JP5019807B2 - Speech synthesis apparatus, speech synthesis method, and program for realizing speech synthesis method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a voice synthesizer for outputting voice suppressive in occurrence of distortion. <P>SOLUTION: A waveform generating device 200 included in the voice synthesizer is provided with: an energy changing section 210 for changing energy information Ep to energy information corresponding to volume indication information Ev; an energy correction section 220 for correcting energy information E based on a correction data prepared beforehand; a waveform decoding section 230 for decoding and outputting a waveform of a time unit defined beforehand, from waveform information for each period defined beforehand; and an amplifier section 240 for amplifying and outputting a waveform S[i] generated by decoding, based on energy information f(E) after correction. <P>COPYRIGHT: (C)2008,JPO&amp;INPIT

Description

  The present invention relates to a technique for synthesizing speech. More specifically, the present invention relates to a speech synthesizer for synthesizing and outputting speech, a speech synthesis method, and a program for realizing the speech synthesis method.

  In a device capable of outputting sound, a digital signal cannot be expressed beyond the amplitude range defined by the data format. In order to obtain a loud volume exceeding the range from a digital signal, an amplifier having a higher amplification factor or an efficient speaker is used. However, changing the amplifier or speaker increases the cost of the device.

  Also, in so-called text-to-speech synthesis or speech decoding, which synthesizes speech using character information, there is a problem that the sound is rapidly distorted if the output digital signal exceeds a specified range. In this case, if the volume is reduced in order not to cause distortion, there may be a problem that when a synthesized speech with a low output level is output, it cannot be heard.

Thus, for example, Japanese Patent Laid-Open No. 2001-109500 (Patent Document 1) discloses a pitch processing technique that does not impair the naturalness of speech.
JP 2001-109500 A

  Furthermore, in order to correct sound distortion, for example, in a compressor as a recording device, a technique for making the output volume constant by sequentially adjusting the amplification factor using the average value of energy in a short time of the input waveform is known. ing. However, in order to perform this technique in real time, it is difficult to adjust the response characteristics, and there is a possibility that processing for obtaining the average value is wasted.

  The present invention has been made to solve the above-described problems, and an object of the present invention is to provide a speech synthesizer that can prevent distortion of output speech.

Another object is to provide a speech synthesis method that can prevent distortion of output speech.
Still another object is to provide a program for causing a computer to implement a speech synthesis method that can prevent distortion of output speech.

In order to solve the above problems, according to one aspect of the present invention, a speech synthesizer that synthesizes and outputs speech is provided. The speech synthesizer includes an acquisition unit that acquires data for generating output sound and designation information for specifying the volume of the output sound, and each of the sounds divided in predetermined units. Stores each waveform information for generating a waveform of a part, energy information of each part, and correction data prepared according to a predetermined relationship between energy before and after correction to correct each energy information. storage means for, each energy information, and changing means for changing each of the energy information corresponding to the designated information, by applying the energy information after the change by the change means to the correction data, energy information of the corrected and correcting means for deriving, based on the waveform information, and decoding means for decoding each waveform of the sound output, based on each energy information after correction Amplifying means for amplifying each waveform decoded by the decoding means, based on each waveform amplified by the amplifying means, and output means for outputting sound.

  Preferably, the correcting unit corrects the energy information changed by the changing unit so as to approach an upper limit value defined in advance for the energy information.

Preferably, the correction unit corrects the energy information nonlinearly.
Preferably, the correcting unit continuously corrects the energy information.

  Preferably, the correction data includes energy information before correction and energy information after correction of energy information before correction.

  Preferably, the output means includes waveform connecting means for connecting the waveforms amplified by the amplifying means to generate a synthesized waveform, and audio output means for outputting sound based on the synthesized waveform.

  Preferably, the output unit further includes a waveform saturation unit that adjusts the combined waveform output from the waveform connection unit so as to gradually approach a predetermined upper limit value. The voice output means outputs a voice based on the synthesized waveform adjusted by the waveform saturation means.

  Preferably, the speech synthesizer further includes a driving unit that is mounted with a removable recording medium and drives the recording medium. The recording medium stores audio data and designation information associated with the audio data. The acquisition unit includes a reading unit that reads out audio data and designation information from a recording medium attached to the driving unit.

