JPH067342B2 - Speech synthesis method - Google Patents

Description

TECHNICAL FIELD OF THE INVENTION The present invention relates to a speech synthesis system capable of generating synthetic speech with high naturalness.

[Technical background of the invention and its problems]

When converting a character string that constitutes a Japanese sentence into a voice, basically, the above character string is converted into a phoneme sequence, and from this phoneme sequence, parameters of a voice unit characterizing a spectrum envelope of the voice are obtained, and these voices are obtained. A speech synthesizer is driven according to a series of parameters generated by combining pieces to obtain synthesized speech.

In such speech synthesis, when a word contains a phoneme that causes phoneme deformation, processing without considering the phoneme deformation significantly deteriorates the naturalness of the synthesized speech. For example, the word "Science" is shown in the Kana series of "Kagak",
The second "ga" from the beginning of the kana sequence is a dull sound of a moth line, and is nasalized depending on the phonological situation before and after that. Therefore,
In order to enhance the naturalness of the synthesized voice, it is necessary to fully consider the phoneme deformation due to the nasalization.

Therefore, in the past, based on the principle that "ga-sound in a word (ga-daku) is nasalized", it is determined whether or not there is a g-sound in the phoneme sequence obtained from the input character string. When there is a line sound, it is converted into a nasal sound and voice synthesis is performed.

However, the conventional speech synthesis method has the following problems. That is, compound words frequently appear in Japanese sentences. Such compound words, for example "Bank of Japan"
When there are moth sounds and moth sounds in the same way, such as or "Kitatama-gun", when the moth sounds and moth sounds are nasalized or not nasalized depending on the nature of the compound word. There is.

Therefore, in the method of nasalizing the paradoxical sound by the gag sound like the conventional speech synthesis method, the naturalness of the synthesized speech is rather deteriorated, and the intelligibility of the synthesized speech is deteriorated. was there.

[Object of the Invention]

The present invention has been made in consideration of such circumstances, and an object thereof is to perform a nasalization test in a phonological sequence indicating a compound word that is particularly problematic with a high degree of accuracy for highly natural speech. It is to provide a speech synthesis method that enables synthesis.

[Outline of Invention]

The speech synthesis method according to the present invention includes a sentence input unit for inputting a Japanese sentence to be used for speech synthesis, a sentence segmentation unit that divides the Japanese sentence input from the sentence input unit into phrases, and words. Corresponding to, its reading, prosody information, a word dictionary storing nasalization information indicating whether or not to nasalize this word when the word is a compound word and contains a moth sound or a moth sound, and the clause A phoneme conversion unit that converts a word in the phrase into its reading by matching a phrase separated by a cutting unit with the word dictionary, and the word dictionary of the word dictionary for the word converted into the reading by this phoneme conversion unit. Using the nasalization information, a nasalization phoneme test unit that performs this nasalization test at a place containing moth sounds or dullness sounds, and the word for the word that has been nasalized test performed by this nasalization phoneme test unit. Using the prosodic information of the dictionary A prosody test unit for performing a temperament test, a voice parameter generation unit for generating a phonological sequence parameter and a prosody parameter for the word according to each test result by the nasalized phoneme test unit and the prosody test unit, and the voice parameter generation unit. And a synthesizer for synthesizing and outputting the word in accordance with the parameter generated by.

〔The invention's effect〕

Thus, according to the present invention, in particular, when a phoneme sequence indicating a compound word contains a ga-sound (ga-daku), whether the ga-sound is nasalized or not is determined from the property of the compound word. Since the verification can be performed with high accuracy, the speech synthesis processing of the word can be correctly performed according to the nasalization verification result.
Therefore, the naturalness of the synthesized voice can be sufficiently enhanced,
It is possible to achieve a practically great effect such as the improvement of the intelligibility.

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 embodiment apparatus.

A Japanese sentence to be used for speech synthesis is input through the sentence input unit 1, guided to the phrase segmentation unit 2 and segmented into segment units, and then provided to the phoneme conversion unit 3. The phonological conversion unit 3 is for phonologically converting the words of each of the above clauses in accordance with the word information registered in the word dictionary storage device 4 in advance in the dictionary. The word information registered in this word dictionary storage device 4 is
For example, as shown in FIG. 2, when the entry word 11, the grammatical information 12, the reading 13 of the word, the prosody information 14, the appearance frequency 15, and the word is a compound word and includes a gag sound or a gad sound. Nasalization information 1 indicating whether or not to nasalize this
It consists of 6 mag. The nasalization information 16, which will be described in detail later, includes, for example, information indicating the non-nasalization mora number of the word.

The phoneme sequence of the words of the phrase thus phonologically converted in this way is guided to the nasalized phoneme verification unit 5, and the nasalized information in the word information registered in the word dictionary storage device 4 is used. A nasalization test is performed on the nasalization candidates included in the word. This nasalization test extracts ga-sounds (ga, gi, g, g, ge, and go) and ga-ga-daku (ga, gyu, and gyo) in the phonological sequence as nasalization candidates, and these ga-sounds (ga It is performed by determining whether or not the muddy sound is nasalized in the word.

