JPS63168699A - Voice synthesizer - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] [Purpose of the Invention] <Industrial Application Fields> The present invention is directed to a speech synthesis method that can synthesize input words, especially proper nouns, given as strings of character codes with naturalness through simple processing. This relates to a synthesis device.

(Prior Art) Recently, prosodic parameters and phonological parameters of a word input as a string of character codes are determined, and speech is regularly synthesized according to these metrical parameters and phonological parameters. This type of speech synthesis, along with speech recognition processing, is an important technology for realizing highly natural man-machine interfaces.

However, in order to obtain synthesized speech with high naturalness and intelligibility in such rule-based synthesis of speech, it is important to accurately reproduce the accent specific to the word.

However, proper nouns such as surnames, company names, place names, etc. often have their own unique accents, and it is extremely difficult to correctly determine the accent unique to that proper noun from the series of input character codes that represent the word. be. Therefore, in conventional speech synthesis devices, for each simple word proper noun, the accent type unique to that simple word and the accent type of a compound word composed of a combination of multiple simple words are registered as an accent dictionary, and the input word is compared with this accent dictionary to find the accent specific to that proper noun.

(Problem to be solved by the invention) However, in conventional speech synthesis devices, when a compound word consisting of multiple simple words is input, even one of the multiple simple words is not registered in the accent dictionary. Even if other simple words can be matched, the matching fails, and the entire compound word is treated as one accent phrase and outputted with a flat accent without referring to the accent information in the accent dictionary. For example, when there is an input such as "Toshiba Koukyou Shiten", even if "Toshiba" and "r Tokyo" can be matched, the collation fails because "Shiten" is not found.

In this case, not only will the accent not be accurate, but r j
, to, -,, l'-shiba 1 tokyo shiten'' should be output, but it may be output as ``toushiba tokyo shiten'', resulting in a synthesized sound that lacks naturalness.

Another option is to use accent information unconditionally for simple words that can be matched, but in this case there is a high possibility that the position of the break between simple words will be incorrect.For example, inputting a "company name" such as "tsubai" If there is, "tsubai" and "pai" are not registered in the accent dictionary, and "tsu" is a "place name"
Assuming that is registered, r
There is a risk that ``tsu'' and ``pai'' will be cut into two accented phrases. In this way, if a word with 1 mora is registered in the accent dictionary, and there is an input string containing that word, the probability of matching with a word with 1 mora is
There was a problem in that the fibers were often separated at the wrong position.

The present invention was made in view of these problems, and its purpose is to efficiently utilize information such as reading information and accent type registered in an accent dictionary to perform accent testing with high accuracy. An object of the present invention is to provide a speech synthesis device that can effectively obtain highly natural synthesized speech.

[Structure of the Invention] (Means for Solving the Problems) In order to achieve the above object, the present invention includes an input unit for inputting words formed into character code strings, and storing accent information and pronunciation information of the words. an accent dictionary, a word matching unit that matches the input word input from the input unit with words stored in the accent dictionary and provides accent information and pronunciation information to the input word, and predetermined conditions are stored. If the input word is a compound word made up of multiple simple words, even if it is impossible to match the condition table with some of the simple words that make up the compound word, as long as the conditions stored in the condition table are met. A compound word accent testing unit instructs to use the determined accent information and reading information for the matched simple words constituting the compound word; and a speech synthesis unit that generates synthesized speech according to the determined prosodic parameters and phonetic parameters.

(Operation) When the input word is a compound word consisting of multiple simple words, even if some of the simple words that make up this compound word are impossible in the word matching section, the conditions stored in the condition table can be satisfied. For example, since the accent information and pronunciation information determined by the accent dictionary are used for matched simple words that make up this compound word, the frequency of use of accent information and pronunciation information increases, and the accuracy of accent testing increases. It will be done.

(Example) An example of the present invention will be described in detail below based on the drawings. FIG. 1 is a block diagram of a speech synthesis device according to an embodiment of the present invention. As shown in the figure, this speech synthesis device includes a character string input section 1, a word matching section 2, an accent dictionary 3, a phoneme series testing section 4, a compound word accent testing section 5, a condition table 6, a speech unit file 7, It consists of a phonetic parameter generation section 8, a prosodic parameter generation section 9, and a speech synthesis section 10.

Information on words to be used for speech synthesis is input from the character string input section 1 as a character code series. This input word information includes, for example, proper nouns such as first and last names, company names, and place names. The word matching unit 2 matches the input word with an accent dictionary 3 in which accent information and pronunciation information of proper nouns are registered in advance as shown in FIG. I'm looking for it from. The accent information and pronunciation information obtained here are given to the phonological sequence test section 4 and the prosodic parameter generation section 9 when the input word is determined to be one simple word, but when the input word is determined to be a complex word consisting of multiple simple words, When it is determined that the words constitute a word, the information is provided to the compound word accent testing section 5. Further, in the case of a compound word, matching information indicating whether or not each simple word has been matched is also provided to the compound word accent testing section 5. Compound word accent test part 5
In this method, when a matched input word constitutes a compound word consisting of a plurality of simple words, the accent in the compound word is tested according to the type of the simple word and the accent information detected by the match. Also, if there is a simple word that cannot be matched among multiple simple words, refer to condition table 6 and give accent information from the accent dictionary only to the simple word that was matched, or judge it as a matching failure. After checking and determining whether to give a flat accent by treating the entire compound word as one accent, give the input word or pronunciation information for the input word to the phonological sequence test section 4, and input it to the prosodic parameter generation section 9. Provides accent information for words.

