JPH01321495A - Speech synthesizer - Google Patents

Abstract

PURPOSE:To make a synthetic speech easy to listen to by easy processing by inserting a specific voiceless section when a word representing a surname or given name appears in an input word string and thus generating the synthetic speech. CONSTITUTION:A character string code which is inputted through a character string input part 1 is supplied to a word matching part 2 and word test processing using an accent dictionary 3 is carried out; and information on the reading, part of speech, and accent type etc., corresponding to each word is supplied to a phoneme and rhythm information generation part 4 and used to generate a phoneme string and a rhythm string. Then when a proper noun showing 'surname' appears except at the word head of the input character string code, the specific voiceless section is interposed between this word and last word and when a proper noun representing a 'given name' appears except at the word tail of the input character string code, the specific voiceless section is interposed between the word and following word, thereby generating synthetic voice. Consequently, even if a simple word consists of word information on a complicate customer's name, etc., the synthetic speech is easily obtained which has high naturalness and is easy to hear.

Description

[Detailed Description of the Invention] [Object of the Invention] (Industrial Application Field) The present invention provides a speech synthesis device that can synthesize speech with good naturalness through simple processing of word information given as a symbol string code. Regarding.

(Prior art) Recently, according to the phonological parameter series and the prosodic parameter series obtained by analyzing character string codes indicating word information,
Various speech synthesis devices that generate synthesized speech based on predetermined rules have been developed.

This type of speech synthesis device using a rule synthesis method has an advantage over a conventional speech synthesis device using a recording/editing method in that it can easily generate synthesized speech representing arbitrary words or sentences. For this reason, along with technological developments in speech recognition processing, it is attracting attention as an important technology for realizing highly natural man-machine interfaces.

By the way, in this type of speech synthesis using rule synthesis,
In order to obtain synthesized speech that is highly natural and easy to hear,
An important challenge is to accurately reproduce the accent unique to each word.

Therefore, conventionally, accent information specific to each simple word or a compound word formed by a combination of simple words is registered in an accent dictionary in advance, and when word information consisting of a character string code is given, , the accent information for the input word is obtained by comparing it with the accent dictionary, and synthesized speech with a word-specific accent is obtained according to this accent information. By checking with such accent dictionaries, it has become possible to relatively easily obtain highly natural synthesized speech.

By the way, this type of speech synthesis device has recently been applied to, for example, automatic transfer machines at banks to confirm payment destinations. Specifically, the automatic transfer device outputs the transfer destination information entered into the device using a synthesized voice to prompt confirmation, and once confirmation is obtained, the transfer process is automatically executed. There is. However, the customer name indicating the remittance destination may be simply given as a combination of "surname" + "first name", but - in the field, it is given as "company name" + "head office/branch name".
It is often given as a complex combination such as + "surname" + "first name". In such a case, the accent information for each individual word is determined by testing using the above-mentioned accent dictionary, and the synthesized speech is generated according to the phonological parameter series and the prosodic parameter series.
Since each word is synthesized continuously, it cannot be denied that the synthesized speech is somewhat difficult to hear. In particular, since the "head office/branch name" and the "surname", which have different meanings, are synthesized in a continuous manner, there is a problem in that the synthesized speech of the "surname" tends to be difficult to hear.

(Problem to be Solved by the Invention) In this way, conventional speech synthesis speech devices test the accent position of each word using an accent dictionary to obtain highly natural synthesized speech. However, when synthesizing complex word information consisting of simple words, such as a sequence of proper nouns such as a customer's name, the synthesized speech becomes difficult to understand due to the unnatural connection (continuity) of the synthesized speech of each word. The problem was that it was becoming difficult.

The present invention was developed in consideration of these circumstances, and its purpose is to create synthetic speech that is highly natural and easy to hear, even when word information such as customer names has a complex structure of simple words. An object of the present invention is to provide a speech synthesis device that can be easily obtained.