Further, the prosody test section 6 uses the word information in the word dictionary storage device 4 to test the prosody of the words of the phrase for which the nasalization test has been performed.

The voice parameter generation unit 7 reads out the voice units (parameters) registered in the voice unit file device 8 according to the test results relating to the words tested by the test units 5 and 6, and combines the read units. It generates phonological sequence parameters and prosodic parameters for words. According to these parameters, the synthesizer 9 speech-synthesizes the word and outputs it.

Thus, according to the speech synthesizer configured as described above, it is determined by referring to the word dictionary whether the nasalization candidate in the word is a nasalized phoneme or a non-nasalized phoneme, and the determination ( Since the phoneme parameter of the word is generated according to the result of the (test), it is possible to obtain a synthetic speech with high naturalness and high quality.

As described above, since the nasalization information 16 indicating whether or not to nasalize a word that is a compound word and includes a gag sound or a daku sound is registered in the word dictionary, there is a conventional problem. It is possible to accurately execute the nasalization process for the compound word that has become. That is, it is convenient to register a compound word, which is generally used with high frequency, in the dictionary in order to reduce the number of times the dictionary is collated and improve the processing efficiency. However, if the phoneme that comes to the boundary of the words that compose the compound word is a gaphonic sound, there are cases in which the gamonic sound is nasalized or not nasalized (nonnasalized) depending on the nature of the compound word. . For example, the proper noun "Bank of Japan" is pronounced as "Nihonginkou", its moth sound "Gi" is non-nasalized phoneme, and "Kitatama-gun" is pronounced as "Kitatamagun""Gu" is a nasalized phoneme. For this reason, if a compound word formed by combining a plurality of simple words is registered in the dictionary in the same manner as a general word, it becomes impossible to know whether or not the gag sound in the word becomes nasal.

Therefore, by giving the nasalization information to such a compound word as described above, the nasalization can be correctly tested. This means that the dictionary registration of the above-mentioned compound device can be performed in the same way as a simple word,
It enables speech synthesis with high processing efficiency without impairing the naturalness of synthesized speech.

By the way, the nasalization information may be given as follows. That is, the word to be voice-synthesized is decomposed into mora as its phoneme sequence. The position of the moth sound in the above word is known in advance by the mora number.
Therefore, since the moth sound in a word has the property of being nasalized in principle, the mora number of the non-nasalized phoneme is set so as to indicate the non-nasalized moth sound in the word. A dictionary may be registered as nasalization information. Specifically, the reading of the word "Japanese synthetic rubber" is given as "Nihon Gosei rubber." In this case, since the two moth sounds in the word are not nasalized, the non-nasalization mora number [4, 8] may be set as the nasalization information. This gives the 4th and 8th of the above words
Even if the second mora (phoneme) is a moth sound, it will not be treated as nasalization. Therefore, it is possible to effectively obtain a synthetic voice that is free from unnaturalness due to unintentional nasalization.

The present invention is not limited to the above embodiment. For example, the format of the nasalization information and the means for the verification processing of the nasalization phoneme may be determined according to the device specifications. In short, 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 a voice synthesizer configured by adopting an embodiment system of the present invention, and FIG. 2 is dictionary word information including nasalization information used in the embodiment. 1 ... Sentence input section, 2 ... Phrase segmentation section, 3 ... Phonological conversion section, 4
... word dictionary storage device, 5 ... nasalized phoneme verification unit, 6 ... prosody verification unit, 7 ... speech parameter generation unit, 8 ... speech unit file device, 9 ... synthesizer.

Claims (1)

[Claims] 1. A sentence input unit for inputting a Japanese sentence to be used for speech synthesis, a sentence segmentation unit for segmenting a Japanese sentence input from this sentence input unit into sentence units, and a corresponding word. A word dictionary storing the reading, prosody information, and nasalization information indicating whether or not to nasalize the sound when the word is a compound word and contains a gag sound or a gagaku sound, and the word segmentation section separates it. A phonological conversion unit that converts the word in the phrase into its reading by matching the obtained phrase with the word dictionary, and nasalization information of the word dictionary for the word converted into reading by this phoneme conversion unit. A nasalized phoneme verification unit that performs this nasalization test when it contains a moth sound or a syllabic sound, and prosody information of the word dictionary for the word that has been nasalized verification performed by this nasalization phoneme test unit. Test the prosody using A prosody test unit, a voice parameter generation unit that generates a phonological sequence parameter and a prosody parameter for the word according to each test result by the nasalized phoneme test unit and the prosody test unit, and a parameter generated by the voice parameter generation unit. And a synthesizer for synthesizing and outputting the word in accordance with the above.