JPH03293398A - Device and method for reading out japanese language sentence - Google Patents

Abstract

PURPOSE:To read out the Japanese language sentence with satisfactory operation convenience and at a high speed without allowing it to pass through a host device by providing a phoneme data holding/sending-out part and sending out the phoneme data concerned to a feature parameter synthesizing part by designation of discrimination data. CONSTITUTION:When the command for showing the correspondence of a character data string and an identification number is sent from a host device 10, a word identification part 11 collates it with a word dictionary 11a, selects an optimal word train and outputs the phoneme information concerned. A rhythm imparting part 12 gives expiration/paragraph information, and information of an accent type, etc., thereto and generates phoneme data. It is written in an analytic text temporary accumulation buffer 15c together with an identification number. Thereafter, if only there is a command of its identification number utterance, the data is read out of the buffer 15c, sent out to a feature parameter synthesizing part 13, subjected to voice synthesizing analog conversion and the synthetic voice can be generated from a loudspeaker 14c.

Description

[Detailed description of the invention] [Table of contents] Overview Industrial field of application Prior art (Figures 5 and 6) Means for solving the problem to be solved by the invention (1st action (1st, 2nd) Figure) Embodiment (Figure 3.4) Effect of the invention Figure 2) (Summary) A word identification unit that analyzes a string of character data representing a Japanese sentence and outputs corresponding phoneme information, and the output phoneme. a prosody adding unit that adds prosodic information such as exhalation paragraph information based on the information; a feature parameter synthesizing unit that analyzes the provided prosodic information and phonetic data including the phonetic information and synthesizes corresponding feature parameters; Regarding a Japanese sentence reading device and method having a synthesized speech generator that converts synthesized feature parameters into synthesized speech and outputs it, the time spent on word identification etc. can be shortened, thereby eliminating the need for higher-level devices. The purpose of the present invention is to provide a high-speed Japanese text reading device and method that is easy to use and performs processing quickly without using a complicated interface. A phonological data holding/sending section is provided which holds corresponding phonological data together with identification data for identifying the string, and when the identification data is designated, extracts phonological information of the corresponding word string and sends it to the feature parameter synthesizing section. The configuration is as follows.

[Industrial application field]

The present invention relates to a Japanese text reading device and method, and more particularly to a word identification unit that analyzes a character data string representing an input Japanese text and outputs phonological information such as the corresponding phonology and accent type; a prosody assignment unit that assigns prosodic information such as exhalation paragraph information, phrase boundary information, accent boundary information, and accent type for each accent phrase based on the phonological information; and phonological data including the assigned prosodic information and the phonological information. The present invention relates to a Japanese sentence reading device and method having a feature parameter synthesis unit that analyzes and synthesizes corresponding feature parameters, and a synthesized speech generator that converts the synthesized feature parameters into synthesized speech and outputs the synthesized speech.

The Japanese reading device according to the present invention can be used in place of a display device used in a CPU, etc., or can be used in fields where the function as one of multimedia is applied, that is, CA
I (computer assisted
It is used as audio guidance in systems such as computer-based educational systems in which individual instruction is conducted through program learning while interacting with the computer.

[Conventional technology]

Conventionally, there has been a Japanese text reading device according to a first conventional example as shown in FIG.

As shown in the figure, this device extracts candidate words by comparing the character sequence data corresponding to the input Japanese text with a word dictionary, and selects the optimal word string from among these word combinations. , a word identification unit 41 that outputs phonological information such as phoneme, accent type, and grammar corresponding to the identified word; and based on the phonological information included in the identified word, exhalation paragraph information, phrase boundary information, A prosody adding unit 42 adds prosodic information such as accent boundary information and accent type for each accent phrase, and analyzes the provided prosodic information and phonetic data including the phonetic information, and extracts the corresponding vocal tract characteristic parameters from the syllable file. Feature parameter synthesis unit 43 that extracts and combines sound source parameters such as pitch and adds them to the obtained vocal tract characteristic parameters.
and a synthesized speech generator 44 that converts the synthesized feature parameters into synthesized speech.

On the other hand, the Japanese text reading device according to the second conventional example was installed in the sixth
As shown in the figure.

