JPH04199195A - Voice synthesizer - Google Patents

Abstract

PURPOSE:To synthesize and output synthesized voices of a number string with good natularity by synthesizing voices by taking the number string as a unit if the information indicating the digit position, such as, for example, number particle or comma, is attached to the number string according to the expression type of the information contents possessed by the number string. CONSTITUTION:The character code string from an input section 1 is applied to a word collating section 2 where this cord string is collated with the accent information of the corresponding word and the information on a part of speech from an accent dictionary 3. This code string is then sent to a number string processing section 4. The number string is first decided by the presence or absence of the comma indicating the break of the digit in the processing section 4. The presence or absence of the specific number particle, such as YEN, BAN or HON, is further decided if such command is not included. Either case, this number string is taken as a unit and the unit reading mode to synthesize the voices is set if such comma or particle exists. The information on the unit word, such as tens thousands, thousand, hundred, or ten, accent and the Japanese rendering, is determined from a unit reading table 5 and is added after the number of the corresponding digit position. This information is outputted via a synthesis parameter synthesizing section 7 and a voice synthesizing section 8.

Description

DETAILED DESCRIPTION OF THE INVENTION [Purpose of the Invention (Field of Industrial Application) The present invention relates to a speech synthesis device that synthesizes and outputs information indicated by a character code string by a rule synthesis method.

(Prior Art) Recently, an input character code string is analyzed to obtain its phonetic and prosodic sequences, and based on the phonetic and prosodic parameters that are generated from this information according to predetermined rules. Various speech synthesis devices that synthesize and output information indicated by the character code string have been researched and developed. A speech synthesis device using this type of rule synthesis method has the feature that it can easily generate arbitrary words and sentences, compared to conventional speech synthesis devices using a general recording and editing method. For this reason, it is attracting attention as an important technology for realizing highly natural man-machine interfaces in conjunction with voice recognition processing technology.

Now, such a speech synthesis device using the rule synthesis method is used in fields such as outputting a sentence created with a word processor as voice, or outputting answer information for various arrangements in a voice response device as voice, etc. The number of users is gradually increasing.

However, conventional speech synthesis devices merely analyze a given character code string, synthesize the text information into speech, and output the result. Therefore, even if a numeric string is given, for example, the digits that make up the numeric string are simply vocalized one by one.

Recently, when synthesizing a number string into speech, scale processing is performed to convert unit words such as [10,000], [1000 ko, [100 ko], [+ko] to each digit according to the digit position of each digit in the numeric string. Attempts have been made to provide a mode for adding voice synthesis later. According to this scale processing technology, when a character code string consisting of r1234J etc. is given, for example, it is possible to scale it as "9n/nihyaku/?njiyuu/fun" and synthesize it into speech. becomes.

By the way, if scale is not used, the voice will be synthesized as "ichi/knee/san/shi".

However, such a voice synthesis mode using scaling is selectively set by the user and executed between a voice synthesis mode that does not perform scaling. This results in problems such as the mode not being switched often or the mode having to be switched frequently during the speech synthesis process.

(Invention or problem to be solved) Conventionally, when a string of numbers is given, there are two ways to synthesize speech: to set the scale and synthesize speech, or to synthesize speech from each number in order without setting scale. There was a problem in that it was necessary to switch the mode each time, and it was very troublesome to switch these modes appropriately and synthesize the number string into speech.

The present invention has been made in consideration of these circumstances, and its purpose is to perform speech synthesis by setting the scale of a number string depending on the meaning of the number string, or to synthesize speech without scaling the number string. The object of the present invention is to provide a highly practical speech synthesis device that can perform the following tasks.

[Structure of the Invention] (Means for Solving the Problems) A speech synthesis device according to the present invention has a character code string that is a string of numbers, and a character code string that accompanies (for example, follows) the string of numbers. When the specified edict is a numeric particle such as yen, ban, hon, etc., or when a symbol indicating a digit separator, such as a comma [,], is inserted in the string of numbers, each number in the string of numbers is After the number, [ according to the digit position]
This system is characterized in that by adding unit words such as [10,000], [1,000], [100], and [+], the number string is scaled and synthesized into speech.

