JP2001034285A - Model sentence table database for insert voice synthesis and method for generating model sentence table - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make retrieval easy at the time of insert synthesis by generating a model sentence table database where meter data of a generated model sentence table are discriminatingly stored by uses or parts of speech. SOLUTION: The voice that, for example, an announcer speaks is recorded, the recorded voice is converted by an A/D converter 11, and the voice waveform converted by an intermediate data generation part 12 is viewed and sectioned by phonemes, labeled, and marked with synchronizing points by cycles of the voice waveform. A meter data extraction part 13 extracts meter data from the data marked with the synchronizing points by cycles of the voice waveform to generate a model sentence table. The generated model sentence table is adjusted by a model sentence table adjustment part 14 to obtain a model sentence table. Then the model sentence table which is generated and adjusted is stored to structure a model sentence database 15. The meter data are stored in the database 15 discriminatingly by usages and parts of speech, so retrieval is made easy at the time of insert synthesis.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a sentence example table database used for in-speech synthesis and a method for generating a new example table using the sentence example table database.

[0002]

2. Description of the Related Art FIG. 1 shows an example of a flowchart of a text-to-speech synthesis system using a sentence example table. In the figure, a Japanese text which is a character string of a sentence mixed with kanji and kana is subjected to language processing (S1). Here, the text is analyzed and divided into words, the pronunciation of the words and accents are given, and the words are converted into character strings of accented pronunciations.
For example, if the text is “Today is fine weather”, the text is converted to “Honshutsuhase” Itinari. At that time, the language dictionary A is used. Next, it is determined whether the prosody data corresponding to the character string exists in the sentence example table database B (S2). Prosody data is data for adding intonation to a character string. If the prosody data corresponding to the character string is in the sentence example table database B (Yes), the character string of the accented reading kana is inserted into this prosody data, and the inset synthesis is performed (S).
3). On the other hand, if there is no prosody data (No), rule synthesis is performed (S4). That is, based on the accented reading kana, the duration of the prosodic rule, the accent,
Determine intonation, poses, etc. In each case, waveform synthesis is performed using the waveform dictionary C (S
5) The synthesized voice is output.

[0003]

In the above system, the sentence example table database B is used for the inset synthesis.
Is used. When generating a sentence example table, for example, a voice uttered by an announcer is recorded, A / D conversion is performed on the recorded voice, a voice waveform is divided from the converted data into phonemes, labeled, and labeled for each cycle of the voice waveform. Mark a sync point. Then, prosody data, that is, data for giving the intonation to the character string is extracted from the data labeled and marked with the synchronization point to generate a sentence example table, which is stored as a sentence example table database.

[0004] However, as described above, it takes a great deal of time to generate a sentence example table. The sentence example table is composed of one sentence or a plurality of sentence units, or prosody data of elements constituting the sentence, and it takes time to search for the prosody data from the sentence example table when performing inlay synthesis. When a new text is created, added, or changed, it is necessary to regenerate the text example table.

[0005]

In order to solve the above problems, according to the present invention, a sentence example table database in which prosody data of a created example example table is stored for each use or part of speech is created and embedded speech synthesis is performed. A search can be performed from prosodic data stored separately in a sentence example table database.

Further, when a new sentence example table is generated, prosody data that has already been divided and stored is used, and these can be generated in combination.

[0007]

FIG. 2 is a diagram showing a configuration for generating a sentence example table. First, for example, a voice uttered by an announcer is recorded, the recorded voice is A / D-converted by an A / D converter 11, a voice waveform converted by the intermediate data generation unit 12 is viewed, and a break is inserted for each phoneme. And a synchronization point is marked for each period of the audio waveform. Then, the prosody data extracting unit 13 extracts prosody data from the data with the labels and the synchronization point marked for each period of the audio waveform, and generates a sentence example table. Next, the generated sentence example table is adjusted by the sentence example table adjustment unit 14 to obtain a sentence example table. The generated and adjusted sentence example table is stored as a sentence example table database 15.

FIG. 3 is a diagram showing an embodiment of a sentence example table database according to the present invention. For example, the following sentences are often used in traffic information. "The Tomei Expressway bound for Nagoya is 20km ahead of Isehara
In a traffic jam, "the sentence," Tomei Expressway "is a street name,
“Nagoya” is a direction, “Isehara” is a place name, “20 km” is a distance, and “traffic jam” is a situation.

