JPH01303498A - Accent dictionary generating device - Google Patents

Abstract

PURPOSE:To easily and accurately specify a correct accent regarding a word and to register it in a dictionary efficiently by outputting the synthesized voice of the word to be registered in order while its accent is made different, and specifying accent information according to the monitor result of the synthesize voice. CONSTITUTION:In the accent dictionary 1, the accent information on the word to be synthesized in voice is registered and stored by words. When a character string indicating the word to be registered is supplied, a voice synthesis part 2 analyzes the character string to obtain a phoneme parameter series and a rhythm parameter series respectively and also obtains accent information regarding the character string by referring to the accent dictionary 1, thereby synthesizing and outputting the voice under a specific rule according to those pieces of information. Consequently, the synthesized voice of the word to be registered which is indicated by single-time key operation is outputted in order while made different in accent automatically and while the difference in accent is easily contrasted with efficiency, the correct accent is accurately specified and the word is registered in the dictionary efficiently.

Description

[Detailed Description of the Invention] [Objective of the Invention] (Field of Industrial Application) The present invention easily and accurately registers word-specific accent information used to generate highly natural synthesized speech in an accent dictionary. This invention relates to an accent dictionary creation device that can create accent dictionaries.

(Prior Art) Recently, a rule has been developed for inputting a character code string indicating word information and generating synthesized speech based on a predetermined rule according to a phonological parameter string and a prosodic parameter string obtained by analyzing this character code string. Speech synthesis devices using synthesis methods are attracting attention. This type of speech synthesis device has an excellent advantage in that it can easily generate arbitrary words and sentences compared to conventional sound recording and editing type speech synthesis devices. For this reason, along with voice recognition technology, it is attracting attention as an important technology for realizing highly natural human-machine interfaces.

However, in order to obtain highly natural synthesized speech in such rule-based synthesis of speech, it is important to accurately reproduce the word-specific accent of each word.

Conventionally, the accent type unique to each simple word is registered in advance in an accent dictionary, and the accent information for each compound word consisting of a combination of multiple simple words is registered in advance in an accent dictionary. Highly natural synthesized speech is generated by referring to an accent dictionary to find accents specific to words (simple words, compound words) to be subjected to speech synthesis.

However, it is extremely difficult to correctly register word-specific accent information in an accent dictionary, and even experienced users often make mistakes. Conventionally, as disclosed in Japanese Patent Application Laid-Open No. 58-171098, for example, accent information is given to input words according to predetermined rules, speech synthesis is actually performed according to this accent information, and the synthesized speech is produced. It has been proposed to make judgments by hearing. At this time, it has also been proposed to change the accent information in response to a request for repeating synthesized speech, perform speech synthesis again, and repeat this process to obtain correct accent information.

However, there was a problem in that it required a lot of time to monitor all synthesized speech with different accents while issuing a repeat request, and to find the correct accent information and register it in the dictionary, resulting in poor processing efficiency. Moreover, since the synthesized speech is only output as a single voice each time a repeat request is given, there is a problem in that it is difficult to compare the different accents and it is difficult to judge the accuracy of the accents.

(Problem to be Solved by the Invention) In this way, in the past, in order to obtain the output of synthesized speech with different accents, monitor it, and find the correct accent, an instruction such as a repeat request is given each time. It was necessary. For this reason, there is a problem in that it takes time to accurately identify the correct accent for a word, and the work of registering it in an accent dictionary is extremely troublesome.

The present invention was made in consideration of these circumstances, and its purpose is to provide an accent that can easily and accurately specify the correct accent for a word and efficiently register the accent information in a dictionary. An object of the present invention is to provide a dictionary creation device.

[Structure of the Invention] (Means for Solving the Problems) When an accent dictionary creation device according to the present invention is instructed to register word information to be used for speech synthesis in an accent dictionary, the accent dictionary creation device stores words to be registered. The accent type given to the reading is automatically illuminated in sequence and outputted to the speech synthesis device together with the reading information, and the speech synthesis device sequentially synthesizes the speech and outputs the synthesized speech according to the monitoring result. The present invention is characterized in that an accent type to be registered in the accent dictionary is determined based on the pronunciation of a word.

In other words, when an instruction is given to register word information in an accent dictionary, the synthesized speech of the word to be registered should be automatically output in order with different accents, and the dictionary should be registered according to the monitoring results of this synthesized speech. This system is characterized by specifying accent information.

(Function) According to the present invention, by simply instructing the registration of word information in the accent dictionary by, for example, one key operation, the synthesized speech of the word to be registered is automatically output in order with different accents. Therefore, it is possible to easily and efficiently compare the four different accents and accurately identify the correct accent. Furthermore, it is possible to obtain correct accent information very efficiently and register it in a dictionary without having to issue a request to output synthesized speech each time as in the past.

