JPH05224689A - Speech synthesizing device - Google Patents

Abstract

PURPOSE:To emphasize a specified part in a synthesized speech to an emphasis level represented with a sensual word like 'large' or 'small'. CONSTITUTION:A character string from a text memory 2 receives pronunciation data at a text-speech conversion part 3 and is converted into a speech parameter string corresponding to the character string. An input/output device 1 is used to specify a part to be emphasized in the character string and also to specify whether the emphasis level is high or low by human sensation. The emphasized part in the speech parameter string is inputted to an emphasizing process part 9 according to specification information on the emphasized part. The specified sensual emphasis level is converted by a sensual quantity-physical quantity conversion part 8 into a sound volume conversion value indicating how much sound volume is to be increased, a speaking speed conversion value which indicates how much a speaking speed is to be slowed down, a conversion value indicating how much a fundamental frequency is to be made higher, and a value indicating how much a pause is to be inserted before and after the emphasized part. An emphasizing process for the speech parameter of the emphasis- specified is performed, and the processed parameter and the speech parameter string of an emphasis-unspecified part are put together and outputted.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice synthesizing device for synthesizing and outputting a voice stored in advance, or converting an input character string into voice and synthesizing and outputting the same, and in particular emphasizes a specific portion in the output voice. The present invention relates to a voice synthesizing device that outputs a voice having excellent expressiveness.

[0002]

2. Description of the Related Art Advances in semiconductor technology such as speech processing technology, language processing technology, and microprocessor have led to practical use of speech analysis / synthesis, band compression, and regular speech synthesis that synthesizes and outputs speech by inputting a character string. The stage has been reached. For example, guidance output in services such as order acceptance, information provision, database search, etc. via communication such as telephone,
Speech synthesis technology has been used to provide data and information.

The synthetic output of voice is performed by a recording / editing / playback system in which a voice stored in advance is edited and output as needed, and a voice is input by inputting a character and automatically giving a reading to the given character. Can be divided into a regular speech synthesis method for synthesizing and outputting. Conventionally, in speech synthesis, research and development have been conducted with an emphasis on how to improve the quality of output speech. However, as the voice synthesis technology becomes more general, there is a demand for a voice synthesis method that enables anyone to easily output expressive voice even if they are not sufficiently familiar with the voice technology.

In the synthesis of voice, as a method of improving the expressive ability of synthesized voice, the quality of voice is controlled to a desired sound.
Although various methods such as changing the speaking rhythm are conceivable, the present invention relates to a voice emphasizing technique for emphasizing and outputting a specific phrase. Conventionally, methods for emphasizing and synthesizing a specific part of speech have been developed.In these methods, a method for transforming parameters necessary for speech synthesis when emphasizing processing is prepared in advance, and automatic A method has been adopted in which the parameters of the emphasized part extracted in advance or the part of the emphasized part designated in advance are modified to output synthetic speech. In this case, examples of parameters to be modified include a fundamental frequency (pitch) parameter of voice expressing intonation, voice strength (sound pressure), and poses added before and after the emphasized portion. Regarding such a speech enhancement method, for example, Takeda and Ichikawa “Creation and Evaluation of Prominence Generation Rule for Japanese Sentence Speech” (Acoustic Society of Japan, vol.
47 No. 6, pp. 397-404, 1991) and the like.

[0005]

However, in these methods, (1) originally, there are various levels of emphasizing, and there are various ways from the way of speaking when strongly emphasizing to the way of speaking when weakly emphasizing. Only two types of expressions, with and without emphasis, can be realized. (2) When changing the emphasis level, specialized knowledge about voice generation / synthesis is required and familiar with these technologies. There was a problem that new users could not add new controls and expressions to ordinary users. In order to solve these problems, it is desired to realize a synthetic output method of emphasized speech that automatically controls the physical parameters in response to the human's sensory input for emphasis and matches the human sense. It was rare.