As shown in the figure, this device extracts candidate words by comparing the character sequence data corresponding to the input Japanese text with a word dictionary, and selects the optimal word string from among the combinations of these words. , the phonology corresponding to the identified word,
A word identification unit 51 outputs phonological information such as accent type and grammar, and extracts exhalation paragraph information, phrase boundary information, accent boundary information, and accent type for each accent phrase based on the phonological information included in the identified word. It has a prosody adding unit 52 that adds prosody information such as, etc., and the receiver receives phoneme data including the added prosody information and the phoneme information, and the receiver sends the taken phoneme data to the feature parameter synthesis unit 53. The phoneme data holding and sending unit is connected to the host device 50.
a feature parameter synthesis unit 53 provided within the phonological data, which analyzes the phoneme data, extracts and combines the corresponding vocal tract characteristic parameters from the syllable file, and calculates sound source parameters such as pitch and adds them to the vocal tract characteristic parameters; The synthesized speech generator 54 converts the synthesized feature parameters into synthetic speech.

This provides the following advantages over the first conventional example.

In other words, since the processing time for feature parameter synthesis and speech synthesis is considerably shorter than the processing time for word identification and prosody assignment, the overhead time from sending data from the host device 50 to starting speech synthesis is made sufficiently small. be able to. Furthermore, since the phonetic data including prosody information can be manipulated by the host device 50, more detailed information regarding occurrence can be added and provided.

[Problem to be solved by the invention]

By the way, although the second conventional example has advantages such as shorter processing time compared to the first conventional example, it also has the following problems.

In other words, the interface (data exchange) between the host device 50 and the Japanese reading device becomes very complicated, increasing the burden on the host device, and in fact, the problem is that the processing on the host device becomes more complicated. was.

Therefore, the present invention has been made with the object of providing a Japanese sentence reading device and method that can shorten processing time and add more detailed information regarding occurrences without increasing the burden on host devices. It is something that

(Means for Solving the Problems) In order to solve the above-mentioned technical problems, the first invention, as shown in FIG. and a word identification unit 1 that outputs phonological information such as accent type, and provides prosodic information such as exhalation paragraph information, phrase boundary information, accent boundary information, and accent type for each accent phrase based on the outputted phonological information. a prosody assigning unit 2 that analyzes the assigned prosody information and the phonetic data including the phonetic information, and a feature parameter synthesizing unit 3 that synthesizes corresponding feature parameters; In a Japanese text reading device having a synthesized speech generator 4 for outputting, for each predetermined word string, identification data for identifying the word string and corresponding phonological data are held, and when identification data is specified, , a phonological data holding/sending section 5 is provided which extracts phonological information of a corresponding word string and sends it to the feature parameter synthesizing section 3.

As shown in Figure 2, the second invention inputs a character data string representing a Japanese sentence (Sl), analyzes the input character data string, and outputs phoneme information such as the corresponding phoneme and accent type. (S2), based on the output phonetic information,
Adding prosodic information such as exhalation paragraph information, phrase boundary emotion group, accent boundary information, and accent type for each accent phrase (S3), and analyzing the attached prosodic emotion σ and phonological data including the phonological information, In a Japanese text reading method that synthesizes corresponding feature parameters (S6), converts the synthesized feature parameters into synthesized speech, and outputs the synthesized speech (S7), the word string is identified for each predetermined word string. The corresponding phoneme data is retained together with the identification data (S4
), and when identification data is specified (S5), the phonetic information of the corresponding word string is extracted, and then feature parameters are synthesized (S6).

(Operation) Next, the operation of the reading device and method according to the first and second inventions will be explained.

To read out a Japanese sentence, as shown in FIGS. 1 and 2, in step S1, a word data string representing the Japanese sentence to be read out corresponds to identification data for identifying a predetermined word string. Then, in step S2, the word identification unit 1 extracts candidate words by comparing the phonological information corresponding to the character data string representing the Japanese sentence with a word dictionary, and selects the selected optimal word. Outputs the phonological information corresponding to the string. The output phonetic information is sent to the prosody adding section 2 in correspondence with the identification data.

In step S3, the prosody imparting unit 2 uses information such as phoneme (reading), accent type, grammar, etc. contained in the identified word to apply exhalation paragraph information, phrase boundary information, etc. to the phoneme information. Information such as the accent type for each accent phrase is given.

Here, the term "exhalation paragraph" refers to a group of sentences uttered with an exhalation, such as a breather.

A "phrase boundary" is a gradual rise and fall in the pitch of each utterance segment corresponding to a phrase, etc., and refers to a boundary where the voice is reformed in intonation.

The term "accent phrase boundary" roughly refers to a boundary that divides each word into groups with a local rise or fall in pitch.

Each word to which prosody information has been added in this way is processed in step S4.
The data is temporarily held in the phoneme data holding and sending unit 5 together with the identification data.

When there is an instruction to read out the retained phonetic information specifying the identification data in step S5, the phonetic information is sent to the feature parameter synthesis section 3 in step S6, and feature parameter synthesis is performed. Become.

