JPH07152740A - Sentence reading device - Google Patents

Abstract

PURPOSE:To read out a sentence by reading intended by a user by previously registering information relating to the reading of a phrase optionally specified in text data to be read out and referring to the registered information at the time of language analysis. CONSTITUTION:A CPU 1 controls the whole document reader and synthesizes voice to be generated through a voice synthesizer 2. A user registering dictionary SRAM 7 rgisters a user registering dictionary including correspondence relation between a specific phrase newly registered by the use and its reading voice and the dictionary stores information indicating the reading, accent type and connection existence of each phrase to be registered. A work DRAM 8 temporarily stores various data on the way of processing of the CPU 1. A ROM 9 stores a program for driving the CPU 1, data 9-1, a language analytical processing dictionary 9-2, and a voice element file 9-3 for voice synthesis.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a text-to-speech device for outputting voice based on text data (e-mail, report, novel, newspaper article, etc.) created by a document creating device or the like.

[0002]

2. Description of the Related Art Recently, a text-to-speech device capable of synthesizing a voice signal based on text data of a text created by a text creating device (including a word processor, a text editor, etc.) and outputting a voice to read the text. Has been developed and used. This type of text-to-speech device has various uses because it can directly hear the content of any text.

The text-to-speech device is specifically an expansion board used in combination with various information processing devices such as a general personal computer (hereinafter abbreviated as PC) and a workstation (hereinafter abbreviated as WS). Type, general PC or WS, IC card type used in combination with a notebook type small electronic device (some have a built-in CPU and can be used alone), or consist of only PC or WS for reading aloud sentences It is realized by executing software.

Here, an outline of the operation of the conventional text-to-speech device will be described. First, a language analysis process is performed on text data of a sentence created by a document creation device or the like. In the linguistic analysis process, the text data is divided into words, and the information in the linguistic analysis dictionary for each word is referenced to generate reading information, accent information, and breathing parts (pauses), and phonetic symbols including these information. Organize in columns.

Next, based on the speech symbol string obtained by the language analysis processing, a speech unit file is referenced to generate speech synthesis parameters that are the basis of speech signal synthesis. The voice synthesizing function regularly synthesizes and outputs a voice signal according to a voice synthesizing parameter. The audio signal is amplified and uttered through an external earphone or through a speaker and converted into audio.

[0006]

As described above, in the conventional text-to-speech device, when converting from text data into a phonetic symbol string including information necessary for actual reading (presence or absence of reading, accent, breath), Information in a language analysis dictionary prepared in advance is referenced.

Therefore, when a word that is not stored in the dictionary exists in the text data, or if there is an uncommon word or phrase even in a combination of registered words, the voice reading expected by the user can be obtained. In some cases, it was not possible to read the text naturally, which hindered the understanding of the text content.

For example, when the word "SRAM" appears in a sentence, the expected reading "Esram" is "Esaaru-Emu" according to the alphabetic reading rules (the actual reading content and this description). Is slightly different).

[0009] For example, when a person named "Fumio Oyama" (reading "Fumio Ooyama") appears in the text, the reading "Daisen" is written for "Oyama" as the first candidate in the language analysis dictionary. ing. In addition, since "Fumio" is not registered in the language analysis dictionary, the reading "Shijo" is generated in accordance with the reading rules of the words not written in the dictionary.

As a result, the expected reading "Oyama Fumio" is read "Daisenjojo". For such reading aloud, the user is instantly distracted by not knowing what was read aloud, which is a great obstacle to understanding not only this phrase but the surrounding text.

The present invention has been made in consideration of the above circumstances, and it includes words that are not registered in the language analysis dictionary,
An object of the present invention is to provide a text-to-speech device capable of natural reading of a text including a phrase formed by an uncommon word combination.

[0012]

SUMMARY OF THE INVENTION The present invention relates to a text-to-speech device for generating voice-synthesized data by linguistic analysis of text data and uttering a voice, with respect to a specific word and phrase in the text data. Registration means for registering the correspondence with information about reading aloud,
It is characterized by further comprising a sentence reading unit for reading a sentence by a language analysis including the correspondence relationship registered by the registration unit.

[0013]

According to this structure, information about reading is registered for a word or phrase arbitrarily specified in the text data to be read aloud, and the user can refer to the information during language analysis. The sentence can be read aloud as intended.