[Objective of the Invention (Means for Solving the Problem) The present invention analyzes a character string code indicating a word to obtain accent information of the word, and also obtains a phonological parameter series and a prosodic parameter series thereof, In a speech synthesis device that generates synthesized speech according to these phonetic parameter series, prosodic parameter series, and the accent information, when a proper noun indicating "surname" appears at a position other than the beginning of the input character string code, this word A predetermined silent interval is inserted between this word and the word immediately before it, or when a proper noun indicating a "name" appears in a position other than the end of the input character string code, between this word and the word immediately after it. This method is characterized in that a predetermined silent period is inserted into the synthesized speech to generate synthesized speech.

(Operation) According to the present invention, when the occurrence of a proper noun consisting of "surname" or "name" in the input character string code is detected by verification using an accent dictionary, the word immediately before it or the word immediately after it is detected. Synthesized speech is generated with a predetermined silent interval provided between the words. As a result, even when word information is given that is a combination of simple words in a complex structure, synthesized speech is generated with silent intervals (pauses) between words that have different meanings. This eliminates the occurrence of unnatural connections between words, making it possible to generate highly natural synthesized speech and improve its audibility. Moreover, there are silent intervals (pauses) in places where words with different meanings are connected, such as immediately before "surname" or immediately after "name".
It becomes possible to effectively generate highly natural synthesized speech by the simple process of providing the above.

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

FIG. 1 is a schematic diagram of an apparatus according to an embodiment of the present invention.
is a character string input section for inputting word information encoded in character strings. The character string code input through the character string input section 1 is given to the word matching section 2, and the accent dictionary 3
It is applied to word test processing using . This accent dictionary 3 includes, for example, the reading 1 of each word information as shown in FIG.
Information such as part of speech (type), accent type, etc. is registered in advance in association with the heading. The word matching unit 2 searches the accent dictionary 3 for the character string code inputted from the input character string (2), obtains information as a word, and performs verification thereof.

Reading 1 corresponding to each word found by this word matching unit 2
Information such as part of speech and accent type is sent to the phonological/prosody information generating section 4
and is used to generate its phonological and prosodic sequences. At this time, the phoneme/prosodic information generation unit 4 activates the silent interval insertion unit 8, which will be described later, according to the word's part-of-speech information, and inserts a predetermined silence between a predetermined type of word and the word immediately before or after it. Information indicating the insertion of a section (pause) is added.

Therefore, the synthesis parameter generation section 5 refers to the speech segment file 6 according to the phoneme sequence for the input character string code given from the phoneme/prosodic information generation section 4, and extracts speech segment parameters corresponding to each phoneme. This method generates a phonological parameter sequence. At the same time, the synthesis parameter generation section 5 generates a prosodic parameter series according to the above-mentioned prosodic information, accent information for each word, and information indicating a silent interval.

The speech synthesis unit 7 constructs a filter that approximates the vocal tract characteristics based on predetermined rules according to the phonetic parameter series and the prosodic parameter series generated in this way, and generates and outputs synthesized speech by passing a sound source through this filter. ing.

In this device basically configured as described above, to explain in detail the insertion of the silent section described above, first, the word matching section 2 performs matching with the accent dictionary 3 for each word information indicated by the input character string code, Along with information on the pronunciation, information on the accent type and the type of part of speech is required. The silent interval insertion unit 8 in the phonetic/prosodic information generation unit 4 examines the part of speech information obtained for each word in the word matching unit 2, and determines whether the type of the part of speech of the word is the “surname” of the first name in the proper noun. Yes, but we are testing whether it is a ``name'' or a surname. When a word of the type "surname" is detected, after confirming that the word is not the beginning of the input string code, a predetermined line is inserted between the word of the type "surname" and the word immediately before it. This is done by inserting silent intervals. In addition, when a word of the type "name" is detected, after confirming that the word is not at the end of the input character string code, add a predetermined value between the word of the type "name" and the word immediately after it. This is done by inserting silent intervals.

