JPH07140996A - Speech rule synthesizer - Google Patents

Abstract

PURPOSE:To easily correct the length of continuation time in a speech rule synthesizer. CONSTITUTION:A text analysis means 101 analyzes an inputted text and obtains an information relative to phonemes and rhythm. Based on the information, a pitch pattern generating means 102 generates the pitch pattern which indicates time variation of fundamental frequencies, a continuation time length generating means 103 generates the trains of continuation time length of each phoneme. An input means 111 inputs the pair of the phoneme trains and the continuation time length of each phoneme included in the phoneme train, a storage means 112 which stores the phoneme trains and the corresponding continuation time length trains. A waveform generating means 104 refers to the means 111 and 112 in accordance with the input of the phoneme trains and when a proper continuation time length train is retrieved, the continuation time length train from the means 103 is replaced by the continuation time length train retrieved from the means 112 and a replacement means 113 transmits the train to the means 104.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice rule synthesizing device used in a voice response system or the like.

[0002] The voice response system is a system that responds by voice to an inquiry by telephone regarding the contents of a database that changes from moment to moment such as recruitment information and ticket reservation information.

In such a voice response system,
Since the response contents are changed in various ways according to the change in the contents of the database, a speech rule synthesizing device capable of synthesizing a reading voice corresponding to an arbitrary input text is used.

[0004]

2. Description of the Related Art FIG. 13 shows the configuration of a conventional Japanese speech rule synthesizing device. In the speech rule synthesizing device, the text analysis unit 201 analyzes the input text with reference to the word dictionary 202 in accordance with the input of a kanji / kana mixed sentence, and generates reading information and prosody information. The reading information and the prosody information are provided to the pitch pattern generation unit 20.
3 and the duration length generation unit 204, and is held in the phoneme sequence buffer 205.

Here, the reading information indicates a phoneme sequence indicating an input text. For example, in response to input of a text "a telegram arrived at a bungalow early in the morning", "Asahaya yak bungalow Nidenpo" The reading information indicating the phoneme sequence "Ugato Doita" is obtained. On the other hand, the prosody information is information about intonation and accent when the corresponding phoneme sequence is pronounced, and is shown by adding an accent symbol or the like to the above-described phoneme sequence, for example.

The pitch pattern generator 203 analyzes accents and phrases indicated by reading information and prosody information, generates a pitch pattern showing a temporal change of the fundamental frequency when uttering the phoneme sequence, and generates a pitch pattern. It is sent to the pattern buffer 206. In addition, the duration generation unit 20
On the basis of the reading information and the prosody information, 4 generates a time length for continuing the phoneme for each phoneme included in the phoneme sequence and sends it to the duration buffer 207.

The waveform generation unit 208 includes a phoneme sequence buffer 2
Based on the phoneme sequence in 05, the phoneme waveforms corresponding to the phonemes are searched from the waveform dictionary 209, the frequencies of these phoneme waveforms are adjusted according to the pitch pattern, and the signals corresponding to the phonemes are respectively adjusted. Only the duration length is generated and sent to the speaker 210.

As described above, since the speech rule synthesizing device adjusts and combines the segment waveforms corresponding to each phoneme based on a predetermined rule, the speech rule synthesizing device corresponds to all phonemes composing Japanese. If one-sided waveform is registered in the waveform dictionary 209, it is possible to obtain a read voice corresponding to any input text.

The duration duration generation unit 204 described above has a correspondence table holding the average duration duration peculiar to the phoneme corresponding to each phoneme, and this correspondence table is provided according to the input of the phoneme sequence. After the basic duration is obtained by reference, the duration corresponding to each phoneme is adjusted in consideration of various conditions.

For example, the duration of each phoneme is adjusted according to the type of adjacent phonemes or the position in the expiratory paragraph which is a unit generated in one breath, or according to the number of mora in the expiratory paragraph. By adjusting the duration time corresponding to all the phonemes included in the corresponding phoneme sequence, it is possible to bring the rhythm of the synthesized speech closer to the rhythm when a human utters it, and obtain a more natural reading voice. .

[0011]

By the way, since an actual language includes many phonological sequences in which utterance is performed with an irregular rhythm, there are exceptions that cannot be adjusted under the above-mentioned conditions. In addition, a rhythmically unnatural reading voice may be synthesized. For example, in a sequence in which a phoneme with a long intrinsic phoneme and a phoneme with a short phoneme are intricately mixed, such as “Settsukutai Show” and “Joho Hawka Shakai”, the duration of each phoneme in synthetic speech and human utterance There is a large difference from the duration of each phoneme in the case of doing, and it often gives a lingering impression.