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the hardware configuration of a text-to-speech device according to an embodiment of the present invention. Figure 1
As shown in FIG. 1, the text-to-speech device comprises a CPU 1, a voice synthesizer 2, a D / A converter 3, a low-pass filter 4, an amplifier 5, an earphone jack 6, a user registration dictionary SRAM.
7, a work DRAM, a ROM 9, an external I / F 10, a bus 11, and a battery 12. Further, the CPU 1, the voice synthesizer 2, and the D / A converter 3
Is configured as a one-chip LSI.

The CPU 1 controls the entire document reading device. In the case of a voice synthesis book, the CPU 1 performs a language analysis process on the input text data, generates a voice synthesis parameter, and outputs the voice synthesis parameter via the voice synthesizer 2. Control to make a voice.

Under the control of the CPU 1, the voice synthesizer 2 subjects the voice synthesis parameters to digital signal processing and converts them into digital vocal data. D
The / A converter 3 actually converts the digital voice data from the voice synthesizer 2 into an analog voice signal.

The low-pass filter 4 removes unnecessary high frequency components contained in the analog audio signal obtained by the D / A converter 3. The amplifier 5 voltage / power amplifies the analog audio signal from which the high-frequency component has been removed by the low-pass filter 4 and outputs it.

The earphone jack 6 is for outputting the signal voltage / power amplified by the amplifier 5 to the outside. An earphone is directly connected to the earphone jack 6 or a speaker with a built-in power amplifier is connected.

The user registration dictionary SRAM 7 is for registering a user registration dictionary having a correspondence relationship between a specific phrase (text data) newly registered by the user and a voice (reading information) read aloud. is there. The user registration dictionary SRAM 7 is backed up by the battery 12 and is configured to retain the contents of the user registration dictionary even when the main power supply is turned off.

The work DRAM 8 is for the CPU 1 to temporarily store various data during processing.
The ROM 9 is for storing a program and data 9-1 for operating the CPU 1, a language analysis processing dictionary 9-2, and a voice unit file 9-3 for voice synthesis.

The external I / F 10 is, for example, a JEID in the case where the text-to-speech device is constructed as an IC card type.
A (Japan Electronic Industry Development Associat
Ion) Ver.4 standard, IC card and PC / W
It is a standard interface for connecting to an information processing device such as S.

The bus 11 connects the above-mentioned units. The battery 12 is, for example, a button-type battery that supplies electric power to each unit constituting the device when the sentence reading device is used alone and backs up the contents of the user registration dictionary SRAM 7 when the device is powered off. .

FIG. 2 shows the contents of the user registration dictionary stored in the user registration dictionary SRAM 7. The user registration dictionary stores information indicating “reading”, accent type, and presence / absence of connection for each word to be registered.

The term to be registered is a character string that appears in the text data to be read out and for which "reading" is designated by the user. “Yomi” is a designated yomi for the word to be registered. Also, the accent type can be specified separately by dividing the reading into a plurality (delimited by / (slash) in the figure). The accent type indicates, for example, the number of beats (reading division corresponding to the voice) up to the reading of the designated word. If "reading" is separated, set for each part. With or without connection
Indicates whether the registered phrase is connected to the previous or subsequent word in the sentence. When words are connected, the accent position is moved according to a predetermined rule.

For example, the word "RAM" is read aloud as "al-emu" according to the contents of the language analysis dictionary prepared in advance, but if the user registration dictionary is registered as shown in FIG. , It is read aloud when the pitch is constant. In the case of the word "ROM", it is almost the same, but it is possible to connect with the previous word. For example, in the case of "mask ROM", the accent of "mask" moves and "maskuro" becomes constant. It is read aloud at the height and becomes low when "mu". The above contents are registered in the user registration dictionary.

FIG. 3 shows a functional block diagram of the text-to-speech device in this embodiment. As shown in FIG.
Language analysis processing means 20, synthesis parameter generation means 21,
It is composed of a voice synthesis unit 22, a language analysis dictionary 23, a user registration dictionary 24, a voice unit file 25, and a dictionary registration unit 26.

The means 20, 21, 22, 26 are mainly C
It is realized by the PU1 and the ROM 9 (a program and data 9-1 for operating the CPU 1 stored therein), and the voice synthesizing means 22 further includes the voice synthesizer 2 and the D / A.
It is realized by including the converter 3.