Specifically, for example, if a character string code such as "XXX・For each word, the type of pronunciation, part of speech, accent type, compound accent information, etc. are obtained.

Then, the phoneme/prosodic information generating section 4 performs a predetermined compound word test, accent test, etc., and obtains the phoneme series and prosody series. By detecting the proper noun "Sasaki" (surname), information indicating a silent interval is added between it and the previous word.

As a result, the synthesis parameter generation unit 5 generates the phonetic parameter series, and also generates the prosodic parameter series according to the accent information and silent section information, for example, in the third
Generate as shown in the figure.

Then, the speech synthesis unit 7 generates the synthesized speech as "XXX×Yanagi Chokojo/Sasaki Yutaka" according to the phonetic parameter series and the prosodic parameter series. Note that [] is an accent nucleus, and [・] is an accent phrase delimiter.

[/ indicates a silent section of 200n+sec.

Similarly, if the character string ``Saito Takashi Shuunyuyaku'' is given, a silent interval is inserted between the Ebetsu word ``Takashi'', which is the ``name'', and the word immediately after it, and ``Saito The synthesized voice is generated as "Ai Takashi Shuunyuyaku".

In this way, according to this device, when inputting a character string consisting of a combination of word information of proper nouns, and outputting the word information indicated by this input character string through speech synthesis, When the type of part of speech is "surname" of the first and last name, a predetermined silent interval is inserted between the word immediately before it and the synthesized speech is generated. Also, the ``name'' of the surname
When a word indicating a word is detected, a predetermined silent period is inserted between the word immediately after it and a synthesized speech is generated. As a result, when a word indicating a first and last name appears in synthesized speech, a silent interval (pause) is inserted between the words before and after it to clearly indicate the break between the words. Therefore, it becomes possible to easily eliminate the unnaturalness of the conventional method. This makes it possible to obtain highly natural synthesized speech with clear word boundaries, and to greatly improve its audibility.

Note that the present invention is not limited to the embodiments described above. For example, it goes without saying that the content of the word information registered in the accent dictionary 3 and the information format given to the synthesis parameter section 5 may be determined according to the specifications required of the apparatus. Furthermore, the length of the silent section inserted before and after the "surname" and "first name" is not particularly limited, and can of course be varied depending on the word content (length, etc.). 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 speech synthesis device according to the present invention, when a word indicating "surname" or "name" appears in an input word series, it can be distinguished from the word immediately before it or the word immediately after it. Since synthesized speech is generated by inserting a predetermined silent interval between the words, it is possible to clearly separate words indicating proper nouns, and it is possible to effectively eliminate unnaturalness in synthesized speech. .

As a result, great practical effects can be achieved, such as making it possible to improve the audibility of synthesized speech through simple processing.

[Brief explanation of the drawing]

The figures show a speech synthesis device according to an embodiment of the present invention. FIG. 1 is a schematic configuration diagram of the embodiment device, FIG. 2 is a diagram showing an example of the configuration of an accent dictionary, and FIG. 3 is a synthesis parameter section. FIG. 3 is a diagram showing a format string of data to be given. 1...Character string input section, 2...Word matching section, 3...
Accent dictionary section, 4...Phonological/prosodic information generation section, 5
. . . Synthesis parameter generation unit, 6 . Speech segment file, 7. Speech synthesis unit, . . . Silent interval insertion unit. Applicant's agent Patent attorney Takehiko Suzue

Claims (1)

[Claims] A character string code indicating a word is input, and this input character string code is analyzed to obtain accent information of the word, and its phonological parameter series and prosodic parameter series are respectively obtained. In a speech synthesizer that generates synthesized speech according to a parameter series, a prosodic parameter series, and the accent information, when a proper noun indicating "surname" appears at a position other than the beginning of the input character string code, this word and the immediately preceding When a proper noun indicating "name" appears, a predetermined silent interval is inserted between this word and the word immediately following it to generate synthesized speech. A speech synthesis device characterized by: