JPH064090A - Method and device for text speech conversion - Google Patents

Abstract

PURPOSE:To clearly represent the intention represented in a text by extracting an emphasis part in the text, and emphasizing and outputting the speech of the extracted emphasis part. CONSTITUTION:An emphasis detection part 3 detects emphasis information described in the text which is outputted from a text input part 1 and outputs it to an emphasis level conversion part 4. Further, a parameter conversion value determination part 5 converts the emphasis level outputted by the emphasis level conversion part 4 into a physical parameter and outputs the parameter conversion value to an emphasis processing part 7 together with emphasis position information. Then a text speech conversion part 6 collates the text contents outputted from a text memory 2 with a previously stored dictionary to grant pronunciation and rhythm information, and combines previously stored phonemes for the pronunciation to generate a speech parameter. The emphasis processing part 7 performs emphasis processing for the speech parameter by using the emphasis position information and parameter conversion value from the speech parameter conversion part 5.

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 conversion method and device for synthesizing speech from text, and particularly, to extract an emphasized portion written in the text and output the emphasized speech of the extracted emphasized portion. By doing so, the present invention relates to a text-to-speech conversion method and device for clearly expressing the intention expressed in text as voice.

[0002]

2. Description of the Related Art With the advance of voice processing technology, synthetic voice devices are used in cash service machines in banks, post offices and other cash handling offices, parking ticket issuing machines in parking lots and other guidance devices. However, synthetic speech is not often adopted in information providing services such as telephone services. This is because the conventional rule-synthesized speech is a monotonous speech because the volume of the output speech and the utterance speed are constant and average, and the intention is not sufficiently transmitted. Due to that.

In the case of character expression, various character styles, parentheses and other expression techniques can be used to emphasize our intention and convey it to the reader. In the conventional text-to-speech conversion method, even if the font is changed in the text, it is ignored. Also, the parentheses in the text were either read as "brackets" as they were, or they were ignored. Therefore, even if parentheses were added in the text or the font was changed in order to emphasize our intention, there was no meaning in the conversion of the text voice. In fact, it goes without saying that it is desirable that the emphasized portion described in the text is output as emphasized voice.

[0004]

DISCLOSURE OF THE INVENTION The present invention solves the problems described above by extracting the emphasized portion described in the text and emphasizing and outputting the voice of the extracted emphasized portion to output the text. The present invention provides a method and apparatus for converting a text voice that clearly expresses the intention expressed in (1) as voice.

[0005]

[Means for Solving the Problems] Collating a prestored dictionary with text, adding reading information and prosodic information to this, and combining prestored speech units based on the reading information. In the text-to-speech conversion method for outputting speech corresponding to text, a text-to-speech conversion method for extracting a emphasized portion written in the text and outputting a sound emphasizing the extracted emphasized portion is configured, and the text Extract the emphasized part by changing the font inside,
A voice that emphasizes the extracted emphasized portion with the determined strength by determining the strength of the emphasized sound of the extracted emphasized portion by the correspondence between the predetermined type of font and the emphasized level of the sound. A text-to-speech conversion method for outputting the emphasized part using parentheses in the text, and the emphasized part extracted by the correspondence between the predetermined bracket type and the emphasis level of the voice. A text-to-speech conversion method for deciding the strength of the emphasis of the voice and outputting a voice in which the strong portion extracted with the determined strength is output, and further comprising a text input unit 1, and a text input unit 1 is provided with a text memory 2 in which the text output is stored, and an emphasis detection unit 3 for detecting emphasis information composed of emphasis position information about an emphasized part described in the text and notation method information. The text content output from the text memory 2 includes an emphasis level converter 4 for converting the emphasis level based on emphasis information, and a parameter conversion value determiner 5 for converting the emphasis level into a physical parameter. Is provided with a text-to-speech conversion unit 6, which includes an emphasis processing unit 7 that performs emphasis processing on the sound parameter based on the emphasis position information and the parameter conversion value output from the sound parameter conversion unit 5. A text-to-speech converter having a speech output unit 8 for electroacoustic conversion of speech parameters output from the processing unit 7 is also configured.

[0006]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT An embodiment of the present invention will be described with reference to FIG. 1 is a text input section. Various methods such as a method of newly inputting text with a keyboard, a method of inputting text from a text already stored on a hard disk or a floppy disk, and other various methods can be adopted for inputting text by the text input unit 1.
The text input by the text input unit 1 is then stored in the text memory 2. In addition, once entered text can be edited using a display, a keyboard, a mouse or other input device.

