JP3862300B2 - Information processing method and apparatus for use in speech synthesis - Google Patents

Description

[0001]
[Industrial application fields]
The present invention relates to a speech synthesis editing method and apparatus, and more particularly to a method and apparatus for processing information used for speech synthesis in a method and apparatus for speech synthesis based on sentence information decomposed into accent phrase unit information.
[0002]
[Prior art]
In general, in a speech synthesizer, text information representing acoustic information called phonetic text is input to an acoustic processing unit.
In order to generate the phonetic text, sentence analysis processing such as morphological analysis and phrase relation analysis is necessary.
[0003]
Using this sentence analysis result, generation and reading of pause information (how long to speak at once (pause phrase) and the length of silent time between pause phrases) and reading (corresponding to changes in reading) ), Accent generation (accent combination), and the like, speech language processing for generating synthesized speech is performed based on information obtained from language analysis, and the result is output as phonetic text. An example of phonetic text is shown in FIG.
[0004]
In the same figure, symbols and numerical values surrounded by “{}” such as H85, T300, etc. are control signals and alphabets enclosed in “[]” which are their values, phonemes, and “/”, “\” Represents the rising and falling points of the accent.
The prosody control unit analyzes (interprets) the phonetic text and extracts information such as accent information, pose information, and pitch frequency necessary for generating prosody information therefrom. Accent information is obtained from “/” and “\”, pause information is T300 to 300 (milliseconds), and pitch frequency is H85 to 85 (Hz). The prosody control unit generates a pitch pattern from these pieces of information. (See Figure 6)
As described above, the phonetic text includes acoustic information such as phoneme information, accent information, and control information.
[0005]
[Problems to be solved by the invention]
However, even if the user corrects the phonetic text to create his or her favorite synthesized speech, the language information is hardly written on the phonetic text, so the information that can be manipulated (changed or corrected) is limited. It has been.
Moreover, even when trying to perform prosodic control using language information, the language information is already missing. If the language information is represented on the phonetic text, there is a problem that the phonetic text becomes difficult to read and difficult to understand.
[0006]
The present invention solves the above-mentioned conventional drawbacks, and does not generate phonetic text from sentence analysis results, but provides speech processing with finer control by providing language information obtained by analysis to the acoustic processing unit, It is an object of the present invention to provide a speech synthesis editing method and apparatus capable of outputting various and high quality synthesized speech.
[0007]
[Means for Solving the Problems]
In order to achieve the above object, an information processing method and apparatus used for speech synthesis of the present invention comprises the following arrangement. That is, a method for processing information used in speech synthesis for performing speech synthesis based on sentence information decomposed into accent phrase units, a part of speech list holding step for holding a list of parts of speech of words included in an accent phrase; A part-of-speech determination step for determining a part-of-speech existing in a part-of- speech list held by the part-of-speech list holding step; a prosody control step for controlling a prosody based on the part-of-speech determined in the part-of-speech determination step; A generating step for generating a synthesized speech parameter based on the prosody controlled in the prosody control step.
[0008]
Another invention is a processing device for information used for speech synthesis for speech synthesis based on sentence information decomposed into units of accent phrases, and holds a list of parts of speech of words included in the accent phrases Part-of-speech list holding means, part-of-speech determination means for determining a part-of-speech existing in a part-of- speech list held by the part-of-speech list holding means, and control of prosody based on the part-of-speech determined by the part-of-speech determination means Prosody control means and generation means for generating synthesized speech parameters based on the prosody controlled by the prosody control means.
[0009]
[Action]
In the above configuration,
Prosody control based on the accent phrase unit information,
A synthesized speech parameter is generated based on the prosody controlled in the prosody control step.
[0010]
Another invention is:
Prosody control means performs prosody control based on the accent phrase unit information,
Generation means generates a synthesized speech parameter based on the controlled prosody.
[0011]
【Example】
First, after summarizing one of the points of the speech synthesizer according to the embodiment of the present invention, a detailed description will be given.
The speech synthesizer of the present embodiment includes an input holding unit that accepts as input a speech language processing result having a frame structure as shown in FIG. 4, and a prosody that generates prosody control information such as pitch, duration, and power. A control unit, a prosody information holding unit that holds prosody information, a voice parameter generation unit that generates synthetic voice parameters such as phoneme, parameter connection, or waveform connection, a voice parameter holding unit that holds voice parameters, A DA converter that converts voice parameters into synthesized voice is basically provided.