[0006]

According to the present invention, there are provided a means for designating a portion to be emphasized and a means for a human to give various levels of emphasis instruction based on sensory designation of the degree of emphasis of the emphasized portion. , Means for automatically generating a physical control parameter required for voice synthesis control from the emphasized level, and a voice according to the specified emphasis level for the emphasized portion using the physical control parameter. It is possible to synthesize and output a voice in which a specific portion of the output voice is subjected to an emphasis process that matches a human sense, and a synthesized output unit.

[0007]

Embodiments of the present invention will be described below in detail with reference to the drawings. The present invention may be applied to a rule synthesizing method for synthesizing and outputting a voice from a character, and may be applied to an analysis and synthesizing method for outputting a voice accumulated in advance.
In the following, the embodiment will be described by taking the case of application to the rule composition method as an example.

In FIG. 1, the user designates an emphasized portion,
The emphasis level is designated by the input / output device 1, and the contents of the text to be synthesized in the text stored in the text memory 2 are displayed on the input / output device 1 and displayed to the user. Ask the user to specify which part of the character string to emphasize. That is, the user designates the portion of the character displayed on the screen to be highlighted by using the input means such as a mouse and a keyboard. Further, the level of emphasis that indicates how much emphasis should be given to the specified part is specified. At this time, as words expressing the emphasis level, "ordinary emphasis", "strong emphasis", "high emphasis level", "middle emphasis level", etc. can be specified by the human sense, so that Even users who are not familiar with specialized knowledge or technical terms can easily use it. Figure 2A
FIG. 7 shows a display example of the display screen 21 when a display device, a keyboard, a mouse or the like is used as the input / output device 1. The user designates the portion of the character string 22 displayed on the screen 21 to be emphasized (the underlined portion in the example of FIG. 2A) and then instructs the ON of the emphasis switch 23 on the screen 21. Next, the level of emphasis for the emphasized part is displayed on the screen 21.
The slide volume 24 displayed above is designated by a mouse or the like and input. A highlight level display 25 is displayed near the slide volume 24 along the slide direction.

In the text-to-speech conversion unit 3 of FIG. 1, reading is added to the character string read from the text memory 2, and the basic voice unit is combined to correspond to the character string. Generate audio parameters. Various methods have been proposed for text-to-speech conversion (for example, Hakoda, Hirokawa, Mizuno, Nakajima "Prototype of text synthesis board using COC method", Institute of Electronics, Information and Communication Engineers, Speech Study Group, SP 90-55 ( (1990) etc.), it is sufficient to select an appropriate method, but since it is necessary to perform emphasis processing later, the sound volume, fundamental frequency,
A method in which the speed can be controlled (for example, LPC synthesis method, L
It is desirable that the system adopts the SP composition system or the like). The voice parameters generated by the text-voice converter 3 are stored in the voice parameter memory 4.

The control signal corresponding to the emphasized portion designation information input to the input device 1 is the emphasized portion designation information input section 5.
Is converted into a positional information parameter indicating which part of the character string is to be emphasized and is output to the voice-parameter correspondence conversion unit 6. The voice-parameter correspondence conversion unit 6 associates the input positional information parameter with the reading of the text memory 2 and the voice parameter memory 4, and converts the character string stored in the voice parameter memory 4 into a voice parameter. The parameter of the portion to be emphasized in the row is extracted.

On the other hand, the sensational emphasis level designation information input to the input device 1 is converted by the emphasis level input section 7 into a sensation level that represents the sensational emphasis level by a numerical value. As this feeling level, for example, the smallest emphasis level is −
A method of using a numerical value in which the highest emphasis level is 1, the normal emphasis level is 0, and the lowest emphasis level is 0, the highest emphasis level is 1, and the normal emphasis level is 0.5. Although various methods such as a method of using a numerical value corresponding to are conceivable, the case of corresponding from -1 to 1 will be described below as an example.