Here, "feature parameters" are parameters specified to express the characteristics of speech, and there are two types: sound source parameters that represent voice quality and vocal tract characteristic parameters that represent linguistic content. includes fundamental frequency (pitch), amplitude, etc., and vocal tract characteristic parameters include LPG coefficient, PARCOR coefficient, LSP coefficient, formant frequency, etc.

"Feature parameter synthesis" involves decoding input phoneme data, extracting vocal tract characteristic parameters of corresponding feature parameters from the corresponding (CV&V) syllable file, and combining them.

In addition, processing such as calculating and obtaining sound source parameters such as pitch and adding them to vocal tract parameters is performed.

The synthesized parameters are sent to the synthesized speech generator 4, and the Japanese sentence input in step S7 is read out using synthesized speech.

〔Example〕

Next, a Japanese text reading and method apparatus according to an embodiment of the present invention will be described.

FIG. 3 shows an overall equipment configuration diagram according to this embodiment.

As shown in the figure, this system includes a host device 10 that outputs reading instructions and data to be read out.
and a Japanese text reading device 20 according to this embodiment.

The host device 10 has a CPU 10a and an interface 10b.
, display unit 10c, file 10d, printer device 10
It is accompanied by input/output devices such as e.

In addition, as shown in the figure, the Japanese text reading device 20 is
It has a CPU 21 that performs word identification, a memory 22 that stores a dictionary, etc. used for word identification, and the synthesized speech generator 14.

In addition, as shown in the figure, the synthesized speech generator 14 includes a DSP 14a as a speech synthesis section that synthesizes speech based on characteristic parameters to be described later, and a speaker control section 14b.
and a speaker 14c.

FIG. 4 functionally shows the Japanese text reading device according to this embodiment, in which reading instructions and Japanese text to be read are input in correspondence with identification numbers as identification data. A host device 10, a command/data analysis processing unit 16 that decodes commands input from the host device 10 and outputs corresponding signals and data, and converts character data strings corresponding to input Kanji/Kanaji/Japanese text into words. A word that extracts candidate words by comparing them with the dictionary 11a, selects the optimal word string from among the combinations of these words, and outputs phonological information such as phoneme, accent type, and grammar corresponding to the identified word. an identification unit 11; and a prosody assignment unit 12 that assigns prosodic information such as exhalation paragraph information, phrase boundary information, accent boundary information, and accent type for each accent phrase based on the phonological information included in the identified word. , analyzes the given prosodic information and the phonological data including the phonological information, extracts and combines the corresponding vocal tract parameters from the syllable file, calculates and obtains sound source parameters such as pitch, and converts the vocal tract parameters into vocal tract parameters. For each predetermined word string, an identification number (
data) as well as corresponding phoneme data, and has a phoneme data holding and sending unit 15 that extracts the phoneme information of the corresponding word string and sends it to the feature parameter synthesis unit 13 when identification data is specified. It is.

Here, the command/data analysis processing section 16, the word identification section 11, the prosody adding section 12, and the feature parameter synthesis section 13 correspond to the CPU 21 and the memory 22. Further, the phoneme data holding/sending section 15 corresponds to the memory 22, etc., and as shown in FIG.
It has the following.

Furthermore, as shown in FIG.
43b.

Next, the operation of the Japanese text reading device according to this embodiment will be explained.

This embodiment differs from the first conventional example in that the following two commands are added in addition to the normal command for simply inputting a Japanese sentence and reading it aloud.

■One command is to input a Japanese sentence and an identification number to identify it from the host device, and analyze it (
This is a command to generate phonetic data including prosody information (word identification + prosody assignment) and store it in the buffer 15c together with the identification number.

■As another command, by giving the identification number from the host device, the analysis text-time accumulation buffer 15
This is a command to perform feature parameter synthesis by the feature parameter synthesis section 13 by extracting phonetic data including prosody information corresponding to it from above c and processing the data.

The host device 10 sends an identification number uln and a command indicating the correspondence between the identification number and a character data string representing the Japanese sentence "Do you want to perform voice processing?".

Then, the word identification unit 11 extracts candidate words by comparing the character data string representing this Japanese sentence with the word dictionary 11a, and selects the optimal word string from among the combinations of these candidate words. and outputs the corresponding phoneme information.

The output phonological information uses the phonology (reading), accent type, grammar, and information included in the word identified by the prosody adding section 12 to apply exhalation paragraph information, phrase boundary information, Information such as accent boundary information and accent types such as accent phrases is provided.