Language analysis dictionary 23 and speech unit file 2
5 is realized by recording its contents in the ROM 9, and the user registration dictionary 24 is realized by registering its contents in the user registration dictionary SRAM 7.

Next, the operation of this embodiment will be described.
Here, it is assumed that the text-to-speech device is configured as an IC card type and is inserted into an IC card slot conforming to JEIDA Ver.

First, an outline of an operation of reading a sentence based on text data will be described. First, a language analysis process is performed on text data of a sentence created by a document creation device or the like. The text data is input to the text reading device from the PC side via the external I / F 10 and stored in the work DRAM 8, for example.

The linguistic analysis processing means 20 divides the text data into words, and refers to the information in the linguistic analysis dictionary 23 and the user registration dictionary 24 regarding each word to read information, accent information, breathing section (pause). ) Is generated and summarized as a phonetic symbol string including these pieces of information. At this time, the user registration dictionary 24 is prioritized over the language analysis dictionary 23. Even if the user registration dictionary 24 is registered in the language analysis dictionary 23, if the target phrase is registered in the user registration dictionary 24, the user registration dictionary 2
Processing is performed according to the contents of 4.

Next, based on the speech symbol string obtained by the language analysis processing, the synthesis parameter generation means 21 refers to the speech unit file 25 to generate the speech synthesis parameter which is the basis of the speech signal synthesis. To do.

The voice synthesizing means 22 regularly synthesizes the voice signal according to the voice synthesizing parameter and outputs it. The audio signal is amplified via the low-pass filter 4 and the amplifier 5 and transferred to the earphone jack 6. Earphone jack 6
Is uttered through an earphone or a speaker and converted into voice.

Next, the operation (registration processing) for registering the dictionary contents as shown in FIG. 2 in the user registration dictionary 24 will be described with reference to the flowchart shown in FIG. As described above, when the text-to-speech device is reading the text, a screen as shown in FIG. 5 is displayed. That is, the content of the sentence to be read is displayed, and the cursor is displayed at the position currently being read (the display position of the cursor moves as the reading progresses). Also, "Reading aloud" and "Interrupting"
There are provided a message line for displaying a message notifying the execution status such as, and a function line indicating an input operation for executing the function.

In the registration process, when the user notices an unnatural reading while reading a sentence, and the reason is that the actual reading is different from the expected reading for a certain constituent phrase in the sentence. It is activated by pressing the "ESC" key of the keyboard provided on the PC.

When the "ESC" key is pressed to instruct the interruption of the text reading, the CPU 1 of the text reading device
Executes a series of processes necessary for reading interruption, and then interrupts the reading process (step S1). On the other hand, the PC side displays a screen as shown in FIG. On this screen,
The cursor is displayed at the character position of the break point.

When the "F2" key is pressed to select the registration mode for registering in the user registration dictionary,
The CPU 1 shifts to the registration module (step S
2). Further, the PC displays a screen (word registration screen) for inputting words to be registered in the user registration dictionary, as shown in FIG. The word registration screen has an area for displaying "registered word", "registered reading", "accent type", and "connection / non-connection". On the PC side, the operation content is displayed on the message line in response to a request from the text reading device. The user proceeds with word registration according to the message displayed on the message line.

When the CPU 1 shifts to the registration module, it outputs a registration phrase range designation request to the PC (step S3). In the PC, the range is designated by designating the start point and the end point of the word to be registered in the sentence read up to that point on the screen by using the word registration specifying cursor. If the registered word is a compound word consisting of multiple words, delimit by "/" (slash) and specify.

When the word registration range is designated, the CPU 1
Outputs a "reading" setting request for the registered word to the PC (step S4). On PC, the registered word "reading"
Is input from the keyboard and is displayed in the "registration reading" field.

When "reading" is input, the CPU 1
Outputs an accent type setting request for the registered word to the PC (step S5). On the PC, the accent type of the registered word is input from the keyboard (may be selected from the menu) and displayed in the "accent type" field. The accent type is a number indicating how many beats the tone of the voice is kept high with respect to the registered “reading”.
When a compound word is set in the registered word, "accent type" is specified for each word.

When the "accent type" is input, the CPU 1 finally outputs to the PC a connection presence / absence setting request before and after the registered phrase (step S6). The PC sets whether (presence or absence) the registered phrase is connected to the preceding and following words. When the connection is designated here, when the actual reading is performed, the language analysis processing means 20 treats the compound as a compound word and moves the accent when the predetermined condition is satisfied.