[0012] An unnatural part in the read-aloud voice is often found by the user, but since knowledge such as phonetics is necessary to set the rule for adjusting the duration, In effect, the user could not make the necessary modifications.

Therefore, conventionally, when the maker of the voice rule synthesizing device improves the device, various rules regarding the adjustment of the duration time are added to meet the request from the user. .

However, it is very difficult to set a rule for adjusting the duration time for all the exceptions. Also, since exceptional utterances are often irregular, it is difficult to extract the rules corresponding to each exception. Furthermore, by applying the rule for dealing with one exception, the duration of other phoneme sequences may be corrected to a bad one.

It is an object of the present invention to provide a speech rule synthesizing device capable of easily modifying the duration length corresponding to an arbitrary phoneme sequence.

[0016]

FIG. 1 is a block diagram showing the principle of a voice rule synthesizing apparatus according to claims 1 and 2.

According to the invention of claim 1, the text analysis means 10 is provided.
1 is a pitch pattern generation means 102 based on information about a phoneme sequence and prosody obtained by analyzing the input text.
And the duration generating means 103 respectively generate a pitch pattern representing a temporal change of the fundamental frequency and a sequence of durations of each phoneme, and the waveform generating means 104 synthesizes a voice based on these pieces of information. In the speech rule synthesizing device, an input means 111 for inputting a pair of a phoneme sequence and a duration of each phoneme included in the phoneme sequence and a storage means 112 for accumulating a sequence of durations corresponding to the phoneme sequence.
And referring to the accumulating means 112 according to the input of the phoneme sequence,
When searching for the corresponding series of durations, the means 1 accumulates the series of durations from the duration generating means 103.
It is characterized in that it is provided with a replacement unit 113 which replaces with the sequence of the duration length searched from 12 and sends it to the waveform generation unit 104.

According to a second aspect of the present invention, in the voice rule synthesizing apparatus according to the first aspect, the input means 111 receives the information indicating the phoneme sequence corresponding to the synthesized voice from the waveform generating means 104, and the input text. A display unit 121 for displaying at least a portion corresponding to a phoneme sequence among them, and a phoneme sequence displayed at the corresponding position is extracted according to an input of a position on the display screen of the display unit 121, and stored in the storage unit 112. And a phoneme sequence extraction means 122 for transmitting.

FIG. 2 is a block diagram showing the principle of the speech rule synthesizing apparatus of claim 3. According to a third aspect of the present invention, in the voice rule synthesizing apparatus according to the first aspect, the input unit 111 selects a corresponding rule from a plurality of preset rules according to the input of a selection instruction. And a correction unit 132 that corrects the duration length by applying the rule selected by the rule selection unit 131 according to the input of the phoneme sequence and sends the obtained correction value to the storage unit 112. Is characterized in that.

FIG. 3 is a block diagram showing the principle of the speech rule synthesizing apparatus of claim 4. According to a fourth aspect of the invention, in the voice rule synthesizing apparatus according to the first aspect, the input unit 111 receives a voice corresponding to a phoneme sequence, and, in response to an input of a boundary designation instruction, outputs each phoneme of a waveform of the voice It is characterized by including a boundary setting means 141 for setting a boundary position and a duration length extraction means 142 for extracting the duration length of each phoneme based on the boundary position of each phoneme in the waveform of the voice. To do.

FIG. 4 is a block diagram showing the principle of the speech rule synthesizing device of the fourth aspect. According to a fifth aspect of the invention, in the voice rule synthesizing apparatus according to the first aspect, the input unit 111 inputs the voice corresponding to the phoneme sequence and the synthesized voice generated by the waveform generating unit 104 corresponding to the phoneme sequence. Based on the matching unit 143 that associates each phoneme in the voice with each phoneme in the synthesized voice and sets the boundary position of each phoneme in the voice, and the boundary position of each phoneme in the voice, The present invention is characterized in that it is provided with a duration extraction unit 142 for extracting the duration.

[0022]

According to the invention of claim 1, the accumulating means 112 accumulates the phoneme sequence inputted through the input means 111 and the corresponding duration length sequence, and the replacing means in accordance with the input of the corresponding phoneme sequence. Reference numeral 113 is for outputting a series of durations in the storage means 112. Therefore, the user inputs a phoneme sequence having a rhythmically unnatural impression in units of syllables or accent phrases, and by inputting a sequence having an appropriate duration, the user can continue the sequence for any phoneme sequence. The length of time can be corrected.