The text memory 2 stores the text output from the text input section 1, and the text is output from the text memory 2 to the text-to-speech conversion section 6 and the emphasis detection section 3, respectively. The emphasis detection unit 3 detects the emphasis information written in the text of the text output from the text input unit 1 and outputs it to the emphasis level conversion unit 4. The emphasis information is composed of emphasis position information and information on the notation method of the emphasis part in the text.

Here, the emphasis information detection method will be described. First, a method in which a font is adopted as a notation method for an emphasized portion in text and the emphasis information is detected from a font change position where the font changes in the text will be described with reference to FIG. The change position of the font is detected by comparing the font of the input text with the standard font. The start position and the end position of the font different from the standard font in the text are output as emphasis position information, and the type of the font detected at that time is used as type information, which are output to the emphasis level conversion unit 4. To do. FIG. 2A is an example of text input in the case where an emphasized portion is extracted according to the change position of the font in the text. In the entered text "It is cloudy in the Kanto region during the daytime, but it will start raining at about 6 pm.", "Kanto region is" is in italics and "from about 6 pm" The parts are shown in bold type. When this example is input, first, the italic font different from the standard font is detected at "Seki", and then up to "ha". The position of “Seki” and the position of “ha” are the emphasis position information. Next, a font “bold” different from the standard font is detected in “evening” and even up to “ra”. The position of "evening" and the position of "ra" are the emphasis position information. The first detected emphasis position and “italic” as the type of font, and the next detected “bold” as the emphasis position and type of font are output to the emphasis level converter 4.

Next, referring to FIG. 2B, description will be given of a method in which parentheses are adopted as a notation method of an emphasized part in text and the emphasized part is detected from the position of the bracket in the text. The position after the position where the parenthesis is detected in the text and the position immediately before the position of the next detected parenthesis are used as the emphasis position information, and the type of the parenthesis at that time is used as the type information information. Output to the emphasis level converter 4. Here, for example, if the left parenthesis and the right parenthesis are different types, or if there is only one parenthesis, and if other parenthesis positions do not correspond to the parenthesis types, ignore these parentheses or display a message on the display. Use a method that outputs. Figure 2
(B) is an example of text input in the case of extracting an emphasized part according to the type of parentheses in the text. In the entered text "The latest" CD radio-cassette "has a" remote control """,the" CD radio-cassette "part is"".
, The "remote control" part is enclosed in brackets "". When this example is entered, the bracket "is detected first, then the bracket" is detected. The position of "C" immediately after the bracket "is detected and the position of" S "immediately before the bracket" is detected becomes the emphasized position information. Next, the bracket "is detected, and then the bracket" is detected. The position of "ri" immediately after the bracket "is detected and the position of" n "immediately before the bracket" is detected becomes the highlighted position information. The first detected emphasis position and "" as the type of brackets and the next detected emphasis position and "" as the type of brackets are
Each is output to the emphasis level converter 4.

The emphasis level conversion unit 4 converts these into emphasis levels based on the emphasis position information of the emphasis part in the text output from the emphasis detection unit 3 and the notation method information. The conversion method of the emphasis level may be any method as long as the correspondence between the notation method and the emphasis level can be obtained, such as a conversion table that associates the notation method of the emphasized part in the text with the emphasis level. The converted emphasis level is output to the parameter conversion value determination unit 5 together with the emphasis position information.

First, a conversion method for converting the emphasis level using a conversion table that associates the type of font and the emphasis level will be described with reference to FIG. Here, the emphasis level is converted based on the type of font output from the emphasis detection unit 3, and the emphasis level is output to the parameter conversion value determination unit 5 together with the emphasis position information. FIG. 3A shows an example of the conversion table of the emphasis level corresponding to the type of font. The higher the emphasis level, the stronger the emphasis. In this example, the correspondence between the type of font and the emphasis level is set to five levels, but any number of levels may be used. When a font that is not on the conversion table of the emphasis level is detected, this is made to correspond to the intermediate level 3, for example. In the text input example of FIG. 2A, the font information “italic” and “boldface” output from the emphasis detection unit 3 is “1” according to the conversion table of FIG. Is converted to "and bold"
Means that the emphasis level is converted to "3".

Next, a conversion method for converting the emphasis level using a conversion table that associates the bracket type with the emphasis level will be described with reference to FIG. Here, the emphasis level is converted based on the type of parenthesis output from the emphasis detection unit 3, and the emphasis level is output to the parameter conversion value determination unit 5 together with the emphasis position information. FIG. 3B shows an example of the conversion table of the emphasis level corresponding to the type of bracket. The higher the emphasis level, the stronger the emphasis. In this example, the correspondence between the bracket type and the emphasis level is set to five levels, but any number of levels may be set. If a parenthesis that is not on the conversion table of the emphasis level is detected, it may be associated with an appropriate emphasis level. In the text input example of FIG. 2B, the bracket type information “” and “” output from the emphasis detection unit 3 is based on the conversion table of FIG. 3B, and the bracket “” has an emphasis level of “2”. And the bracket “” is converted to the emphasis level “4”.