(Example) Hereinafter, an example of the present invention will be described with reference to the drawings.

FIG. 1 is a schematic block diagram of an accent dictionary creation device according to an embodiment of the present invention. The accent dictionary 1 registers and stores accent information of words to be subjected to speech synthesis for each word, for example, in correspondence with character strings (headings) constituting the word. The speech synthesis unit 2, which forms the main body of the speech synthesis device, generates a character string (
When a character code string) is given, the character string is analyzed to obtain a phonetic parameter string and a prosodic parameter string, respectively, and the accent dictionary 1 is referred to to obtain accent information regarding the character string. Then, the speech synthesis unit 2 generates and outputs synthesized speech based on predetermined rules in accordance with this information (phonetic parameter string, prosodic parameter string, accent information). The synthesized speech generated by the speech synthesis section 2 is outputted via the speaker 3.

Now, in FIG. 1, a control section 4 constitutes the main part of the present apparatus, and includes the accent dictionary 1 described above, as well as an accent type table 5. The accent type table 5 is made up of a table in which information on the accent types that a word can have, depending on the number of moras of the word, is registered, as illustrated in FIG. 2, for example. This accent type information indicates at which mora position of the synthesized speech the accent is present. In other words, the accent type of a word is indicated by the mora position at which the accent is located, and the relationship between this accent type information and the number of moras is registered in the accent type table 5.

Therefore, the keyboard 6 inputs, for example, character strings constituting a word to be registered in the dictionary and information on their pronunciations to the control unit 4, and also performs control for registering accent information of the word in the accent dictionary l. This is for inputting information etc. The control unit 4 receives the synthetic speech output request instruction from the keyboard θ, and controls the generation and output of synthetic speech in which the accents of the words are different, as will be described later.

At this time, the control unit 4 displays information regarding the synthesized speech that is being synthesized and output on the display 7, for example as shown in FIG. 3, and presents it to the operator. Note that the information regarding the synthesized speech displayed on the display 7 is, for example, the headword of the word.

It consists of information such as pronunciation, part of speech, accent type, etc., and is displayed for a predetermined period of time including the period in which the synthesized speech is output.

Now, the control section 4 sequentially outputs synthesized speech with different accents and is controlled as follows.

To control the monitor output of this synthesized speech, first input the headword and reading from the keyboard 6, as shown in FIG.

It is activated by inputting part of speech information and determining this as the registration mode for accent dictionary 1 (step a).
. By inputting this information, the control unit 4 determines the number of moras of the word indicated by the input information.

After that, as a control parameter for generating a plurality of synthesized voices with different accents, initial setting of information
] is performed (step b).

In this state, it waits for a key input of the instruction information from the keyboard B, and then determines the type of the input information 1 (step C). This type is determined by determining whether monitor output of synthesized speech is requested or whether an accent type to be registered in the dictionary is directly specified. Specifically, according to the number of moras of the input word, the instruction information is determined by determining whether a number indicating the accent type that the word can have has been designated or whether any other key input has been performed. When an accent type is directly designated by a number, the designated number is registered and set in the accent dictionary 1 as accent type information (step d).

On the other hand, if there is a key other than a number that directly indicates the accent type, first refer to the accent type table 5 and find information on the Xth accent type y corresponding to the number of moras in the reading ( Step e). In addition, the above X
is a control parameter indicating the position of the mora described above.

Thereafter, in accordance with this accent type information, it is checked whether the y-th character in the pronunciation of the word is a long sound "-", a positive sound "n", or a consonant sound "tsu". If the y-th character is one of these long sounds, phlegmatic sounds, and consonant sounds, it is determined that the word cannot take that accent type, and speech synthesis using that accent type is not performed.

After confirming that the y-th character is not a long consonant, a cursive consonant, or a consonant consonant in this manner, information on this accent type is then displayed on the display 7 (step g). Information about the accent type y is displayed on the disk in the form shown in FIG. 3, for example. Thereafter, the information on the accent type y and the information on the reading are given to the speech synthesis section 2 to generate and output synthesized speech with the accent indicated by the accent type y (step h).

After adding the accent indicated by the accent type y to the synthesized speech as described above and outputting it to the monitor, the control parameter is confirmed, and the process from step e described above is repeatedly executed (step j). Through this repetitive control, information on different accent types is sequentially given to the given word information (pronunciation) as described above, and synthesized voices with different accents are sequentially output to the monitor.

If the control parameter X exceeds the number of moras of the input word, the process returns to the control parameter initialization process (step b) described above and waits for an accent type instruction to be input from the keyboard 6.

Incidentally, if accent type information is not input from the keyboard 6 at this time, the above-described monitor output of synthesized speech with sequentially different accents is repeated again.