In this case, since it is not necessary to extract the rule for correcting the duration time, the user can relatively easily correct the duration time. Also, since the correction value of the duration is applied only to that phoneme sequence, unlike the case where a rule for correcting the duration is added,
It does not adversely affect the duration sequence of other phoneme sequences.

Further, by accumulating the correction values corresponding to various phoneme sequences in the accumulating means 112, it becomes possible to flexibly cope with various uses and generate high-quality synthetic speech.

According to the second aspect of the present invention, the display means 121 displays at least a part of the input text in synchronization with the operation of generating the synthesized voice by the waveform generation means 104.
Information about the correspondence between the synthesized voice and the corresponding portion of the text can be specified to the user. Further, the phoneme sequence extraction means 122 extracts the phoneme sequence corresponding to the designated portion of the text based on the position on the display screen, so that the input work of the phoneme sequence can be simplified.

According to the third aspect of the present invention, the correction means 132 corrects the corresponding duration length sequence in accordance with the rule selected by the rule selection means 131, thereby inputting the duration length sequence itself. It becomes possible to replace the work with the work of selecting the rule, and it is possible to simplify the work of correcting the duration time.

According to the fourth aspect of the present invention, by inputting a boundary designating instruction to the boundary setting means 141, the boundary position of each phoneme in the input speech waveform is set, and the duration length extracting means 142 sets the duration time series. Can be extracted.

In this case, since the correction value of the duration time can be input as a voice, it is not necessary to repeat trial and error, and the work of correcting the duration time can be simplified.

Further, according to the invention of claim 5, the matching means 143 automatically sets the boundary position of each phoneme in the input voice waveform by associating each phoneme with the input voice and the synthesized voice. It is possible to further simplify the work of correcting the duration time.

[0030]

Embodiments of the present invention will now be described in detail with reference to the drawings. FIG. 5 shows a block diagram of an embodiment of the speech rule synthesizing apparatus of claim 1.

In FIG. 5, the speech rule synthesizing device is shown in FIG.
In the conventional speech rule synthesizing device shown in FIG. 3, the correction sequence buffer 211, the correction value creating unit 212, and the correction value database 2 are added.
13, a duration correction unit 214, and a user interface unit 220 are added.

In FIG. 5, the user interface section 220 includes a keyboard 221, a mouse 222, an input analysis section 223, a display section 224, and a display data creation section 225.
It consists of and. In the user interface unit 220, the input analysis unit 223 is provided with the keyboard 22.
1 and the instruction input from the mouse, the text analysis unit 201, the waveform generation unit 208, and the correction value generation unit 2
It is configured to send the necessary instructions to 12. Also,
The phoneme sequence to be corrected input from the keyboard 221 is corrected by the correction sequence buffer 211 via the input analysis unit 223.
Sent to. The input analysis unit 223 also displays the display data creation unit 22 in response to an instruction to correct the phoneme duration.
5 is started, and in response to this, the display data creation unit 225
However, a display screen (described later) that assists in correcting the phoneme duration
The display data for displaying is generated and sent to the display unit 224.

In addition, the correction value creation unit 212 creates a correction value for the duration of the phoneme designated by the processing described later in response to the instruction from the input analysis unit 223 and sends it to the duration buffer 207. Then, in response to the instruction to determine the correction value, the correction value of the duration time is sent to the correction value database 213 together with the corresponding phoneme sequence.

The correction value database 213 is stored in the storage means 1.
This is equivalent to 12 and is configured to accumulate a pair of a phoneme sequence from the correction value creation unit 212 and a correction value of the duration time.

The duration correction unit 214 corresponds to the replacing unit 113, searches the correction value database 213 according to the input of the phoneme sequence, and stores the corresponding duration correction values. If there is, the duration length obtained by the duration length generation unit 204 is replaced with a correction value and the result is sent to the duration length buffer 207.

In FIG. 5, the waveform generator 208
Is equivalent to the waveform generation means 104, and performs a normal waveform generation process and a waveform generation process in response to an instruction from the input analysis unit 223 in response to an instruction to correct the duration time. May be configured to execute.

Further, the text analysis unit 201, the pitch pattern generation unit 203, and the duration length generation unit 204 respectively include the text analysis unit 101 and the pitch pattern generation unit 1.
02, which corresponds to the duration length generating means 103.

The correction process of the duration time of the present invention will be described below. FIG. 6 is a flowchart showing the operation of inputting the phoneme sequence to be corrected and the correction value of the duration.