(Function) According to the present invention, according to the analysis results of a number string, when a number particle such as [yenko, [ban], [hon], etc. follows the number string, or when there is a number particle in the number string, In cases where symbols such as commas [, ko, etc. are inserted to indicate digit separators, the number string is scaled and synthesized into speech. In other cases, the number string is directly synthesized into speech, so it is similar to the human speech pattern. It becomes possible to easily generate and output synthesized speech similar to that with good naturalness.

(Embodiment) Hereinafter, a speech synthesis device according to an embodiment of the present invention will be described with reference to the drawings.

FIG. 1 is a diagram showing a schematic configuration of an embodiment apparatus, and numeral 1 indicates an input section for inputting a coded character string or number string. The character code string input through this input section 1 is
It is given to the word matching unit 2 and used for word matching with reference to the accent dictionary 3.

The accent dictionary 3 has accent information, part of speech information, reading information, etc. of various words registered in advance, and the word matching unit 2 registers accent information, part of speech information, reading information, etc. of the word corresponding to the input character code string. etc. are determined from the accent dictionary 3 above.

Information obtained from the accent dictionary 3 regarding the words detected as a match by the word matching section 2 is provided to the number string processing section 4. This number string processing unit 4 checks the number string in the input character code string, and places it in the place reading table 5.
When synthesizing speech by scaling the number string in the character code string above while referring to
・It is used to find unit words such as [1000 ko, [100], shi+].

That is, the numeric string processing unit 4 first determines whether the character code string is a numeric string, for example, according to the processing procedure shown in FIG. 2 (step a).

This determination is made by checking whether a character code indicating a number consisting of [1, 2, ~0] or a comma [,] indicating a digit separator in a number string is given consecutively. Ru. If a numeric string is detected, then it is determined whether the numeric string contains a comma [,] or not (step b). If the numeric string does not contain a comma [,], then [Yen] specified in advance after the column
- [Number] - It is determined whether number particles such as [book] continue (Step C). If the number string contains a comma [, ko, or if a numeric particle follows the number string, set the place reading mode that scales the number string and synthesizes it into speech. In this case, a non-digitation mode is set in which speech synthesis is performed without scaling the number string (step e).

When the place reading mode is set by the above-described determination process, the number string processing unit 4 selects the place value from the place value table 5 according to the digit position of each number making up the number string.
Find unit words such as 1000, 100, and +, and their accent information and pronunciation information, and add each unit word after the number in the corresponding digit position. Then, the phonological information is obtained as the reading information of the numerical string scaled according to this information. This phoneme information is given to the synthesis parameter generation section 7 together with accent type information.

In the case of the non-place reading mode, the number string processing unit 4 performs phonetic conversion of each number in the number string according to the reading information, and converts each number into the synthesis parameter generation unit 7 together with the accent information obtained from the accent dictionary 3. give to

Therefore, in the synthesis parameter generation section 7, the number string processing section 4
A phoneme parameter sequence is generated by obtaining a phoneme segment corresponding to the phoneme sequence given from the phoneme file 6, and this is provided to the voice synthesis section 8.

The synthesis parameter generation section 7 generates a prosodic parameter string for the number string based on the above-mentioned accent type information, and supplies this to the speech synthesis section 8. According to such a phonetic parameter string and a prosodic parameter string, the speech synthesis section 8 generates and outputs speech expressing the number string by a rule synthesis method.

According to this device configured in this way, when a number string consisting of, for example, r1234J is input, it is possible to confirm that there is no comma [, ko] indicating a digit separator in the number string, and that there is no comma [, ko] that follows the number string. Since there is no character string, set non-displacement mode. Then, for example, each of "ichi-niichi" and "sanoyono" constitutes one accent phrase for the above-mentioned number string, and a "type 3" accent is given to these accent phrases. Such information is given to the synthesis parameter generation unit 7, the phoneme parameter sequence corresponding to the phoneme sequence is determined by referring to the speech segment file 6, and the prosodic parameter is generated according to the accent type. A synthesized voice saying "/san?n" will be generated.