In the sensory quantity-physical quantity converter 8, a physical parameter conversion value (physical control) for physically transforming a voice parameter from the sensory level representing the sensory emphasis level output from the emphasis level input section 7. Parameter). It is known that when a certain part is emphasized in human utterance, the tone is emphasized, pronounced strongly and slowly, a gap is placed before the emphasized word, and consonant vowels are carefully articulated ( For example, Minoru Wada "Accent
"Intonation Prominence", edited by Souken Tokugawa, "Accent" (Ronshu Japanese Studies 2), Yusendo (Showa 55)
Year))). In this embodiment, as physical parameters representing emphasis, (1) volume (sound pressure), (2) fundamental frequency (pitch), (3) speech speed, (4) pose to be inserted before the emphasized part, (5) ) A case of using a pose inserted after the emphasized part will be described.

The sensory quantity-physical quantity converter 8 calculates a physical parameter conversion value from the sensory level in accordance with a rule previously obtained. For example, a conversion table showing the correspondence between the sense level and the physical parameter conversion value may be prepared, and the physical parameter conversion value may be obtained from the input sense level. FIG. 3 shows an example in which the contents of the table for obtaining the physical parameter conversion value from the sensation level are shown as a graph. Figure 3
A is an example of a diagram for obtaining a value for converting the sense level to the volume of the emphasized portion. For example, when the value of the sense level is 0.5, the volume of the emphasized portion is higher than that when the volume is not emphasized. Increase the value by 7 dB. Further, FIG. 3B is an example of a diagram showing a coefficient for converting a sensation level into an utterance speed of an emphasized portion. For example, when the value of the sensation level is 0.5, the utterance speed of the emphasized section is multiplied by 0.7. And speak slowly. Specifically, when the value of the sense level is 0.5, for example, the frame period of speech synthesis without emphasis is 8 ms, whereas the frame period of the emphasized part is 11.4 ms. It can be realized by a method such as Similarly, FIG.
C is an example of a diagram showing the correspondence between the coefficient for converting the fundamental frequency of the emphasized portion and the sense level, where the sense level is 0.
In the case of 5, the fundamental frequency of the emphasized part is 1.1 times that in the case where no emphasis is applied. 3D and 3E are diagrams showing the correspondence between the length of a pose inserted before the emphasized part, the length of a pose inserted after the emphasized part, and the sense level, for example, when the sense level is 0.5. If 4 before emphasis
Insert a 20 ms pose and a 260 ms pose after the highlight.

Correspondence between the sense level and the physical parameter conversion value differs depending on the gender of the speaker of the voice that is the material for voice synthesis, the content of the document to be handled, etc.
Human speech uttered at various emphasis levels is analyzed to find correspondence, or the results obtained by allowing multiple subjects to hear comparatively synthesized speech with various physical parameters are measured using a psychometric method. Analyzes and finds the correspondence between the physical parameter value and the feeling level of emphasis, or synthesizes voices with different emphasis levels for multiple subjects by a voice control device, voice editing device, etc. that can operate and control physical parameters required for voice synthesis. Is generated to obtain the correspondence between the emphasis level and the parameter value, and the like.
As shown in FIG. 3, the physical parameters are 0.6 to 0.9 times for speech rate and 3 to 9d for volume.
B, 1.05 to 1.15 times the fundamental frequency, 200 to 550 as the length of the pause inserted before the emphasized part
Milliseconds, 150-350 milliseconds as the length of time of the pose inserted after the emphasized portion, and the sensuous emphasis level -1-
It is desirable to convert corresponding to 1.

In the above, an example of converting a sense level into a physical quantity using a conversion table has been described. However, the contents corresponding to the conversion table are stored as a conversion formula representing an approximate expression, and the sense level is converted into a physical control parameter. It can also be realized by a method of converting by calculation. In the example of FIG. 3, the volume increase (dB), the conversion coefficient of the speech speed, the conversion coefficient of the fundamental frequency, the pause length (milliseconds) inserted before the emphasized portion, and the pause length (millisecond) inserted after the emphasized portion. Sec) is a, α ₁ , α ₂ , p ₁ , p ₂ , respectively,
These values can be obtained by calculating the following formula using the feeling level x.