When a rhythmically unnatural synthesized voice is found, the user is required to use the keyboard 221 or the mouse 22.
Instruct to correct the duration length via 2 and then
For example, a character string indicating the phoneme sequence to be corrected is input in accent phrase units (or syllable units) via the keyboard 221 (step 301).

The input analysis unit 223 uses this step 301.
The character string input in (1) is sent to the correction series buffer 211, and the text analysis unit 201 is instructed to analyze the character string in the correction series buffer 211.
In response to this, the text analysis unit 201, the pitch pattern generation unit 202, and the duration length generation unit 204 generate the phoneme sequence, the pitch pattern, and the duration length in the same manner as the ordinary text, and the phoneme sequence buffer 205, respectively. ，
It is held in the pitch pattern buffer 206 and the duration buffer 207.

Next, the input analysis section 223 uses the waveform generation section 2
08 is instructed to generate a synthetic voice, and in response to this, the waveform generation unit 208 generates a corresponding synthetic voice based on the contents of the phoneme sequence buffer 205, the pitch pattern buffer 206, and the duration buffer 207, and outputs the generated voice to the speaker. 21
It is output to 0 (step 302).

At this time, the input analysis section 223 activates the display data creation section 225, and in response to this, the display data creation section 225 causes the display data creation section 225 to display the waveform data obtained by the waveform generation section 208 and the duration buffer. Display data is created based on the contents of 207 and sent to the display unit 224 (step 303).

The display data creation unit 225, for example, together with the display data representing the waveform of the above-mentioned synthesized voice, based on the contents of the duration buffer 207, the boundary position of each phoneme in the waveform of the synthesized voice and continuation of each phoneme. Display data indicating time is created, and these display data are displayed on the display unit 2.
It is sufficient to send it to 24. As a result, the display unit 224 obtains a display screen as shown in FIG. 7 for assisting the input of the correction value.

In FIG. 7, as an example of the waveform of the synthetic voice, a part of the waveform of the synthetic voice corresponding to the phonological sequence "Settsukutaisho" is shown. Further, in FIG. 7, the phonemes that make up each phoneme are shown in Roman letters, and the corresponding waveform portions are separated by solid lines to show the respective durations.

The user may operate the mouse 222 while looking at the display screen as shown in FIG. 7 to indicate the change of the boundary position of each phoneme. At this time, the input analysis unit 223
The position data input by the user operating the mouse 222 is sent to the display data creating unit 225 and the correction value creating unit 212. In response to this, the display data creation unit 225 updates the display data, while the correction value creation unit 2
Reference numeral 12 obtains a change amount of the boundary position of the corresponding phoneme from the position data described above, converts the change amount of the boundary position into a change amount of the duration time of each phoneme, and calculates a correction value of the duration time. The contents of the duration buffer 207 are updated (steps 304 and 305).

That is, the user interface section 220
The correction value creation unit 212 realizes the function of the input unit 111, and can input the phoneme sequence to be corrected and the correction value of the duration time. in this case,
The duration time can be corrected for each phoneme forming each phoneme.

Further, at this time, the input analysis section 223 instructs the waveform generation section 208 to generate a synthetic voice, and in response thereto, the correction target of the correction target is corrected using the correction value of the duration length generated in step 305. The synthesized speech corresponding to the phoneme sequence of is generated (step 306).

By listening to the synthesized voice obtained in step 306, the user may determine whether or not the rhythm of the synthesized voice is appropriate, and input the determination result by operating the mouse 222 or the keyboard 221. .

When the judgment result indicating that the synthesized voice is not appropriate is inputted, the negative judgment is made in step 307, the flow returns to step 303 again, and the correction processing by the new correction value is repeated.

Steps 303 to 307 described above
By repeating the above, it becomes possible for the user to correct the series of the duration length while repeating trial and error. In this way, by correcting the series of durations, when a natural synthesized speech is obtained, a determination result indicating that the synthesized speech is appropriate may be input.

In this case, an affirmative decision is made in step 307, and the correction value creating section 212 sends the calculated correction value and the corresponding phoneme sequence to the correction value database 213 (step 308). As a result, the correction value of the duration length corresponding to the corresponding phoneme sequence is stored in the correction value database 2
Then, the correction processing of the duration time may be finished.

In this way, when synthetic speech is perceived as rhythmically unnatural, by inputting the corresponding phoneme sequence and the correction value of the corresponding duration length sequence, the correction value database Phonological sequences as shown in Table 1 (Table 1
In Romanized notation) and the duration of each phoneme is stored as a pair.