[0012]
With this basic configuration, the output of sentence analysis can be expressed in a frame structure, which will be described later, and fine control can be performed based on the data of this frame structure, enabling a variety of synthesized speech outputs.
In the following, the audio configuration apparatus of the present embodiment will be described in detail with reference to the drawings.
FIG. 2 is a block diagram showing the configuration of the speech synthesizer of this embodiment. In the figure, 201 is an input holding unit that inputs and holds frame structure data that is the result of sentence analysis, 202 is a frame structure data that is input from the input holding unit 201, and includes prosody, pitch, duration length, This is a prosody control unit that controls power and the like. Reference numeral 203 denotes a prosody information holding unit that holds prosody control information. Reference numeral 204 denotes a voice generation unit that performs processing such as prosody, parameter connection, and waveform connection. Reference numeral 205 denotes a voice parameter holding unit that holds parameters for generating a synthesized voice. Reference numeral 206 denotes a DA converter that converts voice parameters into synthesized voice.
[0013]
Next, the operation of the present apparatus will be described with reference to the flowchart shown in FIG.
First, in step S301, it is determined whether there is an input of a speech language processing result represented by a frame structure. Here, the frame structure refers to that shown in FIG.
The frame structure is a set of pairs consisting of a key and its value. In FIG. 4, the frame number, notation, accent type, etc. are keys, and each value is “1”, “Today”, “1”. is there. Further, the value can point to the next frame by using the key as a pointer (corresponding to the arrow in FIG. 4). The frame in this figure has an accent phrase as a unit. An accent phrase is a unit having one accent kernel.
[0014]
The “phonetic notation” is a phonetic string that is uttered when generating synthesized speech and is expressed in katakana. The “accent type” expresses the position of the accent nucleus by the number of mora from the beginning of each accent phrase. Here, the “number of mora” indicates the total number of vowels, prompting sounds, and repelling sounds included in the accent phrase. “Component part of speech” is a list of parts of speech of morphemes constituting the accent phrase. Furthermore, the “pause length” represents the length between silent phrases generated after the accent phrase.
[0015]
Returning to FIG. 3, if there is an input in step S301, the process proceeds to step S302. If there is no input, step S301 is repeated.
In step S302, the prosody control unit 201 generates prosody control information such as prosody control, pitch generation, and duration time. The generated prosodic information is held in the prosodic information holding unit 203, and the process proceeds to step S303.
[0016]
The details of the prosody control method in this step are as follows. FIG. 7 shows a control flow of the prosody control unit 201, and will be described for each step. FIG. 8 shows a detailed configuration of the prosody control unit 201.
Here, it is assumed that the leading element of the input accent phrase frame is set in the accent phrase information holding unit 801 in step S301 described above.
[0017]
In step S <b> 701, the accent phrase existence determination unit 802 determines whether an accent phrase is set in the accent phrase information holding unit 801. If it does not exist, the process ends. If it exists, the process proceeds to step S702.
In step S702, the noun determination unit 803 determines whether there is a noun in the morpheme constituting the accent phrase. If not, the process proceeds to step S704. If it exists, the process moves to step S703.
[0018]
In step S703, prosodic information is generated. The pitch frequency setting unit 807 adds a width of +10 to the difference between the maximum value and the minimum value of the pitch frequency given as a default value. In other words, this increases the amplitude value of the pitch frequency and produces a pseudo enhancement effect. In order to increase the ease of listening, the phoneme time length setting unit 808 extends the phoneme time length, and the power setting unit 809 adjusts the power level. Then, the next frame of the accent phrase is set in the accent phrase information holding unit 801, and the process returns to step S701.
[0019]
Hereinafter, from step S704 to step S710, processing for changing the pitch frequency, phoneme length, and power is performed for each part of speech of each constituent morpheme.
In step S704, the moving verb determination unit 803 determines whether there is a moving verb in the morphemes constituting the accent phrase. If not, the process moves to step S706. If it exists, the process moves to step S705.
[0020]
In step S705, prosodic information is generated. The pitch frequency setting unit 807 adds a width of +30 to the difference between the maximum value and the minimum value of the pitch frequency given as a default value. In other words, this increases the amplitude value of the pitch frequency and produces a pseudo enhancement effect. In order to increase the ease of listening, the phoneme time length setting unit 808 extends the phoneme time length, and the power setting unit 809 adjusts the power level. Then, the next frame of the accent phrase is set in the accent phrase information holding unit 801, and the process returns to step S701.