The following sentences are often used in news reports. "○○ Mr. arrived in London to 11 o'clock." In the above sentence, "○○ Mr." is a person's name, "11
"Time" is a date, "London" is a place name, and "Arrival" is a situation. In this way, the sentence "person name",
Words such as "place name", "time", "distance", and "situation" are included according to the purpose, and prosody data is generated for each.

Therefore, in the embodiment, as shown in FIG. 3, a sentence example table database storing prosody data such as "person name", "place name", "time", "distance", and "situation" is constructed. FIG. 4 is a diagram showing another embodiment of the sentence example table database according to the present invention. For example,
The following sentences are often used in traffic information.

[0011] "Gulf line is 5km traffic jam." In the above sentence, "is" is a particle, "it is" is an auxiliary verb. These are "Gulf line" or the "traffic jam
This is the part of speech that appears at the end of the prosody data assigned to "". Therefore, in the present embodiment, as shown in FIG. 4, the prosody data of the parts of speech appearing at the end of these prosody data are collectively stored and constitute a sentence example table database.

FIG. 5 is a diagram showing another embodiment of the sentence example table database according to the present invention. In the embodiment shown in FIG. 4, the prosodic data of the parts of speech appearing at the end of the prosodic data are stored collectively, but in the present embodiment, the "particles" are further stored separately for each particle having the same accent pattern. It is. For example, as shown in FIG. 5, “ga”, “to”, “ni”, etc. are stored in section A, “no” is stored in section B, “ne” and “kana” are stored in section C, respectively. I do.

FIG. 6 is a diagram showing another embodiment of the sentence example table database according to the present invention. In the embodiment illustrated in FIG. 5, the prosody data is stored by dividing “particles” into particles having the same accent pattern. In the present embodiment, “auxiliary particles” are stored by dividing into auxiliary particles having the same accent pattern. Things. For example, as shown in FIG. 6, "da" is stored in section a, "is" and "look" are stored in section b, and "u" and "you" are stored in section c.

FIG. 7 is a diagram showing another embodiment of the sentence example table database according to the present invention. In the embodiment shown in FIG. 5 and FIG. 6, the prosody data is stored by dividing the “particles” or “auxiliary verbs” into patterns having the same accent, but in the present embodiment, as shown in FIG. And "auxiliary verbs" are classified into patterns having the same accent and store prosody data.

FIG. 8 is a diagram showing another embodiment of the sentence example table database according to the present invention. In this embodiment,
This is a combination of the classification of prosody data by use and the classification by part of speech. As shown in FIG. 8, "person name", "place name",
The prosody data in which data classified by use such as “time”, “distance”, and “situation” are combined with data classified into “particles” and “auxiliary verbs” are stored. For example, the prosody data corresponding to “Mr. XX” is stored by combining “person name” and “particles”. The prosody data corresponding to "congestion" is stored by combining "situation" and "auxiliary verb".

FIG. 9 is a diagram showing another embodiment of the sentence example table database according to the present invention. In the embodiment shown in FIG. 8, the classification of the prosody data by use and the classification by part of speech are stored in combination, but in this embodiment, as shown in FIG. 9, "person name", "place name", "time" , Prosodic data in which “particles” are classified into patterns having the same accents are stored in data classified according to uses such as “distance”.

FIG. 10 is a diagram showing another embodiment of the sentence example table database according to the present invention. In the present embodiment, as shown in FIG. 9, "person name", "place name", "time",
The prosody data in which "auxiliary verbs" are combined with those classified into the same accent pattern are stored in those classified by use such as "distance". FIG. 11 is a diagram showing another embodiment of the sentence example table database according to the present invention.
In the present embodiment, as shown in FIGS.
The prosody data in which "particles" and "auxiliary verbs" are combined with the ones in which the accents are divided into the same pattern is stored in the ones classified by use such as "place name", "time", "distance", and "situation".

For example, the prosody data is stored as a combination of "place name" + "particle", such as "Tokyo is a fine weather" or "situation" + "auxiliary verb". Next, a method of generating a new sentence example table using the sentence example table database will be described. FIG. 12 is a diagram showing a configuration of a sentence example table generating apparatus used in a method of newly generating a sentence example table using the sentence example table database. When the accented reading kana, which is input information, is input, the information analysis unit 21 performs processing such as dividing a character string into exhalation paragraphs. Next, the prosody data search unit 22 searches the prosody data from the sentence example table database 23 based on the character string cut in the exhalation paragraph. Next, a sentence example table is generated by combining the prosody data retrieved by the sentence example table generation unit 24. Then, the generated sentence example table is adjusted by the sentence example table adjustment unit 25 to be a sentence example table.

