JPH0650435B2 - Speech synthesizer - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a speech synthesizing apparatus capable of converting a word formed by an input character string into a synthetic speech with high intelligibility by a simple process.

[Technical background of the invention and its problems]

In the rule synthesizing process of a voice in which a word formed by an input character string is converted and output as a synthetic voice, it is an important task to increase the intelligibility of the synthesized voice.

Therefore, words are roughly classified into simple words and compound words based on their structure. However, in the conventional device, the speech synthesis is performed based on the simple word exclusively, and the compound word is treated as one word to perform the speech synthesis processing.
Therefore, when a compound word is given, there is a problem in terms of the intelligibility of the synthesized output voice.

That is, the reading and accent of a compound word often change depending on the structure of the compound word and the types of prefixes and suffixes.
For example, the two compound words "new literature" and "newspaper studies" have the same reading. However, the compound word structure has a relationship of "prefix + independent word" as "new + literature" and a relationship of "independent word + suffix" as "newspaper + science".

In spite of this, in the conventional device that integrates the independent word and the prefix (suffix) word and synthesizes the speech as one independent word, both the above-mentioned words (composite words) are "6 mora type 3". You will get the accent test result of "Accent". For this reason, the synthesized voices both become "shinbungaku" and cannot be distinguished.

[Object of the Invention]

The present invention has been made in consideration of such circumstances, and an object of the present invention is to provide a speech synthesizing apparatus capable of obtaining a synthetic speech with high intelligibility for Japanese words, particularly compound words. Especially.

[Outline of Invention]

The present invention collates a word character string with an independent word dictionary, and generates a prosody parameter and a phoneme parameter according to accent type and reading information registered in the independent word dictionary for a word collated as a simple word, and the independent word For character strings that could not be matched as simple words in the word dictionary, in addition to the independent word dictionary, the prefix dictionary and suffix dictionary are respectively matched, and separation from the subsequent words registered in the prefix dictionary or By using the information indicating the connection, the accent pattern for the compound word formed by the character string is tested to generate its prosody parameter and phonological parameter.

〔The invention's effect〕

Thus, according to the present invention, it is possible to highly accurately and easily test a change in the prosody of a compound word formed by combining a prefix or a suffix having a high degree of occurrence in Japanese as an independent word. Become. As a result, it is possible to accurately obtain the prosodic parameter and the phonological parameter for the compound word to effectively create a synthetic speech with a high degree of intelligibility. Therefore, even for compound words having the same reading, it is possible to clearly distinguish them from the difference in their prosody.

Example of Invention

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a schematic configuration diagram of the apparatus of the embodiment. Character input part 1
Is for inputting a character string of a word that is commonly used for speech synthesis, and the input character string is divided into words (simple words or compound words) and provided for speech synthesis processing.

The simple word dictionary matching unit 2 matches the word (input character string) input via the input unit 1 with the independent word dictionary 3 to search for an independent word corresponding to the input character string.

The input character string obtained by the collation result in the independent word dictionary 3 is verified as a simple word, and the reading information about the independent word and the accent information registered in the independent word dictionary 3 are read out. Then, according to these pieces of information, the simple word prosody parameter generation unit 4 generates prosody parameters for the simple words formed by the input character strings, and the simple word phoneme parameter generation unit 5 generates phonological parameters for the simple words. These pieces of parameter information are sent to the synthesizer 11 to be described later and used for the voice synthesis processing.

On the other hand, an input character string that is not tested as a simple word by the simple word matching unit 2 matching with the independent word dictionary 3 is
It is verified as a compound word and given to the compound word collation unit 6. This compound word dictionary collation unit 6 collates the input character string using a prefix dictionary 7 and a suffix dictionary 8 in addition to the independent word dictionary 3.

That is, the compound word dictionary collating unit 6 first collates, for example, several characters sequentially from the beginning of the input character string with the independent word dictionary 3. Then, the character string part that could be collated with the independent word dictionary 3 is removed, and the remaining character string is collated with the independent word dictionary 3 in the same manner. By repeating this process, a compound word consisting of a combination of independent words is tested.

If the remaining character string fails to be collated with the independent word dictionary 3, the character string is collated using the suffix dictionary 8. By this matching process, a compound word that is a combination of an independent word and a suffix is tested. When the verification by the independent word dictionary 3 fails, the collation using the prefix dictionary 7 is performed from the leading character portion. Then, the collated prefix is removed and collation is again performed with the independent word dictionary 3, and a combination of the prefix and the independent word is detected by this collation processing. Note that these processes are performed in combination with each other, and a compound word in which a prefix, an independent word, and a suffix are combined is also tested. In addition, it is determined that the input character string that has failed the above test does not form a word.