  Preferably, the acquisition unit includes a reception unit that receives a signal including character information and an input unit that receives an input of designation information.

  Preferably, the acquisition means includes a microphone that receives an utterance and outputs an audio signal corresponding to the utterance, a waveform information analysis means that analyzes the audio signal and outputs waveform information corresponding to the utterance, and energy information corresponding to the utterance Prosodic analysis means for outputting prosodic information including.

A speech synthesizer according to another aspect of the present invention corrects each waveform information for generating a waveform of each part obtained by dividing speech in a predetermined unit, each energy information of each part, and each energy information Therefore, correction data prepared according to a predetermined relationship between energy before and after correction, a memory for storing a program , and a processor for executing the program are provided. The processor obtains data for generating the output sound and designation information for designating the volume of the output sound, and each energy information is converted into energy information corresponding to the designation information. A change step to be changed, a correction step for deriving the energy information after correction by applying each energy information after the change in the change step to the correction data, and each waveform of the sound output based on each waveform information And a step of amplifying each waveform decoded by the step of decoding based on the corrected energy information . The speech synthesizer further includes an output unit that outputs speech based on each amplified waveform.

Preferably, the correction step, such that approaches the predefined upper limit value for the energy information, comprising the step of correcting the energy information changed by the changing step.

Preferably, the correction step includes a step of correcting the energy information in a non-linear.

Preferably, the correction step includes a step of continuously correcting the energy information.

Preferably, the processor further executes a waveform connection step of connecting the waveforms amplified in the amplification step to generate a composite waveform . The output unit outputs sound based on the synthesized waveform.

Preferably, the processor further executes a waveform saturation step of adjusting the composite waveform generated in the waveform connection step so as to approach the predetermined upper limit value . The output unit outputs sound based on the synthesized waveform after adjustment in the waveform saturation step.

  Preferably, the speech synthesizer further includes a drive device that is mounted with a removable recording medium and drives the recording medium. The recording medium stores audio data and designation information associated with the audio data. The acquisition step includes a reading step of reading out audio data and designation information from the recording medium attached to the driving step.

  Preferably, the obtaining step includes a receiving step for receiving a signal including character information and an input step for receiving input of designation information.

  Preferably, the speech synthesizer further includes a microphone that receives an utterance and outputs an audio signal corresponding to the utterance. The obtaining step includes a step of analyzing the audio signal and outputting waveform information corresponding to the utterance, and a step of outputting prosodic information including energy information corresponding to the utterance.

According to another aspect of the present invention, a speech synthesis method for synthesizing and outputting speech is provided. In this method, a step of obtaining data for generating output sound and designation information for designating a volume of the output sound, and a waveform of each part obtained by dividing the sound into predetermined units Loading each waveform information for generating the energy information, each energy information of each part, and correction data prepared according to a predetermined relationship between the energy before and after correction to correct each energy information; A step of changing each energy information to energy information corresponding to the designated information, a step of deriving energy information after correction by applying each energy information after the change in the changing step to the correction data , A step of decoding each waveform of the output sound based on each waveform information and a recovery based on each energy information after correction. A step of amplifying each waveform decoded by the step of, based on each waveform amplified by the step of amplifying comprises a step of outputting speech.

When further another situation of this invention is followed, the program for making a computer provided with memory and a processor implement | achieve the speech synthesis method is provided. The program obtains , in the processor, data for generating the output sound from the memory and designation information for specifying the volume of the output sound, and the sound is defined in advance from the memory. Each waveform information for generating a waveform of each part divided by unit, each energy information of each part, and each energy information were prepared according to a pre-defined relationship between energy before and after correction. deriving a step of reading the correction data, each energy information, and changing the respective energy information corresponding to the designated information, by applying the energy information after the change at the next to the correction data, energy information of the corrected a correction step of, based on each waveform information, the step of decoding each waveform of the sound output, the energy of the corrected Based on the distribution, the steps of amplifying each waveform decoded by the step of decoding, based on the waveform that is amplified, and a step of outputting the audio signal.

  The speech synthesizer according to the present invention can prevent distortion of output speech. According to the speech synthesis method of the present invention, it is possible to output speech while preventing distortion of the speech that is output. According to the program according to the present invention, the computer can realize a speech synthesis method capable of preventing distortion of output speech.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the following description, the same parts are denoted by the same reference numerals. Their names and functions are also the same. Therefore, detailed description thereof will not be repeated.

<First Embodiment>
With reference to FIG. 1, the waveform generation apparatus 100 with which a speech synthesizer is provided is demonstrated. FIG. 1 is a block diagram showing a configuration of functions realized by conventional waveform generation apparatus 100. The waveform generation device 100 includes a waveform decoding unit 110 and amplification units 120 and 130.

  The waveform decoding unit 110 is configured to receive a signal input to the waveform generation device 100. Specifically, the waveform decoding unit 110 receives input of waveform information (spectral information or sound source information). The waveform decoding unit 110 decodes a short-time waveform based on the waveform information, and outputs the decoded waveform (S [i]).

  The amplifying unit 120 is connected to the waveform decoding unit 110 so as to operate based on the output from the waveform decoding unit 110. Specifically, the waveform signal output from the waveform decoding unit 110 and energy information (Ep) corresponding to the waveform signal are input to the amplification unit 120. The amplifying unit 120 amplifies the decoded waveform using the energy information Ep, and outputs the amplified waveform signal.

  The amplifying unit 130 is connected to the amplifying unit 120 so as to operate based on the output from the amplifying unit 120. Specifically, a signal having a waveform amplified by the amplification unit 120 is input to the amplification unit 130. Furthermore, the volume designation information given to the waveform generation device 100 is input to the amplification unit 130. The amplifying unit 130 amplifies the waveform output from the amplifying unit 120 based on the volume designation information Ev, and outputs the waveform data with the volume adjusted.

  With reference to FIG. 2, a waveform generation apparatus 200 provided in the speech synthesizer according to the first embodiment of the present invention will be described. FIG. 2 is a block diagram showing a configuration of functions realized by waveform generating apparatus 200 according to the present embodiment. The waveform generation device 200 includes an energy change unit 210, an energy correction unit 220, a waveform decoding unit 230, and an amplification unit 240 as main components. Specifically, the voice synthesizer is realized by an information processing apparatus or the like including at least a voice output device such as a computer system, a mobile phone, a game device, or the like. In the following description, for speech synthesis, at least processing for processing previously created speech data, processing for lowering digital data based on speech signals acquired by utterance to a microphone, and input characters And a process of synthesizing speech based on character information included in the sequence or received signal.

  The energy changing unit 210 receives input of the volume designation information Ev and the energy information Ep. The volume designation information Ev is given from the outside of the waveform generation device 200. For example, the volume designation information Ev is given as input of volume designation information (for example, operation of a volume button, setting of a level that defines the volume, etc.) to a speech synthesizer including the waveform generation apparatus 200. Alternatively, the volume designation information Ev is written in advance in a memory provided in the speech synthesizer. Furthermore, in another aspect, when the speech synthesizer accepts input of speech data, the volume designation information is associated with the speech data. The energy changing unit 210 changes the energy information Ep to energy information corresponding to the sound volume designation information Ev and outputs it.

  The energy correcting unit 220 is connected to the energy changing unit 210 so as to operate based on the output from the energy changing unit 210. More specifically, the energy correction unit 220 receives input of energy information E changed by the energy change unit 210. The energy correction unit 220 corrects the energy information E based on correction data prepared in advance.

  Preferably, the energy correcting unit 220 corrects the energy information E changed by the energy changing unit 210 so as to gradually approach an upper limit value that is defined in advance for the corrected energy information. Such correction is performed, for example, in accordance with a pre-defined relationship between the energy E before correction and the value f (E) after correction of the energy E.

  In another aspect, energy correction unit 220 corrects energy information E nonlinearly. Here, the non-linearity refers to non-linearity in a case where the relationship between energy information before correction and energy information after correction is defined by a non-linear function as described later. The non-linear relationship is expressed, for example, as a function of second order or higher, or as an exponential function. Note that the entire range of the energy E before correction may not be corrected nonlinearly. For example, a proportional relationship may be included in a partial range of energy before correction.

  In another aspect, the energy correction unit 220 may correct the energy information E based on a table that preliminarily defines the correspondence between the energy information before correction and the energy information after correction. The table is defined as an array of correction coefficients applied to a certain range of energy information, for example.

  In still another aspect, the energy correction unit 220 corrects energy information for each frame to be processed. Further, for example, in PSOLA (Pitch Synchronous OverLap Add) type speech synthesis, the amplitude (energy) of one pitch waveform to be added in each frame is corrected by the energy correction unit 220 and then added by shifting the time by the pitch interval. However, the waveform of the frame may be generated.

  The waveform decoding unit 230 is configured to accept input of data to the waveform generation device 200. Specifically, waveform information for each predetermined section (for example, several milliseconds to several tens of milliseconds) is input to the waveform decoding unit 230. The waveform information includes spectrum information and sound source information. The waveform decoding unit 230 decodes and outputs a predetermined time unit waveform from the waveform information.

  The amplification unit 240 is connected to the energy correction unit 220 and the waveform decoding unit 230 so as to operate based on the output from the energy correction unit 220 and the output from the waveform decoding unit 230. Specifically, the amplification unit 240 receives the corrected energy information f (E) generated by the energy correction unit 220 and the waveform S [i] decoded by the waveform decoding unit 230. 240 is input. The amplification unit 240 amplifies and outputs the waveform S [i] generated by decoding based on the corrected energy information f (E). The data output by the amplifying unit 240 is input, for example, to a digital / analog conversion circuit (not shown) as waveform data whose volume has been adjusted.

  The correction of energy information will be described with reference to FIG. The energy information here is the energy of the speech waveform for each short time. For example, the sum of the squares of the waveform values, the average of the square of the amplitude per sample divided by the number of samples, The average of the amplitude per sample for which the square root is obtained is represented. FIG. 3 is a diagram showing the relationship between energy information E before correction and energy information f (E) after correction. Here, regarding the value of the corrected energy information f (E), the energy information Y (1) is an upper limit value for energy information for preventing distortion in the reproduced sound. The energy information Y (2) is the maximum value that the energy information before and after correction can take. In this case, the reproduced sound may be distorted within the range of the value Y (1) to the value Y (2).

  The energy information E is corrected by a function that defines the relationship (for example, the nonlinear function 310 or the broken line function 320). Data defining the nonlinear function 310 or the broken line function 320 is stored in advance in a memory area of the energy correction unit 220, for example. The energy correcting unit 220 corrects the energy information E output by the energy changing unit 210 using any function.

  Note that the function for correcting the energy information E is not limited to that shown in FIG. It is sufficient that the value of the energy information f (E) after correction does not exceed the upper limit value Y (1).

  As a method for realizing the correction of the energy information E, as described above, in addition to the method of correcting the energy information itself, a method of obtaining the corrected amplification factor at the time of conversion from the energy information to the amplification factor may be used.

  Waveform amplification will be described with reference to FIG. FIG. 4 is a diagram illustrating the relationship between energy information and amplification factor. This relationship is represented, for example, either as an exponential function 410 or by a function 420 that has a linear relationship with the exponential function.

  In FIG. 4, for example, in a range exceeding the energy information E (1), the function 420 is defined to give a lower amplification factor than the exponential function 410. Therefore, in the range exceeding the energy information E (1), waveform amplification is suppressed. That is, the amplification unit 240 suppresses the degree of amplification of the waveform S [i] output from the waveform decoding unit 230 when the corrected energy information exceeding the energy information E (1) is input. Output. In this way, a signal corresponding to the processing capability of the speaker can be output to the speaker that performs audio processing using the volume adjustment waveform data output from the amplification unit 240. As a result, the speaker can output sound without generating sound distortion.

  With reference to FIG. 5, a waveform generation algorithm applied in waveform generation apparatus 200 according to the present embodiment will be described. FIG. 5 is a flowchart showing an operation executed by each unit of the waveform generation device 200.

  In step S510, waveform generation device 200 accepts input of waveform information and energy information Ep. Specifically, the waveform decoding unit 230 accepts input of waveform information. The energy changing unit 210 receives input of energy information Ep. In step S520, waveform decoding section 230 decodes the waveform (S [i]) using the waveform information. In step S530, energy changing unit 210 changes energy information Ep based on Ev of the volume designation information (E ← E + V).

  In step S540, the energy correction unit 220 corrects the energy information E changed by the energy change unit 210 (E ← f (E)). In step S550, amplification section 240 converts the corrected energy information generated by energy correction section 220 into a waveform amplification factor (G ← exp (E)). In step S560, amplification section 240 amplifies waveform S [i] using amplification factor G (S [i] ← S [i] × G).

  As described above, waveform generation is realized. Here, the order in which each operation is executed can be changed as long as the dependency relationship between the other operations is not impaired. For example, the order of step S520 and step S530 may be interchanged. Such a change in order can be easily understood, for example, by the block diagram shown in FIG.

  Note that the above-described operation is described as information processing by software stored in the waveform generation device 200 being realized using hardware included in the waveform generation device 200. Specifically, the software is a program for realizing the operation shown in each step, and is stored in a storage device included in the waveform generation device 200. The hardware includes a CPU (Central Processing Unit) or other arithmetic processing device that specifically implements a speech synthesizer including the waveform generation device 200, and a storage device that stores the program.

  However, the waveform generation can be realized not only by a combination of hardware and software, but also by a combination of only hardware such as circuit elements for realizing each operation.

  A speech synthesizer 600 using the waveform generation configuration according to the present embodiment will be described with reference to FIG. FIG. 6 is a block diagram showing a functional configuration of speech synthesizer 600. The speech synthesizer 600 includes a text analysis unit 610, a waveform dictionary storage unit 620, a volume setting unit 630, a prosody generation unit 640, a waveform data selection unit 650, a waveform generation unit 660, a waveform superposition unit 670, And an amplifying unit 680.

  The text analysis unit 610 is configured to receive input of text data. The text data is given to the speech synthesizer 600 when the user of the speech synthesizer 600 inputs a character string. Alternatively, when the speech synthesizer 600 can receive a signal including character information, the text data is given by the speech synthesizer 600 receiving the signal and acquiring the character information from the signal.

  The text analysis unit 610 analyzes the text data thus given (hereinafter referred to as input text), and outputs the reading of each word and accent information. In other aspects, text analysis unit 610 outputs part of speech information and the like. When the input text is a kanji-kana mixed sentence, the text analysis unit 610 generates each of the above information using a language dictionary (not shown). Alternatively, when the input text is a kana input or a phonetic symbol input such as an alphabet, the text analysis unit 610 generates each piece of information described above using accent information that is input simultaneously with the kana. For example, the text specifying the boundary between the accent position and the accent phrase is input at the same time, such as “Honjitsuha / Setena”.

  The waveform dictionary storage unit 620 stores segment data used for synthesizing speech. The edge data is given in advance by the manufacturer of the speech synthesizer 600. Or the structure which the user of the speech synthesizer 600 can input the waveform dictionary containing segment data may be sufficient. The data structure of the waveform dictionary storage unit 620 will be described later (FIG. 7).

  The volume setting unit 630 is connected to the text analysis unit 610 so as to be operable based on the output from the text analysis unit 610. The volume setting unit 630 sets the volume output from a device that uses the speech synthesizer 600 and outputs the set value as volume designation information Ev. Here, a device that uses the speech synthesizer 600 or a device that functions as the speech synthesizer 600 is realized as a PC (Personal Computer) having a speech output function, a mobile phone or other information communication terminal, or a game device. The volume setting unit 630 defines a volume output based on an operation on an operation unit (for example, a keyboard, a mouse, a numeric button, a volume adjustment dial, etc.) included in these devices.

  For example, when the volume setting unit 630 is realized by a volume adjustment dial, the output volume is set by the user of the speech synthesizer. The volume setting unit 630 sends a signal corresponding to the setting to the waveform generation apparatus 200 as volume designation information Ev. The waveform generation device 200 changes and corrects the energy information of all the sections according to the setting.

  Alternatively, when the volume designation information is attached to data input to these devices, the volume setting unit 630 sets the volume using the accompanying information and outputs the volume designation information Ev. To do. For example, when the speech synthesizer 600 is used as a game device, the game device drives a cartridge for realizing a game, and reads video data and audio data. In this case, if the volume designation information is associated with either video data or audio data, the volume setting unit 630 is output from the voice synthesizer 600 using the volume designation information acquired from the cartridge. Set as audio volume.