In condition table 6, as shown in Figure 3, the number of matching words (the number of simple words that could be matched among the input words) is n, and the total number of matching moras (the total number of moras of simple words that could be matched) is m. When (n > 1) (n < m), the following conditions are registered. Here, "・" means "and". The prosodic parameter generation unit 9 determines the prosodic parameters of the character code series subjected to the accent test in this manner, and the prosodic parameter series is provided to the speech synthesis unit 10.

The phoneme sequence test section 8 tests whether the phoneme becomes nasal (hi) or devoiced based on the reading information obtained from the word matching section 2 or the compound word accent test section 5. to generate a phoneme sequence corresponding to the reading information and provide this to the phoneme parameter generation section 8.The phoneme parameter generation section 8 refers to the speech segment file 7 and creates a sequence of musical parameters corresponding to the phoneme sequence. The speech synthesis unit 10 synthesizes speech of the word indicated by the input character code series using a rule synthesis method in accordance with the series of phonetic parameters thus created and the series of prosodic parameters described above.

Next, the operation of this embodiment will be explained.0 character string input section 1
For example, if an input such as ``kabushikigaisha・'toushibasuzukikiyotaka'' is given, the word matching section 2 performs a comparison with the accent dictionary 3 having the contents shown in FIG. 2, for example. As a result of the matching, the three simple words "Kabushikigaisha", "1 Toshiba", and "Suzuki" match, but "Kiyotaka" does not match because it is not in the accent dictionary 3. Since the input word is a compound word, the contents of the accent dictionary for each simple word are obtained from the compound word accent test section 5. In the compound word accent test section 5, since there is a 7jl pure word in the input word that does not match the contents of the accent dictionary 3, it checks the condition table 6 as shown in FIG. In this example, "Kabushikigaisha" is 7 mora,
rl)'r7shi)<)'' is 4 moras, "Suzuki" is 3 moras, the number of matching words n is "3", and the total number of matching moras n
is 7+4+3=“14”, and condition (n
-1) (n<m), it is determined that partial matching is performed, and the information in the accent dictionary is used only for the above three simple words.

On the other hand, for "Kiyotaka", it is a name not found in the accent dictionary 3, and "Kiyotaka", which is the same as the input, is used as the pronunciation information, and a flat accent is added. The results obtained in this way are shown in FIG.

By providing these phoneme information to the phoneme sequence verification section 11 and the accent type to the prosodic parameter generation section 9, the speech synthesis section 10 finally outputs a synthesized speech sound.

In this way, in this embodiment, when the input word constitutes a compound word consisting of multiple simple words, even if the input word includes a simple word that fails to match with the accent dictionary 3, if the predetermined condition is met, For simple words successfully matched with the accent dictionary 3, accent information and pronunciation information can be easily and accurately determined, and can be verified with high precision. Therefore, based on this accent information, it is possible to express the accent unique to the input word and obtain a series of prosodic parameters, convert the reading information into a phonological series, and obtain phonological parameters based on that, making it possible to create an unnatural accent. In other words, it is possible to obtain clear and highly natural synthesized speech that reflects the word-specific accent.

Note that the present invention can be modified in various ways within the scope of its technical idea. For example, it goes without saying that the number of matching words and the number of matching moras are not limited to the above example.

[Effects of the Invention] As described in detail above, according to the present invention, accent dictionaries are used frequently and accent testing is performed with high accuracy, resulting in synthesized speech with high naturalness and clarity.

[Brief explanation of the drawing]

1 is a configuration block diagram of a speech synthesis device according to an embodiment of the present invention, FIG. 2 is a diagram showing the contents of the accent word f13, and FIG. 3 is a diagram showing the conditions stored in the condition table 6. FIG. 4 is a diagram showing the output contents of the compound word accent testing section 5. 1...Character string input section 2...Word matching section 3...Accent dictionary 4...Phonology Series test section 5... Compound word accent test section 6...
......Condition table 8...Phonological parameter generation section 9...
... Prosodic parameter generation unit 10 ...
・Speech Synthesis Department Applicant Toshiba Corporation Patent Attorney Sasa Suyama - (Character string) Figure 1

Claims (2)

[Claims] (1) An input unit for inputting words in character code strings, an accent dictionary for storing accent information and pronunciation information of words, and input words input from the input unit for words stored in the accent dictionary. a word matching unit that provides accent information and pronunciation information to the input word; a condition table that stores predetermined conditions; Even if it is impossible to match some simple words, if the conditions stored in the condition table are met, the matched simple words that make up the compound word will be matched with the determined accent information. a compound word accent testing unit that gives instructions using reading information; means for determining prosodic parameters and phonological parameters based on the accent information and reading information; and a voice generating synthesized speech according to the determined prosodic parameters and phonological parameters. A speech synthesis device comprising: a synthesis section; (2) If the number of simple words that can be matched by the word matching unit among the simple words that make up a compound word is n, and the total number of moras of the simple words that were matched is m, then (n>1) and (n<m). 2. The speech synthesis device according to claim 1, wherein the conditions to be satisfied are those stored in the condition table.