[0016] a = 0.6x ² + 2.6x + 6.0 [dB] ^{_{α 1 = 0.77x 2 -0.11x-0.03}} α 2 = 0.02x 2 + 0.04x + 1.07 p 1 = 44x 2 + 154x + 331 [Milliseconds] p ₂ = 20x ² + 76x + 221 [milliseconds] In FIG. 1, the emphasis processing unit 9 extracts a voice parameter corresponding to the emphasis processing target position obtained by the voice-parameter correspondence conversion unit 6 from the voice parameter memory 4. As well as
Using the physical parameter conversion value (physical control parameter) for the emphasized part obtained by the sensation-physical quantity converter 8, the sound parameter of the emphasized sound part is emphasized. The speech enhancement process is performed by a general method such as converting the synthesis frame period for the speech speed and changing the interval of the speech synthesis drive pulse for the fundamental frequency (for example,
This can be realized by Japanese Patent Application No. 3-180812 "Voice enhancement device".

In the speech synthesizer 10, first, the emphasis processor 9
The speech parameter of the emphasized portion which has been subjected to the emphasizing process and the speech parameter other than the emphasized portion among the speech parameters stored in the speech parameter memory 4 are referred to and combined. At this time, since the modification of the emphasis process is added to the voice parameter of the emphasized portion, the parameter discontinuity occurs before and after that. Therefore, next, the speech parameters before and after the emphasized portion are smoothed to remove the discontinuity of the speech parameters. Finally, a voice signal to be output is synthesized from the obtained final voice parameters, and is further digital-analog converted by the output device 11, and is designated by a sense amount with respect to a section designated by the user from a speaker or the like. The synthesized speech subjected to the enhanced processing is output.

Although the present invention has been described above in the case of being applied to the speech synthesis by the rule synthesizing method, when it is applied to the analysis and synthesizing method for synthesizing and outputting the speech which is analyzed and stored in advance, In FIG. 1, a voice memory is used instead of the text memory 2 and the text-to-speech conversion unit 3, and the voice memory stores the raw voice waveform, power, or analysis code, and the input / output device 1 With respect to, the stored contents of the voice memory (voice synthesis in the case of code storage) is displayed and the portion to be emphasized is designated. FIG. 2B shows a display example of the screen 21 of the input / output device 1 when applied to the analysis and synthesis method. In this case, a portion of the voice waveform 26 or the voice power 27 displayed on the screen 21 to be emphasized is designated by using an input means such as a mouse (in FIG. 2B, a shaded portion below the voice waveform is indicated. ), The emphasis switch 23 is instructed to be turned on, and then, similarly to the example shown in FIG. 2A, the sensory emphasis level is input by operating the slide volume 24 displayed on the screen with a mouse or the like. The emphasis section and the emphasis level can be input by various input means such as a keyboard, a mouse, a volume, and a joystick other than the method shown in this example.

[0019]

As described above, according to the speech synthesizing apparatus of the present invention, the synthesized speech can be changed while freely changing the emphasis level expressed by the sensory words for the designated keyword or phrase. It is possible to output, and as a result, the expressiveness of synthetic speech, which tends to be monotonous, increases, not only can the intention and emotion be expressed by synthetic speech, but also specialized knowledge of speech synthesis. Even a user who does not have the ability to freely control the output of synthetic speech with rich expressiveness in which the intention is embedded.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a speech synthesizer according to the present invention.

FIG. 2A is a diagram showing an example of an input screen when applied to a rule composition method, and B is a diagram showing an example of an input screen when applied to an analysis composition method.

FIG. 3 is a graph showing various correspondences between a sense level and a physical parameter conversion value.

Claims (1)

[Claims] 1. A voice synthesizer for synthesizing and outputting a voice from a stored voice or a character, wherein an emphasized portion designation input means for designating and inputting a portion to be emphasized in output speech, and a degree of enhancement of the emphasized portion are designated. An emphasis level input means for inputting, a physical control parameter converting means for converting the input emphasis level into a plurality of physical control parameters for voice synthesis control, and the designated portion using the physical control parameters A voice synthesizing device comprising: a voice output unit that synthesizes and outputs a voice according to a designated emphasis level.