When all items are set in this way and the registration execution is instructed by pressing the "F2" key, the text-to-speech device reads out the registered phrase based on the set information ( Step S7). That is, the user confirms whether the registered contents (“reading registration” or “accent type”) are correct.

Here, the CPU 1 continues the registration mode /
Although an end selection request is output (step S8), if it is determined that the registered contents need to be reset, or if the end is not selected to continue to register another word,
The continuation of the registration mode is selected as it is (step S
8).

When the end of the registration mode is instructed,
The CPU 1 registers the registered content in the user registration dictionary SRAM 7 in a predetermined format, and ends the registration module (step S9) (step S9). Then, the reading is restarted from the interruption position or the designated position according to the sentence reading instruction (step S10).

In the above embodiment, the text-to-speech device is an IC card type equipped with the CPU 1.
Although it has been described as being realized by being mounted in the IC card slot provided in the PC, other hardware configurations may be adopted. That is, a general-purpose PC, which is an expansion board type used in combination with various information processing devices such as PC and WS.
, WS, notebook type small electronic device used in combination with I
C card type (no built-in CPU) or P
It may be configured only by C or WS and realized by executing software for reading a sentence.

Further, the user registration dictionary 24 is stored in the IC card as a battery backup SR for the user registration dictionary.
Although it has been described as being stored in the AM7, it is stored in the flash memory mounted on the expansion board or the IC card or in the SRAM with battery backup in the PC or WS, and is turned off and then reused. Also,
The registered content may be referred to without the need for re-registration.

In this way, the user can arbitrarily set how to read a specific phrase in a sentence, and the specific phrase ("registered phrase") and its "reading" can be set.
By storing “accent type” and “presence / absence of connection” as a user registration dictionary in association with each other, it is possible to naturalize words that are not registered in the language analysis dictionary or sentences that include phrases that are not common word combinations. It is possible to read aloud.

By storing the user registration dictionary in the non-volatile storage device (SRAM 7 for user registration dictionary), the contents are not lost even when the main power supply of the device is turned off. It can be used effectively without the need.

[0049]

As described above, according to the present invention, it is possible to store information relating to reading in association with an arbitrary phrase in a sentence to be read out, and use it when reading out a sentence. This allows natural reading of words that are not registered in the linguistic analysis dictionary or sentences that include phrases that are combinations of unusual words.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of a text-to-speech device according to an embodiment of the present invention.

FIG. 2 is a SRAM 7 for user registration dictionary in the present embodiment.
Showing the contents of the user registration dictionary stored in.

FIG. 3 is a functional block diagram showing the configuration of a text-to-speech device according to the present embodiment.

FIG. 4 is a flowchart showing a registration procedure in a user registration dictionary in this embodiment.

FIG. 5 is a diagram showing an example of a screen displayed when reading a sentence according to the present embodiment.

FIG. 6 is a diagram showing an example of a screen displayed when text reading is interrupted in the present embodiment.

FIG. 7 is a diagram showing an example of a screen displayed when registering in the user registration dictionary according to the present embodiment.

[Explanation of symbols]

1 ... CPU, 2 ... voice synthesizer, 3 ... D / A converter,
4 ... Low-pass filter, 5 ... Amplifier, 6 ... Earphone jack, 7 ... User registration dictionary SRAM, 8 ... Work D
RAM, 9 ... ROM, 10 ... External I / F, 11 ... Bus,
12 ... Battery, 20 ... Language analysis writing means, 21 ... Synthesis parameter generation means, 22 ... Speech synthesis means, 23 ... Language analysis dictionary, 24 ... User registration dictionary, 25 ... Speech element file, 26 ... Dictionary registration means.

Claims (4)

[Claims] 1. A text-to-speech device for generating speech synthesis data by linguistic analysis of text data and uttering a voice, wherein a correspondence between a specific phrase in the text data and information regarding reading of the specific phrase is defined. A text-to-speech device comprising: a registration means for registering; and a text-to-speech means for reading a sentence by language analysis including the correspondence relationship registered by the registration means. 2. The text-to-speech device according to claim 1, wherein the content registered by the registration means is stored in a non-volatile storage means. 3. The text-to-speech device according to claim 1, which is configured as an IC card type. 4. The reading information registered by the registration means includes at least one of reading of the specific phrase, accent type, and information about presence / absence of connection with preceding and following words. The sentence reading device according to item 1.