With respect to the compound word tested in this way, the compound word accent test unit 9 obtains the structural information of the character strings forming the compound word as the matching result of the compound word dictionary matching unit 6, and the dictionary 3,
The accent information required from 7 and 8 is entered.

When the input character string is tested as [prefix + independent word], the compound word accent verification unit 9 inputs information about the prefix registered in the prefix dictionary 7, for example, as shown in FIG. To do. In this prefix dictionary 3, the "reading", "accent type", and "separation / join information" are registered with the prefix as a heading. According to such information, for example, the input character string (compound word) for which [separation] is specified is accented and separated into its prefix and independent word. In addition, the input character string (compound word) for which [Combination] is specified is not accented. The compound word accent test unit 9 generates prosody parameters for the prefix and the independent word according to the accent pattern for the compound word thus tested.

When the input character string is tested as [independent word + suffix], the compound word accent verification unit 9 accentuates the prefix with the preceding independent word. By this combining process, the prosody parameters for the input character strings forming the compound word are generated as having one connection.

On the other hand, the compound word phonological parameter test unit 10 obtains information about "reading" registered in each of the dictionaries 3, 7, and 8 for the independent word, the prefix, and the suffix matched as described above, A phonological parameter for the compound word is generated. Such a phoneme parameter and the prosody parameter are given to the synthesizer 11 so that the compound word is speech-synthesized.

Thus, according to this device, for example, when a compound word "good boy" is input, it is verified that it is a combination of the prefix "good" and the independent word "boy", and the information about the accent is [separated]. Since the instruction is given, the reading "Koudanshi" is separated into "Kou (good)" and "Danshi (boy)", and they are processed as different words in the speech synthesis. In this case, however, if the two words are completely independently speech-synthesized, an unnatural voice is generated as a compound word. Therefore, for example, in the duration of the fundamental frequency of the voice expressing the accent, a process that does not cause a sudden change in the connected portion. Should be done. As a result, it becomes possible to obtain the time-varying pattern of the logarithmic fundamental frequency of the output voice as shown in FIG.

On the other hand, when the compound word "Kodanshi" is input, it is verified that the structure is the independent word "Kodan" and the suffix "shi". Therefore, as mentioned above, the "suffix"
Is accent-integrated with "independent word", the reading "Koudanshi" is expressed by one continuous accent pattern and synthesized into speech. As a result, for the compound word "Kodanshi", it is possible to obtain a time-varying pattern of the logarithmic fundamental frequency in the output voice as shown in FIG. 3 (b), for example.

Therefore, even for the compound words “good boy” and “Kodanshi” which have the same “reading”, these are output as synthesized speech of different accent patterns. Therefore, it is possible to perform speech synthesis with high intelligibility that enables the identification.

The present invention is not limited to the above embodiment. For example, the test process for compound words is performed first in the prefix dictionary.
You may start by checking with 7. Further, the synthesizer 11 may automatically perform the processing for the combined portion of the compound words instructed to be separated in accent. In addition, the present invention can be variously modified and implemented without departing from the scope of the invention.

[Brief description of drawings]

FIG. 1 is a schematic configuration diagram of an embodiment device of the present invention, FIG. 2 is a diagram showing a configuration example of a prefix dictionary in the embodiment device, and FIG.
(A) and (b) are diagrams showing accent patterns of a compound word subjected to accent test processing. 1 ... Character string input unit, 2 ... Simple word dictionary matching unit, 3 ...
Independent word dictionary, 4 ... Simple word prosody parameter generator, 5 ...
... Simple word phoneme parameter generation unit, 6 ... Compound word dictionary matching unit, 7 ... Prefix dictionary, 8 ... Suffix dictionary, 9 ... Compound word accent test unit, 10 ... Compound word phoneme parameter test unit, 11 ... Voice synthesizer.

Claims (1)

[Claims] 1. A simple word dictionary matching unit for matching a word character string with an independent word dictionary, and a prosody parameter of a word matched as a simple word by the simple word dictionary matching unit according to an accent type registered in the independent word dictionary. A simple word prosody parameter generating unit that generates, a simple word phoneme parameter generating unit that generates a phoneme parameter according to the reading information of the independent word registered in the independent word dictionary, and a simple dictionary matching unit that can be matched as a simple word. Using the information indicating the separation or combination of the compound word collation unit that collates the character string that did not exist with the prefix dictionary and the suffix dictionary in addition to the independent word dictionary, and the subsequent words registered in the prefix dictionary. Shows the separation or combination of the compound word accent test means for testing the accent pattern for the compound word and generating the prosody parameter thereof and the following word. Speech synthesis apparatus characterized by comprising a compound word phoneme parameter generating unit for generating the phoneme parameter using the broadcast.