In this way, the phoneme data "Onsesho Rishima*Suka?" added to the phoneme information is generated.

The generated phoneme data "ONSATION RISIMA*SKA?" is written together with the identification number "1" into the analysis text-time storage buffer 15c by the writing section 15a.

Thereafter, the identification number “1” is sent to the phoneme data holding/sending unit 15.
When a command is sent to say the sentence ``pa'',
When the reading unit 15b receives the phonetic data stored corresponding to the identification number "1," the phonetic data "Onsation Risima*Ska?" containing the corresponding prosody information shown earlier is analyzed. It is read out from the text-time accumulation buffer 15c and sent to the feature parameter synthesis section 13.

The characteristic parameters are sent to the DSP as the speech synthesis section 14a of the synthesized speech generator 14, where they are synthesized into speech, and further converted into analog by the speaker control section 14b, so that synthesized speech is generated from the speaker 14c.

Note that when attempting to use voice in various applications such as database searches, it is common to synthesize a message that prompts the operator to perform an operation, for example, as shown below.

In that case, the host device needs to send the identification number and the corresponding Japanese text to this device before starting the application, but if the application wants to perform speech synthesis, any arbitrary By simply specifying an identification number, it is possible to instantly synthesize the corresponding voice.

In other words, in this embodiment, when speech synthesis is treated as a multimedia voice guidance function,
``Since the time-consuming process of word identification and prosody assignment is performed in advance, the time from sending a command from the host device to starting to synthesize speech can be shortened.■ Interaction with applications on the host device. It has two advantages: it does not require a complicated interface.

〔Effect of the invention〕

As described above, in the present invention, the phoneme data to which prosody has been added is once held in the phoneme data holding and sending section, and when the corresponding identification data is specified, the corresponding phoneme data is outputted and the characteristics of the phoneme data are output. It is input to the parameter synthesis section.

Therefore, the corresponding phoneme data can be sent simply by inputting the identification data, without bothering the host device as in the past.

Therefore, the time spent on word identification etc. is shortened, and the host device is not bothered by this. It's the end,
It is possible to provide a high-speed Japanese sentence reading device and method that is easy to use and performs processing quickly without using a complicated interface, and can also add more detailed information regarding the occurrence. become.

[Brief explanation of drawings]

Fig. 1 is a block diagram of the principle of the first invention, Fig. 2 is a flowchart of the principle of the second invention, Fig. 3 is an overall equipment configuration diagram of the embodiment, and Fig. 4 is a reading of Japanese text related to the embodiment. A block diagram showing the device, FIG. 5 is a block diagram related to the first conventional example,
and FIG. 6 is a diagram showing a Japanese text reading device according to a second conventional example. 1...Word identification unit 2...Prosody assignment unit 3...Feature parameter synthesis unit 4...Synthesized speech generator 5...Phonological data holding and sending unit Conventional art of applicant Fujitsu Ltd. Block diagram according to the example Fig. 5

Claims (2)

[Claims] (1) A word identification unit (1) that analyzes a character data string representing an input Japanese sentence and outputs phonological information such as the corresponding phoneme and accent type; A prosody assignment unit (2) that assigns prosodic information such as paragraph information, phrase boundary information, accent boundary information, and accent type for each accent phrase; and a prosody assignment unit (2) that assigns prosodic information such as paragraph information, phrase boundary information, accent boundary information, and accent type for each accent phrase; A Japanese sentence reading device includes a feature parameter synthesis unit (3) that synthesizes feature parameters to be processed, and a synthesized speech generator (4) that converts the synthesized feature parameters into synthesized speech and outputs the synthesized speech. For each word string, identification data for identifying the word string and corresponding phonological data are held, and when identification data is specified, the phonological information of the corresponding word string is extracted and sent to the feature parameter synthesis unit (3). A Japanese sentence reading device characterized in that it is provided with a phoneme data holding and transmitting section (5). (2) Input the character data string representing the Japanese sentence (S1)
, analyzes the input character data string, outputs corresponding phoneme information such as phoneme and accent type (S2), and based on the output phoneme information, exhalation paragraph information, phrase boundary information, accent boundary information, accent Prosodic information such as accent type for each phrase is added (S3), the added prosodic information and phonetic data including the phonetic information are analyzed, and corresponding feature parameters are synthesized (S6), and the synthesized feature parameters are Convert to synthesized voice and output synthesized voice (S7)
In the Japanese text reading method, for each predetermined word string, identification data for identifying the word string and corresponding phonological data are retained (S4).
However, when identification data is specified (S5), the phonological information of the corresponding word string is extracted, and then feature parameters are synthesized (S6).