[0021]
In step S706, the numeral determination unit 805 determines whether there is a numeral in the morpheme constituting the accent phrase. If not, the process moves to step S708. If it exists, the process moves to step S707.
In step S707, prosodic information is generated. The pitch frequency setting unit 807 adds a width of +10 to the difference between the maximum value and the minimum value of the pitch frequency given as a default value. In other words, this increases the amplitude value of the pitch frequency and produces a pseudo enhancement effect. In order to increase the ease of listening, the phoneme time length setting unit 808 extends the phoneme time length, and the power setting unit 809 adjusts the power level. Then, the next frame of the accent phrase is set in the accent phrase information holding unit 801, and the process returns to step S701.
[0022]
In step S708, the particle determining unit 806 determines whether there is a particle in the morpheme constituting the accent phrase. If not, the process moves to step S710. If it exists, the process moves to step S709.
In step S709, prosodic information is generated. The pitch frequency setting unit 807 adds a width of −10 to the difference between the maximum value and the minimum value of the pitch frequency given as a default value. That is, this reduces the amplitude value of the pitch frequency. The phoneme time length setting unit 808 shortens the phoneme time length, and the power setting unit 809 adjusts the power level. Then, the next frame of the accent phrase is set in the accent phrase information holding unit 801, and the process returns to step S701.
[0023]
In step S710, prosodic information is generated. In the pitch frequency setting unit 807, the difference between the maximum value and the minimum value of the pitch frequency given as a default value is used as it is. The phoneme time length setting unit 808 sets the phoneme time length as a default value, and the power setting unit 809 sets the default power level. Then, the next frame of the accent phrase is set in the accent phrase information holding unit 801, and the process returns to step S701.
[0024]
Appropriate prosodic control information can be generated by the processing of each step described above.
Next, returning to FIG. 3, the processing from step S303 will be described.
In step S303, the speech generation unit 204 uses the prosody information stored in the prosody information storage unit 203 to generate phonemes, generate waveforms from parameters, or connect waveforms. The generated voice synthesis signal is held in the voice signal holding unit 205, and the process proceeds to step S304.
[0025]
In step S304, the voice parameter information is D / A converted into actual synthesized voice. When the synthesized voice is uttered, all the processes are terminated.
FIG. 5 shows the configuration of an embodiment of the present invention. A control memory 501 stores a control program according to a control procedure as shown in the flowchart of FIG. A central processing unit 502 performs determination / calculation in accordance with a control procedure held in the control memory 501. A memory 503 includes an input holding unit 201, a prosody control unit 202, a prosody information holding unit 203, a voice generation unit 204, and a voice signal holding unit 205. Reference numeral 504 denotes a disk device, and 505 denotes a bus.
(Other examples)
1. In the above embodiment, the case where the speech parameter generation is performed after the prosody control in FIG. 3 has been described, but the present invention is not limited to this, and the prosody control and the speech parameter generation may be performed in parallel. Also,
2. In the above embodiment, the case where the frame structure is input has been described. However, the present invention is not limited to this, and any input format having a similar structure may be used.
3. In the above embodiment, the example of prosody generation related to the case where the accent component and the natural effect component are separated has been described. However, the present invention is not limited to this. You may keep it.
4). In the above embodiment, the case where the prosodic information is generated by the component parts of speech has been described. However, the present invention is not limited to this, and other factors such as presence / absence of pause, presence / absence of emphasis information, and the number of mora are used. Prosodic information may be generated.
[0026]
The present invention may be applied to a system composed of a plurality of devices or an apparatus composed of a single device. Needless to say, the present invention can also be applied to a case where the present invention is achieved by supplying a program to a system or apparatus. As described above, according to the present invention, since frame-structured data is input instead of phonetic text, it is possible to handle more information than before and generate finer synthesized speech. The effect is obtained.
[0027]
【The invention's effect】
As described above, according to the present invention,
By providing linguistic information obtained by the analysis to the acoustic processing unit, it is possible to control speech synthesis more finely and to output various and high quality synthesized speech.
[Brief description of the drawings]
FIG. 1 is a block diagram showing an example of phonetic text.
FIG. 2 is a block diagram showing a configuration of an embodiment of the present invention.
FIG. 3 is an operation flowchart showing a processing procedure according to the embodiment of the present invention.
FIG. 4 is a diagram illustrating an example of a frame structure.
FIG. 5 is a basic configuration diagram of a synthetic speech creation support apparatus according to the present invention.