  In still another aspect, the volume designation information may be designated by an operation by a user of the device. For example, when the user operates the volume setting unit 630, an electrical signal output in response to the operation may be used as the volume designation information. Here, the operation includes rotation of a dial, pressing of a button, input of a setting value for defining a volume, and the like.

  The prosody generation unit 640 is connected to the text analysis unit 610 so as to be operable based on the output from the text analysis unit 610. The prosody generation unit 640 generates and outputs prosodic information based on accent information or sentence boundaries. The prosody information includes, for example, time length, pitch, energy (power) information, and the like. In general, prosodic information is obtained in units of phonemes, and then generated as information for each frame by interpolation.

  The waveform data selection unit 650 is connected to the text analysis unit 610 and the waveform dictionary storage unit 620 so as to operate based on the output from the text analysis unit 610 and the data stored in the waveform dictionary storage unit 620. Is done. The waveform data selection unit 650 selects segment data that matches the conditions for each phonetic symbol from the waveform dictionary storage unit 620 according to the phonetic symbol string set from the reading of each word. The waveform data selection unit 650 acquires the waveform information (FIG. 8) for each frame from the selected piece information and outputs it.

  The waveform generation unit 660 is configured so that the volume setting unit 630 and the prosody setting unit are operable based on the output from the volume setting unit 630, the output from the prosody generation unit 640, and the output from the waveform data selection unit 650. 640 and the waveform data selection unit 650 are connected. The waveform generation unit 660 corresponds to the waveform generation device 200 shown in FIG. 2 and realizes functions realized by the waveform generation device 200.

  Specifically, the sound volume designation information Ev, the energy information Ep, and the waveform information are input to the waveform generation unit 660. The waveform generation unit 660 generates and outputs waveform data for outputting sound whose volume is adjusted using these pieces of information.

  The waveform superimposing unit 670 is connected to the prosody generation unit 640 and the waveform generation unit 660 so as to operate based on the output from the prosody generation unit 640 and the output from the waveform generation unit 660. The waveform superimposing unit 670 receives input of data output from the prosody generation unit 640 in addition to data output from the waveform generation unit 660. Specifically, the waveform superimposing unit 670 receives input of waveform data, pitch information, and time length information. The waveform superimposing unit 670 superimposes waveform data extracted for each frame at a sample interval derived from pitch information corresponding to each frame, and outputs the result. For example, when the output sound is a voiced sound, the pitch interval corresponds to the fundamental frequency of the sound. When the voice is an unvoiced sound, the waveform data is superimposed at a fixed-length frame interval.

  The amplifying unit 680 is connected to the waveform superimposing unit 670 so as to be operable based on the output from the waveform superimposing unit 670. The amplifying unit 680 amplifies the waveform generated by the waveform superimposing unit 670 and outputs it as audio data. The audio data is converted into an analog signal by a digital / analog converter and sent to a speaker (not shown). The speaker outputs sound whose volume is adjusted based on the signal.

  Here, it should be noted that the output from the volume setting unit 630 is directed to the waveform generation unit 660, not the amplification unit 680. That is, in the conventional configuration, the volume designation information Ev (for example, an adjustment value for adjusting the volume) is input to the configuration corresponding to the amplification unit 680. In such a configuration, when the sound volume designation information Ev is changed by the user in order to suppress the distortion of the sound, the amplification factor is changed at the stage where the sound signal is output. Therefore, an audio signal that does not need to be adjusted for distortion (that is, a signal whose output level does not exceed the specified value of the speaker) is also subject to adjustment, and it is difficult to hear a sound with a low volume.

  However, in the speech synthesizer 600 according to the present embodiment, the volume designation information Ev is input to the waveform generation unit 660, as is apparent from FIG. Therefore, only the adjustment of audio data that needs to be adjusted to suppress the occurrence of distortion can be realized. As a result, the output level of the audio signal that originally does not require distortion adjustment is not adjusted, and the sound with a low volume is output as it is. As a result, the user of the speech synthesizer 600 can comfortably listen to both high volume sound and low volume sound.

  Note that the output mode of data after superposition by the waveform superimposing unit 670 is not limited to the above. For example, a digital amplifier may be used instead of the digital-analog converter.