According to the apparatus thus configured, the operator only has to instruct the output of the synthesized speech on the monitor, and the synthesized speech with different accents is sequentially output. At this time, since the accent type information of the synthesized voice being output on the monitor is displayed on the display 7 corresponding to the synthesized voice, the operator can judge whether the accent is correct or not while listening to the synthesized voice, and It becomes possible to read the accent type from the information displayed on the display 7.

As a result, by comparing synthesized speech with different accents that are continuously output on a monitor with just one key operation, we can distinguish accent type information by distinguishing between the different accents and obtaining synthesized speech with the correct accent. It becomes possible to read from the display 7. By inputting this accent type information manually, it is possible to accurately register the accent type information in the accent dictionary 1.

In particular, compared to conventional devices, there is no need to instruct the output of synthesized speech each time, so all accent forms that an input word can take can be effectively compared without any troublesome operations. It becomes possible to very accurately specify the accent of a word and register it in the accent dictionary 1.

By the way, as described above, when the accents of the synthesized speech are sequentially changed and outputted to the monitor, the correct accent may be found before all of them are outputted. In such a case, it is extremely inefficient to perform dictionary registration after outputting synthesized speech for all accents.

Therefore, if such a problem is to be taken into account, the processing procedure may be controlled as shown in FIG. 5, for example.

That is, after synthesizing speech according to accent type y and outputting the synthesized speech to a monitor, a silent section of a certain period of time is set (step k). Then, a routine for accepting interrupt processing from the keyboard 6 is activated during the period between the monitor output period of the synthesized voice and the above-mentioned silent period, and it is determined whether or not information specifying the accent type is input (step m). .

Then, when information specifying the accent type is input by this interrupt processing, registration settings in the accent dictionary 1 are performed according to the input information (step n
). On the other hand, when information other than the information specifying the accent type is input, or when no information is input from the keyboard 6, the process returns to the above-described processing procedure.

By incorporating such a processing procedure, for example, when the correct accent is confirmed in the initial stage of monitor output of synthesized speech with different accents, the subsequent output of synthesized speech is stopped and the output is changed to the accent dictionary 1. It becomes possible to perform dictionary registration, and it becomes possible to improve the efficiency of the processing procedure.

As described above, according to this device, synthesized speech with different accents for input words is sequentially output to the monitor, so differences in accents can be identified very easily and clearly by comparing the synthesized speech. becomes possible. As a result, compared to conventional devices that require the output of synthesized speech with different accents each time, it is possible to determine whether an accent is correct or incorrect very easily and efficiently, and the correct accent information can be effectively registered in the dictionary. becomes possible.

Note that the present invention is not limited to the embodiments described above. For example, the control flow for sequentially outputting synthesized speech with different accents is not particularly limited. Furthermore, the order in which accent types are assigned is not particularly limited. For example, accent types may be given in order of frequency based on the number of moras.

Furthermore, it is also possible to change the monitor output order according to part of speech information. Furthermore, although the explanation has been given here by taking new registration into the accent dictionary 1 as an example, it goes without saying that the present invention can be similarly applied when changing or modifying information that has already been registered in the accent dictionary 1. 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, the accent information for the word to be registered in the dictionary is automatically changed and sequentially added to the reading information, and synthesized speech with different accents is sequentially created. Since the synthesized speech is output, by sequentially comparing and monitoring the synthesized speech, it is possible to very easily identify the correct accent and register the accent information in a dictionary. In addition, great practical effects can be achieved, such as the ability to register accent information in the accent dictionary very simply, efficiently, and accurately without any troublesome operating procedures.

[Brief explanation of the drawing]

The figures show an accent dictionary creation device according to an embodiment of the present invention, in which FIG. 1 is a schematic block diagram of the embodiment device, FIG. 2 is a diagram showing an example of the configuration of an accent type table, and FIG. The figure shows a display example of accent information, and FIGS. 4 and 5 each show a characteristic processing control procedure of the present invention. ■...Accent dictionary, 2...Speech synthesis unit, 3...
- Speaker, 4... Control unit, 5... Accent type table, 6... Keyboard, 7... Display. Applicant's Representative Patent Attorney Takehiko Suzue Figure 1 Figure 2 Figure 3 Figure 4 Figure 5

Claims (1)

[Claims] When an instruction is given to register word information to be subjected to speech synthesis into an accent dictionary, the accent types given to the pronunciation of the word to be registered are sequentially varied and sequentially output to the speech synthesis device along with the pronunciation information. and means for determining an accent type to be registered in the accent dictionary for the pronunciation of the word according to the monitoring results of synthesized speech synthesized and sequentially output by the speech synthesis device. Accent dictionary creation device.