FIG. 13 is a flowchart for explaining a method of generating a text example table according to the present invention. When the generation of the sentence example table is started, the sentence example table database in which the prosody data is divided and stored is searched (S
1). For example, when creating a sentence example table of a sentence "○ is / ◇◇"("/" is an exhalation paragraph), it is determined whether or not there is a corresponding prosody data in this sentence example table (S2). In the case of Yes, that is, “○○” and “◇◇
If there is prosody data of "is," these prosody data are combined to generate a sentence example table of "XX is / ◇◇."

Next, the parameters of the prosody data are adjusted (S9). FIG. 14 shows a frequency pattern given between the above-mentioned exhalation paragraphs. The horizontal axis is time (t) and the vertical axis is frequency (f). For example, if the newly generated sentence example table is a sentence example table corresponding to the sentence “Your current location is / East side of Yokohama Station”, the frequency according to the pattern shown in FIG. It is given as prosody data of a character string between paragraphs, and the parameters of the prosody data are adjusted. For example, assume that the parameter of the first “current location” has a high frequency (high mountain shape in FIG. 14),
“Yokohama Station” has the next highest frequency, and “East” is the lowest frequency.

Next, it is determined whether necessary prosody data has been prepared, that is, whether all the prosody data of the newly generated sentence example table have been prepared (S5). In the case of No, the process returns to S1, but in the case of Yes, a sentence example table is generated by combining the retrieved prosody data or the prosody data synthesized by rules described later (S6). Finally, the accent and the like of the sentence example table are adjusted and corrected by trial listening (S7).

On the other hand, if there is no corresponding prosody data in S2 (No), it is determined whether there is prosody data of particles and auxiliary verbs of the same pattern (S3). In the case of Yes, prosody data of particles and auxiliary verbs is used (S8). For example,
When generating a sentence example table for the sentence "XX is / ◇◇.", Only the prosodic data of "XX is / ◇◇." It is assumed that there is no prosody data of "○○○". In this case, the sentence example table database does not have the prosody data of "XX", but has the prosody data of "XX" and "XX". As shown in FIG. 5, "ga" is classified into the same pattern A as "ha". Therefore, the prosody data of “XX” is used as the prosody data of “XX”. Next, as described above, the parameters of the prosody data are adjusted (S9).

In the above description, prosody data that is not in the sentence example table is generated by rule synthesis described later and used in combination. In the case of No in S3, that is, when there is no prosody data of particles and auxiliary verbs of the same pattern, prosody data is extracted from the speech generated by the rule synthesis (S4). That is, prosody data is generated based on phoneme length information and pitch information in accordance with an average Japanese rule extracted from various Japanese sentences with accented reading kana. Next, it is determined whether necessary prosody data has been prepared, that is, whether all the prosody data of the newly generated sentence example table have been prepared (S
5). If No, the process returns to S1, but if Yes, the prosody data is combined to generate a sentence example table (S1).
6). Finally, the sentence example table is adjusted and corrected by trial listening (S7).

[0024]

According to the present invention, the prosody data is classified by application.
Since it is stored in the database by being classified according to the part of speech, it becomes easy to search when performing inlay synthesis. In addition, when a new sentence example table is generated, the sentence example table can be generated using the prosody data already stored in the database, so that the generation is not troublesome and the time for generation can be reduced.

[Brief description of the drawings]

FIG. 1 is a diagram showing a flowchart of a text-to-speech synthesis system using a sentence example table.

FIG. 2 is a diagram illustrating a configuration for generating a sentence example table;

FIG. 3 is a diagram showing an embodiment of a sentence example table database according to the present invention.

FIG. 4 is a diagram showing another embodiment of a sentence example table database according to the present invention.

FIG. 5 is a diagram showing another embodiment of the sentence example table database according to the present invention.

FIG. 6 is a diagram showing another embodiment of the sentence example table database according to the present invention.

FIG. 7 is a diagram showing another embodiment of the sentence example table database according to the present invention.