[0053]

[Table 1]

Next, a method of correcting the duration time using the correction value database 213 accumulated as described above will be described. FIG. 8 is a flowchart showing the voice synthesis operation of the present invention.

The text to be read is input in the same manner as in the conventional case (step 401), and the text analysis unit 201, the pitch pattern generation unit 203, and the duration length generation unit 204 perform information and pitch regarding the phoneme sequence and prosody corresponding to the input text. A pattern and a duration of each phoneme are generated (step 402).

Next, the correction processing unit 214 proceeds to step 40.
With respect to the phoneme sequence obtained in 2, the correction value database 213 is referred to for each accent phrase, and a corresponding correction value is searched (step 403).

If the corresponding correction value is obtained in step 403, an affirmative decision is made in step 404,
The series of the corresponding duration in the duration buffer 207 is replaced with the correction value retrieved from the correction value database 213 (step 405), and the process proceeds to step 406. On the other hand, in the case of negative determination in step 404, the process may proceed to step 406 as it is.

In this way, after the correction processing for each accent phrase is completed, it is determined whether the correction for all accent phrases is completed (step 40).
6) If the determination is negative, the process returns to step 403 and the correction process for the next accent phrase is performed. If the determination in step 406 is affirmative, the phoneme sequence buffer 205,
Pitch pattern buffer 206, duration buffer 2
Based on the contents of 07 and the waveform dictionary 209, the waveform generation unit 208 may generate a synthesized voice (step 407) and end the voice synthesis process.

As described above, for the phoneme sequence subjected to the correction process, the sequence of the duration time obtained by the duration time generation unit 204 is replaced with the correction value of the duration time, so that another phoneme sequence is obtained. It is possible to obtain natural synthesized speech for the corresponding phoneme sequence without adversely affecting it.

Further, in this case, since it is not necessary to derive the rule from the phenomenon, the correction can be easily made by the user. By accumulating the correction values in the correction value database 213, various input texts can be obtained. It is possible to respond flexibly to.

In this way, the user can easily modify the duration of the phoneme, so that the synthetic speech that gives an unnatural impression is effectively corrected, and the quality of the synthesized speech produced by the speech rule synthesizing device is improved. Can be improved.

Since the input text analysis process, the duration replacement process, and the synthetic voice generation process can be performed independently of each other, these processes may be performed in parallel with an appropriate time difference. Good.

Next, an embodiment of the speech rule synthesizing apparatus of claim 2 will be described. FIG. 9 is a block diagram of an embodiment of the speech rule synthesizing apparatus of claim 2. In FIG. 9, the speech rule synthesizing device adds a text buffer 215 to the speech rule synthesizing device shown in FIG. 5, and replaces the keyboard 221 of the user interface unit 220 with a text display data creating unit 226 and a correction sequence extracting unit. And a part 227.

The above-mentioned text buffer 215 holds the input text and at the same time the text analysis unit 20.
It is configured to be sent to 1. Further, the text display data creation unit 226 receives the notification of the phoneme sequence that is the target of the waveform generation processing from the waveform generation unit 208, and based on this notification and the contents of the text buffer 215, the waveform generation unit 2
The text display data for displaying the text indicating the corresponding phoneme sequence is created in synchronization with the voice synthesis operation by 08.

This text display data creating section 226
For example, the character codes included in the text in the text buffer 215 may be arranged according to the display screen to create text display data, and the text display data may be sent to a display data buffer (not shown) in the display unit 224. Further, in accordance with the notification of the phoneme sequence described above, the corresponding display attribute data in the display data buffer described above may be updated for the text display data of the part corresponding to the corresponding accent phrase or syllable.

The display unit 224 performs a display operation based on the text display data obtained in this way, in synchronization with the voice synthesis operation by the waveform generation unit 208.
Input text can be displayed.

Further, in this case, the input analysis unit 223
When the mouse 222 is operated by the user to specify the range on the display screen, the correction sequence extraction unit 227 may be notified of the position on the display screen. In response to this, the correction series extraction unit 227 refers to the above-mentioned display data buffer to extract the text data corresponding to the specified range, and the correction series buffer 211
It may be configured to send to the.

That is, the text display data creation unit 22
6 and the display unit 224 realize the function of the display unit 121, and the mouse 222, the input analysis unit 223, and the correction sequence extraction unit 227 realize the function of the phoneme sequence extraction unit 122.