Note that the [/ mark above indicates a break in an accent phrase, and the [ mark] indicates an accent position.

On the other hand, if a number string consisting of "r4321 yen" is input, the place reading mode is set because there is a character (character string) following the number string and the part of speech of "yen" is a number particle. be done. Then, reading information and accent type information for the number string are obtained from the place reading table 5, and at the same time, unit words corresponding to the digit positions of each number are obtained.

As a result, from the phonetic parameter string and the prosodic parameter string for the character string to which the unit word has been added, in this case, a synthesized speech of "Yonosen/Sanobyaku/Ninni/Ichen" is generated.

Further, when a number string consisting of r4.321J is input, the place reading mode is set because commas [, ko] indicating digit positions are inserted in the number string. Then, reading information and accent type information for the number string are obtained from the place reading table 5, and at the same time, unit words corresponding to the digit positions of each number are obtained. As a result, from the phonetic parameter string and the prosodic parameter string for the character string to which the unit word has been added, in this case, a synthesized speech of "Yonpun/♂Nbyaku/Ninyu/Ichi" is generated.

According to this device configured in this manner, depending on the type of word following the number string, if the following syntax is a predetermined numeric particle, or if a comma indicating a digit position is used in the number string, C, ] is inserted, it is determined that this is numerical information that should be automatically scaled and synthesized into speech, and it is determined that it is numeric information that should be automatically synthesized into speech, and it is determined that it is numeric information that should be synthesized by voice synthesis. Speech synthesis is performed. For other number strings, speech synthesis is performed as is without scaling. As a result, there is no need to incur the trouble of switching the mote of speech synthesis based on the user's specifications, and it is possible to
It becomes possible to synthesize and output numeric sounds while changing the scale appropriately.

Moreover, it is possible to automatically switch the voice synthesis mode depending on the situation in the sentence in which the number string appears, and to synthesize the number voice in a natural way, that is, in a form that is close to human speech. In practical terms, great effects can be achieved, such as:

Note that the present invention is not limited to the embodiments described above. For example, the algorithm for determining whether to switch the reading mode of a number string may be determined according to the specifications of the device, and the type of number particle to be read may be determined as appropriate. Furthermore, it is also possible to target not only the numbers following the number string but also numeric particles added before the number string, such as [¥ko and [$]. In addition, the present invention can be implemented with various modifications without departing from the gist thereof.

[Effects of the Invention] As explained above, according to the present invention, depending on the expression format of the information content of a numeric string, for example, when a number particle is attached or there is information indicating the digit position such as a comma, etc. In this case, the scale is set and the speech is synthesized, so that it is possible to produce a natural speech synthesis output of a number string according to the situation, which has a great practical effect.

[Brief explanation of the drawing]

FIG. 1 is a schematic configuration diagram of a speech synthesis device according to an embodiment of the present invention, and FIG. 2 is a diagram showing a procedure for determining a speech synthesis mode switching in the embodiment device. 1... Input section, 2... Word matching section, 3... Accent dictionary, 4... Number string processing section, 5... Place reading table, 6... Speech element file, 7. ...Synthesis parameter generation unit, 8...Speech synthesis unit. Kawagami's agent Patent attorney Takehiko Suzue (text input) (audio output) Figure 1

Claims (2)

[Claims] (1) In a speech synthesis device that analyzes a character code string and synthesizes information indicated by the character code string into speech using a rule synthesis method, the character code string is a string of numbers, and the character code attached to this string of numbers is used. A speech synthesis device characterized in that when a string indicates a pre-specified word or when a symbol indicating a digit separator is inserted in the string of numbers, the string of numbers is scaled and synthesized into speech. (2) The pre-designated word is a numeric particle, and the scale of a string of numbers is determined by adding a unit word after each number according to the digit position of each number in the string of numbers. A speech synthesis device according to claim (1).