrase-specific voice designation means for designating one or a plurality of the voice types for each accent phrase in a sentence that generates the voice. The speech synthesis apparatus according to any one of the above. The voice type designation means includes part-by-part voice type designation means for designating the voice types of the first part and the second part, respectively.
The voice generating means outputs the voice type specified in the second part after the voice of the voice type specified in the first part is output for each predetermined block of the sentence that generates the voice. 5. The apparatus according to claim 1, further comprising: a repetitive sound generating unit that outputs a sound and synthesizes the sound so that the sound of the second part repeats the sound of the first part. Voice synthesizer. The voice type designation means is
A multi-part audio type specifying means for specifying the audio types of a plurality of parts,
Order designating means for designating the order of outputting the sound among the plurality of parts,
The voice generation means includes
First, the sound of a predetermined block of the sentence that generates the sound of the first part in the order specified by the order specifying unit is output, and the second and subsequent parts of the part that are in the second order are the previous ones. Rotating to synthesize the sound so that the sound of each part circulates in the order specified by the order specifying means so that the sound output is started when the sound output of the first block of the part is completed. 6. The speech synthesis apparatus according to claim 1, further comprising speech generation means.   The speech synthesizer according to claim 5 or 6, wherein the block is at least one of a sentence, an accent phrase, a word, or a character delimited by a predetermined symbol.   The voice type designation means is a designation method for selecting one of the accent phrase voice designation means, the whole voice type designation means, the part-by-part voice type designation means, and the multi-part voice type designation means. The speech synthesizer according to claim 3, further comprising a selection unit.   A speech synthesis program for causing a computer to function as various processing means of the speech synthesizer according to claim 1.

Images