In this case, the display unit 224 clearly indicates the correspondence between the synthesized voice and the corresponding portion of the text. Therefore, the user can see the text displayed on the display unit 224 to display the waveform generation unit. It is possible to intuitively understand the text portion corresponding to the voice synthesized in 208. Further, by operating the mouse 222 to specify the corresponding portion, it is possible to easily specify an accent phrase or syllable having an unnatural impression, so that the input work of the phonological sequence is easily performed. be able to.

By the way, in the above phoneme duration correction work, the duration of all phonemes included in the accent phrase can be corrected, and since the degree of freedom of correction is great, no knowledge of phonetics is given. For users,
On the contrary, it may be difficult to correct.

On the other hand, the operation of selecting the most appropriate one from several modifications can be easily performed by a user who does not have specialized knowledge. The invention of claim 3 is to replace the phoneme duration correction work with a selection work from the correction examples to provide a system that is easy to use for users who do not have specialized knowledge.

FIG. 10 shows a block diagram of an embodiment of the speech rule synthesizing apparatus of claim 3. In FIG. 10, the speech rule synthesizing device is the correction value creation unit 2 of the speech rule synthesizing device shown in FIG.
Instead of 12, the number of rule correction units 216 _{1 to} 216 _n , a selector 217, and a demultiplexer (denoted by DMPX in FIG. 10) 218 are provided, and the demultiplexer 218 outputs the output of the selector 217 to the duration buffer 2.
07 or correction value database 213.

The rule modifying units 216 _{1 to} 216 _n are configured to modify the duration of each phoneme by using different rules according to the input of the phoneme sequence and the sequence of the corresponding duration. There is.

As the rules applied by the rule correction units 216 _{1 to} 216 _n , the following can be considered, for example. (a) Selectively correct the duration of repellant and consonant sounds. (b) The average duration is approximated according to the weight set for each phoneme. (c) Selectively reduce the duration of continuous unvoiced fricatives and unvoiced plosives. (d) Selectively extend the duration of consecutive voiced fricatives and voiced plosives.

In this case, the display data creating section 225
In response to an instruction from the input analysis unit 223, the rule correction unit 21
It suffices to create display data representing the contents of the rules applied by 6 _{1 to} 216 _n and the parameters that can be set, and send it to the display unit 224 to prompt the user to select the rules and input the parameters.

Further, when the user operates the mouse 222 or the like to input the rule and the parameter to be selected, the input analysis section 223 causes the corresponding rule correction section 21 to operate.
It is sufficient to send out the parameters input to No. 6 and to instruct the selector 217 to select the output of the corresponding rule modification unit 216.

As a result, the correction value of the duration length obtained by applying the rule selected by the user to the phoneme sequence to be corrected is obtained and sent to the duration buffer 207 via the demultiplexer 218. By the waveform generation unit 208,
A synthetic voice is generated using this modified value.

That is, the rule correction units 216 _{1 to} 216 _n
Further, the selector 217 is configured to realize the same function as that of the rule selection means 131 and the correction means 132.

In this case, the user listens to the synthesized voice to confirm the correction result, changes the rule to be applied, changes the parameter to be set, and repeats trial and error to obtain the best one. At this time, the mouse 222 or the like may be operated to give an instruction to determine the correction result as the correction value.

In response to this, the input analysis unit 223 switches the demultiplexer 218 to send the output of the selector 217 to the correction value database 213, and the user selects the best correction result for the duration time. It may be accumulated as a correction value of the series.

As described above, by limiting the types of phonemes to be modified according to the rules selected by the user, excessive freedom can be reduced, and even users without specialized knowledge can continue effectively. It is possible to correct the time length series. In addition, it is possible to flexibly cope with various phoneme sequences by making it possible to change the parameters and weights for expanding and contracting the duration length for each rule used for correction.

On the other hand, it is also possible that the user utters the phoneme sequence to be corrected and extracts the correction value of the duration time from this voice information. FIG. 11 shows a block diagram of an embodiment of the speech rule synthesizing apparatus of claim 4.

In FIG. 11, the speech rule synthesizing device includes a duration extracting unit 219 instead of the correction value creating unit 212 of the speech rule synthesizing device shown in FIG. 5, and a user interface unit 220 for speech input. The microphone 228 and the voice input buffer 229 are added, and the voice input through the microphone 228 is held in the voice input buffer 229.

Further, in this case, the display data creating section 22
5 displays the waveform of the input voice held in the voice input buffer 229, instead of the waveform of the synthesized voice obtained by the waveform generation unit 208. In addition, information indicating the position designated by the mouse 222 is input via the input analysis unit 223, and in response to this, the display data creation unit 225 displays the boundary position of each phoneme in the waveform of the input speech described above. Create the display data.