FIG. 6 is a diagram showing a general pitch pattern generation method.
FIG. 7 is a flowchart showing a flow of processing in the prosody control unit.
FIG. 8 is a diagram illustrating a configuration of a prosody control unit.
[Explanation of symbols]
201 Input Holding Unit 202 Prosody Control Unit 203 Prosody Information Holding Unit 204 Speech Generation Unit 205 Speech Parameter Holding Unit 206 DA Converter 501 Control Memory 502 Central Processing Unit 503 Memory 504 Disk Device 505 Bus 801 Accent Phrase Information Holding Unit 802 Accent Phrase Information Presence Determination Unit 803 noun presence determination unit 804 impression verb presence determination unit 805 number presence determination unit 806 particle presence determination unit 807 pitch frequency setting unit 808 phoneme time length setting unit 809 power setting unit

Claims (18)

An information processing method used for speech synthesis for performing speech synthesis based on sentence information decomposed into accent phrase units,
A part-of-speech list holding step for holding a part-of-speech list of words included in the accent phrase;
A part-of-speech determination step for determining a part-of-speech existing in a part-of- speech list held by the part-of-speech list holding step;
Based on the part of speech determined in the part of speech determination step, the prosody control step of controlling the prosody,
And a generation step of generating a synthesized speech parameter based on the prosody controlled in the prosody control step. A method for processing information used for speech synthesis, comprising: The processing of information used for speech synthesis according to claim 1, wherein the prosody control step controls the pitch frequency which is one of the prosody based on the type of part of speech determined in the part of speech determination step. Method. The method for processing information used for speech synthesis according to claim 1, wherein the prosody is a phoneme duration. The information processing method used for speech synthesis according to claim 1, wherein the prosody is sound power. The method for processing information used for speech synthesis according to claim 1, further comprising a conversion step of converting into synthesized speech based on the synthesized speech parameter generated in the generating step. 3. The voice according to claim 2, wherein the prosody control step performs control to increase the maximum value and the minimum value of the pitch frequency if the part of speech determined in the part of speech determination step is a noun. Information processing method used for synthesis. The processing of information used for speech synthesis according to claim 3, wherein the prosodic control step performs control to increase the phonological duration if the part of speech determined in the part of speech determination step is a noun. Method. The method for processing information used for speech synthesis according to claim 4, wherein the prosody control step performs control to increase the sound power if the part of speech determined in the part of speech determination step is a noun. The part-of-speech determination step, the processing of the case of a noun in the list of parts of speech to be held by the part of speech list holding step is present, information according to claim 1, wherein the determining a part of speech to use該名the prosody generation Method. An information processing apparatus used for speech synthesis for performing speech synthesis based on sentence information decomposed into accent phrase units,
A part of speech list holding means for holding a list of parts of speech of words included in the accent phrase;
A part of speech determination means for determining a part of speech existing in a list of part of speech held by the part of speech list holding means;
Based on the part of speech determined by the part of speech determination means, the prosody control means for controlling the prosody;
An apparatus for processing information used for speech synthesis, comprising: generating means for generating a synthesized speech parameter based on the prosody controlled by the prosodic control means. The processing of information used for speech synthesis according to claim 10, wherein the prosody control means controls the pitch frequency which is one of the prosody based on the type of part of speech determined by the part of speech determination means. apparatus. The information processing apparatus used for speech synthesis according to claim 10, wherein the prosody is a phoneme duration. The information processing apparatus used for speech synthesis according to claim 10, wherein the prosody is sound power. The apparatus for processing information used for speech synthesis according to claim 10, further comprising conversion means for converting into synthesized speech based on the synthesized speech parameter generated by the generating means. The voice according to claim 11, wherein the prosodic control means performs control to increase the maximum value and the minimum value of the pitch frequency if the part of speech determined by the part of speech determination means is a noun. Information processing device used for synthesis. The processing of information used for speech synthesis according to claim 12, wherein if the part of speech determined by the part of speech determination unit is a noun, the prosody control unit performs control to increase the phonological duration. apparatus. 14. The information processing apparatus for use in speech synthesis according to claim 13, wherein the prosody control means performs control to increase the sound power if the part of speech determined by the part of speech determination means is a noun. The part-of-speech determination unit, the processing of the case where a noun in the list of parts of speech to be held by the part of speech list holding means are present, information according to claim 10, characterized in that the determination as part of speech using a該名the prosody generation Equipment .

Images