FIG. 8 is a diagram showing another embodiment of the sentence example table database according to the present invention.

FIG. 9 is a diagram showing another embodiment of the sentence example table database according to the present invention.

FIG. 10 is a diagram showing another embodiment of the sentence example table database according to the present invention.

FIG. 11 is a diagram showing another embodiment of the sentence example table database according to the present invention.

FIG. 12 is a diagram illustrating a configuration of a sentence example table generating apparatus used in a method of newly generating a sentence example table using the sentence example table database of the present invention.

FIG. 13 is a flowchart illustrating a method of generating a new sentence example table using the sentence example table database according to the present invention.

FIG. 14 is a diagram illustrating an example of a frequency pattern given to prosody data.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 11 ... A / D converter 12 ... Intermediate data generation part 13 ... Prosodic data extraction part 14 ... Sentence example table adjustment part 15 ... Sentence example table database 21 ... Information analysis part 22 ... Prosody data search part 23 ... Sentence example table database 24 ... Sentence example table Generation unit 25: sentence example table adjustment unit

 ──────────────────────────────────────────────────続 き Continuing from the front page (72) Inventor Shinichi Nakaishi 1-2-28 Goshodori, Hyogo-ku, Kobe, Hyogo Prefecture Inside Fujitsu Ten Limited (72) Inventor Yoshitaka Hirashima 1 Goshodori, Hyogo-ku, Kobe-shi, Hyogo F-term (reference) in Fujitsu Ten Co., Ltd. 2-28 Chome 5D045 AA07 9A001 EE05 FF03 HH11 HH18 JJ01 JJ77 KK02

Claims (11)

[Claims] 1. A speech synthesis sentence example table database composed of sentence units or prosodic data of elements constituting a sentence used for generating a speech waveform in in-speech speech synthesis, wherein the database is divided into parts of speech at the end. A sentence example table database having stored prosody data. 2. The sentence example table database according to claim 1, wherein the part of speech is a particle and an auxiliary verb, and the prosody data includes a particle and an auxiliary verb in which accents are further classified and stored for the same pattern. 3. The sentence example table database according to claim 1, wherein the part-of-speech is a particle or an auxiliary verb, and has prosody data in which the particle or the auxiliary verb is further stored by classifying accents by the same pattern. 4. A speech synthesis sentence example database comprising sentence units or prosodic data of elements constituting a sentence used for generating a speech waveform in in-speech speech synthesis, wherein the sentence example table is classified according to the use at the time of use. A sentence example table database having stored prosody data. 5. A speech synthesis sentence example table database composed of prosody data of sentence units or elements constituting a sentence used for generating a speech waveform in in-speech synthesis, wherein the prosody is classified by use at the time of use. A sentence example table database having prosody data stored in combination with data and prosody data classified according to part of speech at the end. 6. A prosody which is a combination of a part-of-speech which is a particle and an auxiliary verb, and which is a combination of prosody data in which particles and auxiliary verbs are further divided into patterns having the same accent, and prosody data which is divided in accordance with the intended use. The sentence example table database according to claim 5 having data. 7. The prosody which is a combination of a prosody data in which the part of speech is a particle or an auxiliary verb, and the prosody data in which the particle or the auxiliary verb is further classified by the same accent and the prosody data which is classified by the use at the time of use. The sentence example table database according to claim 5 having data. 8. A method for generating a speech synthesis sentence example table composed of prosody data used for generating a speech waveform in in-speech speech synthesis, wherein the prosody data constituting the sentence example table is stored in a table. 7. A method of generating a sentence example table by searching from the sentence example table database described in any one of 7 and combining the searched prosody data. 9. A method for generating a speech synthesis sentence example table composed of prosody data used for generating a speech waveform in in-speech synthesis, wherein the prosody data constituting the sentence example table is stored. A method of generating a sentence example table by combining the searched prosodic data from the sentence example table database described in any one of 3, 6, and 7, wherein the prosody data of the searched particle or auxiliary verb is stored in the sentence example table database. If not, a sentence example table is generated using the prosody data of the particle or auxiliary verb in the sentence example table database in the same pattern as the particle or auxiliary verb in the same pattern. 10. The method according to claim 8, wherein a parameter of the retrieved prosody data is adjusted. 11. The method of generating a sentence example table according to claim 8, wherein as a result of the search, when prosody data is not in the database, the prosody data generated by rule synthesis is used.