In this way, the display data creating section 225 performs a display corresponding to the input voice and inputs the boundary position of each phoneme in the waveform via the mouse 222, thereby realizing the function of the boundary setting means 141. The boundary position of each phoneme in the input voice can be set.

The duration extraction unit 219 corresponds to the duration extraction unit 142, and based on the display data created by the display data creation unit 225, the boundary position of each phoneme set as described above. Is calculated, the duration of each phoneme is calculated from these boundary positions, and registered in the correction value database 213.

As a result, the user utters the phoneme sequence to be corrected, corrects the sequence of duration length corresponding to the corresponding phoneme sequence based on this voice information, and stores the correction value in the correction value database 213. You can accumulate.

In this case, if the boundary position of the phoneme is appropriately indicated in the waveform of the input voice, a series of duration length extremely close to human utterance can be obtained, and it is not necessary to repeat trial and error. Therefore, the correction work of the duration time can be performed quickly and appropriately.

Further, in the display data creating section 225,
Based on the sequence of durations generated by the duration generator 204, display data indicating the delimiter position of each phoneme before correction is created, and the above-described input is performed in the same manner as the display screen shown in FIG. It may be displayed together with the sound waveform.

In this case, the user can be given an approximate guideline about the delimiter position of each phoneme included in the input voice, so that the work of inputting the boundary of the phoneme in the input voice can be made easier. .

It is also possible to automatically set the boundary position of the phoneme by comparing the waveform of the synthesized voice before correction with the waveform of the input voice. FIG. 13 shows a block diagram of an embodiment of the speech rule synthesizing device of the fifth aspect.

In FIG. 13, the voice rule synthesizing device is the display data creating section 22 of the voice rule synthesizing device shown in FIG.
Instead of 5, the matching processing unit 231 is provided.

The matching processing section 231 corresponds to the matching means 143, and uses, for example, dynamic programming (DP, Dynamic Programming) to form the waveform and waveform of the input voice held in the voice input buffer 229. Generator 2
It is configured to match the waveform of the synthetic voice generated in 08.

As a result, the boundary position of each phoneme in the input voice can be set by automatically associating each phoneme in the input voice with each phoneme in the synthetic voice, so that the duration extraction unit 219 Matching processing unit 23
The duration length may be calculated based on the boundary position of each phoneme obtained in 1 and registered in the correction value database 213.

In this case, the sequence of the corresponding duration can be corrected simply by designating the accent phrase to be corrected and uttering the accent phrase. Therefore, the correction process of the duration is simpler. It is possible to provide a system that is easy for users to use.

[0096]

As described above, according to the present invention, the phoneme sequence to be corrected and the correction value of the duration time corresponding to each phoneme are input, and the phoneme sequence and the correction value are stored as a pair. , When synthesizing the reading voice of the corresponding phoneme sequence,
The accumulated correction value can be applied. As a result, when a user discovers a phoneme sequence that feels rhythmically unnatural, it can be easily corrected without adversely affecting other phoneme sequences, and can be flexibly adapted to the user's application. As a result, it is possible to provide a speech rule synthesizing device that generates high quality synthetic speech.

According to the invention of claim 2, at least a part of the text is displayed in synchronization with the voice synthesis operation.
Clarify the correspondence between the synthesized voice and the corresponding part of the text,
The work of inputting the phoneme sequence to be corrected can be simplified.

Further, the invention of claim 3 can further simplify the correction work by replacing the work of creating the correction value of the duration time with the work of selecting one from several candidates. As a result, it is possible to provide a voice rule synthesizing device having a system that is easy to use even for a user who does not have specialized knowledge.

Further, the inventions of claims 4 and 5 are
By inputting the voice when the user utters the phoneme sequence to be corrected, by extracting the boundary position of each phoneme in this input voice according to the user's instructions or according to the matching result with the synthesized voice, The correction value of the duration time can be obtained. As a result, it is possible to provide a voice rule synthesizing device having a system that simplifies the correction work and is easy to use even for users who do not have specialized knowledge.

[Brief description of drawings]

FIG. 1 is a principle block diagram of a voice rule synthesizing apparatus according to claims 1 and 2.

FIG. 2 is a block diagram showing the principle of the speech rule synthesizing device of claim 3;

FIG. 3 is a block diagram showing the principle of the speech rule synthesizing device of claim 4;

FIG. 4 is a block diagram showing the principle of the speech rule synthesizing device of claim 5;

FIG. 5 is a configuration diagram of an embodiment of a voice rule synthesizing device according to claim 1;

FIG. 6 is a flowchart showing an operation of inputting a correction value of a duration time.