The parameter conversion value determination unit 5 converts the emphasis level output from the emphasis level conversion unit 4 into a physical parameter, and outputs the parameter conversion value to the emphasis processing unit 7 together with the emphasis position information. As the conversion method, any method may be used as long as the correspondence between the emphasis level and the parameter conversion value is obtained by a conversion table or any other method can be obtained as long as the correspondence between the emphasis level and the voice parameter is obtained. Parameter conversion is performed by, for example, "Speech synthesizer (Japanese Patent Application No.
800) ”can be adopted.
An example of the parameter conversion table corresponding to the emphasis level is shown in FIG. In the example shown in the figure, the emphasis level and the voice parameter are converted in five steps.
It suffices to prepare a voice parameter conversion value corresponding to the emphasis level in.

In the text input example shown in FIG. 2A, the emphasis level "1" output from the emphasis level conversion unit 4 is increased in volume by 3 dB and the fundamental frequency is 1. 03 times, speech rate 0.85
Double, a 200 ms pause is inserted before the emphasized part, and a 150 ms pause is inserted after the emphasized part. The emphasis level “3” output from the emphasis level conversion unit 4 is increased by 6 dB in the volume and 1.05 times in the fundamental frequency according to the conversion table of FIG.
The utterance speed is 0.75 times, a 330 ms pause is inserted before the emphasized part, and a 210 ms pause is inserted after the emphasized part.

In the text input example of FIG. 2B,
Enhancement level “2” output from the enhancement level conversion unit 4
4 increases the volume by the conversion table of FIG. 4, the fundamental frequency is 1.04 times, the speech rate is 0.80 times, a 280 ms pause is inserted before the emphasized part, and 1 after the emphasized part.
An 80-millisecond pause is inserted and converted into a parameter conversion value. In addition, the emphasis level “4” output from the emphasis level conversion unit 4 has a volume increased by 7 dB and a fundamental frequency of 1.
10 times, speech speed 0.70 times, 400 before emphasis
It is converted into a parameter conversion value of inserting a millisecond pause and inserting a 240 millisecond pause after the emphasized portion.

Reference numeral 6 denotes a text-to-speech conversion unit, which collates the text contents output from the text memory 2 with a pre-stored dictionary and adds reading and prosody information to the dictionary.
Combine speech units stored in advance based on reading,
Generate voice parameters. When converting from text to reading, any suitable conversion method proposed to date can be selected and adopted, but the volume, fundamental frequency, and speed of the sound can be changed because of the emphasis processing described later. A conversion method that can be controlled, such as an LPC synthesis method,
It is desirable that the method adopts the LSP synthesis method. The voice parameter generated here is output to the emphasis processing unit 7. In the text input example in Figure 2 (a), "Kanto region is cloudy during the day, but it will start to rain from around 6pm." Is converted into a voice parameter, but changes in the font are converted into voice parameters. It will not be done. In the text input example of FIG. 2B, "the recent CD radio-cassette has a remote controller." Is converted into a voice parameter, but the parentheses are not converted into a voice parameter.

The emphasis processing unit 7 applies emphasis processing to the voice parameter output from the text-to-speech conversion unit 6 based on the emphasis position information and the parameter conversion value output from the voice parameter conversion unit 5. As the processing method, for example, a general method is adopted in which the interval of the drive pulse for voice synthesis is changed for the fundamental frequency and the frame period of synthesis is converted for the speech rate. Since the emphasized voice parameter is modified in the voice parameter of the emphasized portion, a parameter discontinuity occurs with the voice parameter other than the emphasized portion. In order to remove this, the parameters before and after the emphasized portion are smoothed and then output to the audio output unit 8.

For the text input example shown in FIG. 2 (a), the voice parameter "Kanto region is cloudy during the daytime, but it will start to rain from around 6pm in the evening. Is emphasized based on the emphasis position and the parameter conversion value output from the voice parameter converter 5. From the initial emphasis position and the parameter conversion value, “Kanto region is” is subjected to the emphasis process of the parameter conversion value of the emphasis level “1”, and the emphasized voice parameter is created. From the next emphasis position and the parameter conversion value, the parameter conversion value of the emphasis level “3” is emphasized for “from about 6 o'clock in the evening” to create the emphasized voice parameter. The portion that does not correspond to the emphasized position is the voice parameter output from the text-to-speech conversion unit 6 as it is, and no emphasis process is performed.