  Here, the data structure of the speech synthesizer 600 will be described with reference to FIGS. FIG. 7 is a diagram conceptually showing one mode of data storage in waveform dictionary storage unit 620. The waveform dictionary storage unit 620 includes areas 710 to 790 for storing data. Each area stores data of a segment dictionary (segment data) that uses phonemes as basic units. For example, in the area 710, data “a” is stored as segment data. Each piece data has more detailed information.

  That is, referring to FIG. 8, region 790 includes a region 810 for storing a frame number, a region 820 for storing energy information, and a region 830 for storing waveform information. For example, the area 790 has energy information and waveform information for M frames.

  The generation of waveform information will be further described with reference to FIG. The speech segment dictionary is created by analyzing the original speech data for each basic unit. Generally, phonemes, syllables, VCVs (Vowel-Consonant-Vowel) are used as basic units. In FIG. 8, the case where phonemes are used as the basic unit is illustrated for the sake of simplicity of explanation, but the following description is also true for other basic units.

  First, the voice cut out in the basic unit is analyzed for each frame and decomposed into waveform information and energy information. As a process for acquiring the waveform information, when the voice is a voiced sound, generally, a process of removing a periodicity using a cepstrum analysis and extracting a one pitch waveform is performed. When the voice is an unvoiced sound, generally, a process for cutting out the original waveform into a frame length is performed. For the purpose of data compression of waveform information, the waveform information is decomposed into spectral information such as cepstrum and LSP (Linear Spectrum Pair) and residual waveform, and the spectrum information is either voiced or unvoiced. It may be broken down into flags. In this case, it is common to obtain a waveform data by obtaining a filter response to an impulse for voiced sound using a filter that represents a spectrum, and to obtain a waveform data by obtaining a filter response to random noise for an unvoiced sound. is there.

  It is desirable to normalize the waveform information so that it has a constant amplitude level in order to maintain accuracy. For example, when the analyzed waveform is X [i] (i = 1,..., N), the average amplitude Px is

It is represented by
The normalized waveform information S [i] is
S [i] = X [i] × A / Px
Normalize like this.

  Here, A is set to a value that makes the amplitude as large as possible without exceeding the amplitude limit range (for example, -32767 to 32768 for 16 bits). Specifically, a value such as A = 2048 is set.

  For energy information, the short-time average energy Px of each frame at the time of analysis is used as energy information. In addition, it is preferable to logarithmically use the energy information value rather than using the energy information value itself, because various controls such as changing the volume can be performed by “addition processing”.

  In the case of waveform generation in speech synthesis, decoding processing is not necessary if the speech unit dictionary remains waveform information. However, when the speech unit dictionary is compressed using the spectrum information, a general decoding process is performed. At this time, if the average amplitude of the decoded waveform is not A, gain adjustment is once performed so that the average amplitude A is obtained.

  At the time of speech synthesis, when target energy information E of a frame to be processed for synthesis (hereinafter referred to as a target frame) is equal to A in average amplitude, a waveform is generated with an amplification factor of 1. On the other hand, when the value of the target energy information E is not equal to A, a waveform with an average amplitude E is obtained by amplification with the amplification factor E / A.

  Moreover, the waveform generation in the waveform generation unit according to the embodiment of the present invention is more specifically as follows. The target energy E of the target frame is calculated by the formula “E = Ep × Ev” using energy information Ep based on prosody generation and Ev based on sound volume designation information. Here, the value of Ev is a relative value where the standard value is 1.

  Assuming that the normalized average amplitude A is A = 2048, the peak amplitude when the peak amplitude is the largest among the actual waveform information is assumed to be about 16384, which is eight times the average amplitude A, and the maximum of 16 bits. For the amplitude (−32768 to 32767), the peak amplitude can only be amplified up to twice the average amplitude A. Therefore, a restriction is imposed on the amplification factor for the generated waveform. This constraint is represented, for example, as E <2 × A.