FIG. 7 is an explanatory diagram of a display screen that assists in inputting a correction value.

FIG. 8 is a flowchart showing a voice synthesizing operation of the present invention.

FIG. 9 is a configuration diagram of an embodiment of a voice rule synthesizing device according to claim 2;

FIG. 10 is a configuration diagram of an embodiment of a voice rule synthesizing apparatus according to claim 3;

FIG. 11 is a configuration diagram of an embodiment of a voice rule synthesizing device according to claim 4;

FIG. 12 is a configuration diagram of an embodiment of a voice rule synthesizing apparatus according to claim 5;

FIG. 13 is a block diagram of a conventional speech rule synthesizing device.

[Explanation of symbols]

 101 Text Analysis Means 102 Pitch Pattern Generation Means 103 Duration Length Generation Means 104 Waveform Generation Means 111 Input Means 112 Accumulation Means 113 Replacement Means 121 Display Means 122 Phoneme Sequence Extraction Means 131 Rule Selection Means 132 Correction Means 141 Boundary Setting Means 142 Duration Times Length extraction means 143 Matching means 201 Text analysis section 202 Word dictionary 203 Pitch pattern generation section 204 Duration duration generation section 205 Phonological sequence buffer 206 Pitch pattern buffer 207 Duration duration buffer 208 Waveform generation section 209 Waveform dictionary 210 Speaker 211 Correction sequence buffer 212 correction value creation unit 213 correction value database 214 duration correction unit 215 text buffer 216 rule correction unit 217 selector 218 demultiplexer DMPX) 219 duration extraction unit 220 the user interface unit 221 keyboard 222 mouse 223 input analyzer 224 display unit 225 display data producing unit 226 text display data producing unit 227 correcting sequence extraction unit 228 microphone 229 audio input buffer 231 matching processor

Claims (5)

[Claims] 1. A pitch pattern generating means (102) and a duration length generating means (103) are based on information about a phoneme sequence and prosody obtained by analyzing the input text by a text analyzing means (101). In a speech rule synthesizing device in which a pitch pattern representing a temporal change in frequency and a sequence of durations of phonemes are respectively generated, and a waveform generation means (104) synthesizes a speech based on these information, An input unit (111) for inputting a pair with a duration time of each phoneme included in the phoneme sequence, a storage unit (112) for accumulating the phoneme sequence and a sequence of a corresponding duration time, and a phoneme sequence The series of durations from the duration generating means (103) is stored when the series of the corresponding durations is searched by referring to the accumulating means (112) according to the input. A speech rule synthesizing device comprising: a replacement unit (113) for replacing with a sequence of durations retrieved from the multiplication unit (112) and sending it to the waveform generation unit (104). 2. The voice rule synthesizing apparatus according to claim 1, wherein the input unit (111) receives information indicating a phoneme sequence corresponding to the synthesized voice from the waveform generation unit (104), A display unit (121) for displaying at least a portion corresponding to the phoneme sequence, and a phoneme sequence displayed at the corresponding position is extracted in response to an input of a position on the display screen of the display unit (121). And a phoneme sequence extraction means (122) for sending to a storage means (112). 3. The voice rule synthesizing apparatus according to claim 1, wherein the input unit (111) selects a corresponding rule from a plurality of preset rules in response to the input of the selection instruction. 131) and the rule selection means (131) according to the input of the phoneme sequence.
A voice rule synthesizing system, comprising: a correction means (132) for applying the rule selected in step 1 to correct the duration and sending the obtained correction value to the storage means (112). apparatus. 4. The voice rule synthesizing apparatus according to claim 1, wherein the input unit (111) receives a voice corresponding to a phoneme sequence, and each phoneme in the waveform of the voice according to an input of a boundary designation instruction. Boundary setting means (141) for setting the boundary position of the phoneme, and based on the boundary position of each phoneme in the waveform of the voice,
Duration extracting means (1) for extracting the duration of each phoneme
42) and a voice rule synthesizing device. 5. The speech rule synthesizing apparatus according to claim 1, wherein the input unit (111) generates a voice corresponding to a phoneme sequence and a synthesized voice generated by a waveform generation unit (104) corresponding to the phoneme sequence. Is input, each phoneme in the voice is associated with each phoneme in the synthesized voice, and a matching unit (143) that sets a boundary position of each phoneme in the voice, and a boundary position of each phoneme in the voice. A speech rule synthesizing device comprising: a duration duration extracting means (142) for extracting duration duration of each phoneme in the speech based on the above.