In the case of the text input example shown in FIG. 2 (b), voice parameter conversion is performed for the voice parameter "latest CD radio-cassette is equipped with a remote controller." The emphasis processing is performed by the emphasis position and the parameter conversion value output from the unit 5. With respect to the “CD radio cassette player”, the parameter conversion value of the emphasis level “2” is emphasized based on the initial emphasis position and the parameter conversion value, and the emphasized voice parameter is created. With respect to the "remote control" from the next emphasis position and the parameter conversion value, the parameter conversion value of the emphasis level "4" is emphasized, and the emphasized voice parameter is created. The portion that does not correspond to the emphasized position is the voice parameter output from the text-to-speech conversion unit 6 as it is, and no emphasis process is performed.

Reference numeral 8 is a voice output unit. In the voice output unit 8, the digital voice signal which is the voice parameter output from the emphasis processing unit 7 is digital-analog converted, and the voice is output by a speaker or other electroacoustic converter. The text input examples shown in FIGS. 2A and 2B are examples in which two emphasized parts are included in the text, but the same applies even if the number of emphasized parts included in the text is large. The emphasized voice can be output. Although the conversion between the type of font or the type of bracket and the emphasis level is set in five steps, the emphasis level may be set in any number of steps. When converting the emphasis written in the text to the emphasis level, the parameter conversion value corresponding to the emphasis level may be changed. For example, when conversion is performed using the conversion table, the parameter conversion table in the parameter conversion value determination unit 5 should be prepared with parameter conversion values having the same emphasis level as the number of emphasis levels in the emphasis level conversion table in the emphasis level conversion unit 4. Good.

[0021]

As described above, according to the text-to-speech conversion method and apparatus of the present invention, it is possible to output a voice that reflects the type of parentheses and fonts in the input text, resulting in a sentence. It is possible to express the keyword that the creator of the above expresses on the text and the voice that makes the focus of the topic clear. Also, with respect to a sentence that has already been created, it is possible to output expressive voice by a simple operation of enclosing the portion to be emphasized in parentheses or changing the font. According to the text-to-speech conversion method of the present invention, it is expected that the intention and emotion of the text creator can be reflected in the synthesized voice, which tends to be monotonous, and the range of application of the synthesized voice is expanded.

[Brief description of drawings]

FIG. 1 is a block diagram of an embodiment of the present invention.

2A and 2B are diagrams showing an example of text input for detecting an emphasized part, FIG. 2A is a diagram showing an example of text input for detecting an emphasized part from a change position of a font, and FIG. It is a figure which shows the example of the text input which detects a part.

FIG. 3 is an emphasis level conversion table, in which (a) is an emphasis level conversion table for font types and (b) is an emphasis level conversion table for bracket types.

FIG. 4 is a parameter conversion table for emphasis levels.

[Explanation of symbols]

 1 Text Input Section 2 Text Memory 3 Emphasis Detection Section 4 Emphasis Level Conversion Section 5 Parameter Conversion Value Determining Section 6 Text Speech Conversion Section 7 Enhancement Processing Section 8 Speech Output Section

Claims (4)

[Claims] 1. Corresponding to text by collating a pre-stored dictionary with text, adding reading information and prosodic information to this, and combining pre-stored speech units based on the reading information. In the text-to-speech conversion method for outputting a speech, the emphasized portion described in the text is extracted, and the speech in which the extracted emphasized portion is emphasized is output. 2. The text-to-speech conversion method according to claim 1, wherein an emphasized portion is extracted by a change of a font in the text, and a predetermined correspondence between the font type and the voice emphasis level is used. A text-to-speech conversion method, characterized in that the strength of emphasis of the extracted emphasized part is determined, and the emphasized part of the extracted part is output with the determined strength. 3. The text-to-speech conversion method according to claim 1, wherein the emphasized portion is extracted by using parentheses in the text, and a predetermined correspondence between the type of parentheses and the emphasis level of voice is used. A text-to-speech conversion method, characterized in that the strength of emphasis of a voice of an extracted emphasized portion is determined, and a voice in which the extracted strong portion is emphasized with the determined strength is output. 4. A text input section is provided, and a text memory for storing the text output from the text input section is provided, and emphasis position information and notation method information about an emphasized part described in the text. And a parameter conversion value determining unit for converting the emphasis level into a physical parameter. , A text-to-speech conversion unit that converts the text content output from the text memory into a speech parameter, and performs emphasis processing on the speech parameter based on the emphasis position information and the parameter conversion value output from the speech parameter conversion unit 5. An audio output unit that includes an emphasis processing unit and performs electroacoustic conversion of the audio parameter output from the emphasis processing unit. A text-to-speech converter characterized by being provided.