  However, since the value of the energy information Ep based on the prosodic information is based on rules and statistics, it is necessary to adjust complex rules or modify the statistics to limit the energy information Ep. Therefore, in the embodiment of the present invention, in order to avoid these adjustments and corrections, f (E) is defined as a function of E, and the function f () is set so that f (E) <2 × A. Is used. As an example of the function f (E), a non-linear function 310 or a function 320 defined as a broken line as shown in FIG. 3 can be considered. Here, an example of the nonlinear function is as follows, for example.

f (E) = {log (E + 1) ^ (2 × A)} / log (4 × A)
This function is defined so that f (E) = 2 × A when E = (4 × A−1). The nonlinear function is not limited to this function, and various other functions can be considered.

An example of a function of a broken line is as follows, for example.
f (E) = E (if E <B, where B <2 × A)
f (E) = B + (EB) / 2 (when E ≧ B)
Here, various methods for setting the function of the polygonal line are conceivable and are not limited to those shown in the present embodiment.

  In another aspect, as another example, energy information may be defined as a logarithmic value. In this case, it is necessary to convert the value into an amplification factor. Generally, if energy information is input as a value of El = 20 log (E), the amplification factor G is expressed by the following equation. It is.

G = E / A = exp (El / 20) / A
The relational expression for calculating the amplification factor G is not limited to the above. A method may be used in which at least a value larger than El = 20 log (A / 2) corresponding to E = A / 2 is not suddenly increased by using a linear function connected continuously.

  By adjusting the amplification factor so as not to become extremely larger than a certain value in this way, the waveform generation apparatus 200 according to the embodiment of the present invention operates, so that distortion of the output waveform can be suppressed, and The volume can be set. As a result, it is possible to output at a volume larger than the conventional output volume using the same speaker.

  The data shown in FIGS. 7 and 8 is stored in the waveform dictionary storage unit 620 by the manufacturer of the speech synthesizer 600, but may be re-input by the user of the speech synthesizer 600.

  The waveform generation apparatus according to the present embodiment can also be applied to an apparatus that encodes and then decodes speech. Therefore, speech coding / decoding apparatus 900 having a waveform generation configuration according to the present embodiment will be described with reference to FIG. FIG. 9 is a block diagram showing a configuration of functions implemented by speech coding / decoding apparatus 900. The speech encoding / decoding device 900 includes an encoding device 910 and a decoding device 940. The decoding device 940 is connected to the encoding device 910 so as to be operable based on the output from the encoding device 910. The encoding device 910 includes a prosody analysis unit 920 and a waveform information analysis unit 930. Decoding apparatus 940 includes a volume setting unit 950, a waveform generation unit 960, a waveform superimposition unit 970, and an amplification unit 980.

  The decoding device 910 receives an input of an audio signal corresponding to the audio input to the audio encoding / decoding device 900. The speech is input, for example, via a microphone (not shown) provided in speech coding / decoding apparatus 900 or connected to speech coding / decoding apparatus 900. The audio signal is input to the prosody analysis unit 920 and the waveform information analysis unit 930, respectively.

  The prosody analysis unit 920 generates and outputs prosodic information of the input speech based on the accent information of the input speech, the speech interval, and the like. The waveform information analysis unit 930 analyzes the input audio signal, acquires waveform information of the audio signal, and outputs the waveform information. A signal output from the encoding device 910 is input to the decoding device 940. Specifically, information output from the prosody analysis unit 920 is input to the waveform generation unit 960 and the waveform superimposition unit 970. The waveform information output from the waveform information analysis unit 930 is input to the waveform generation unit 960. The volume setting unit 950 realizes the same function as the volume setting unit 630 shown in FIG.

  The waveform generation unit 960 and the prosody analysis unit 920 and the waveform information analysis so as to be operable based on the output from the prosody analysis unit 920, the output from the waveform information analysis unit 930, and the output from the volume setting unit 950. The unit 930 and the volume setting unit 950 are connected. The waveform generation unit 960 realizes the same function as that realized by the waveform generation apparatus 200 shown in FIG. That is, the waveform generation unit 960 decodes the waveform from the waveform information analyzed for each frame by the waveform information analysis unit 930, and outputs the energy information output by the prosody analysis unit 920 and the volume specification output by the volume setting unit 950. The waveform is amplified based on the information, and a waveform with an adjusted amplitude is output.

  The waveform superimposing unit 970 is connected to the prosody analysis unit 920 and the waveform generation unit 960 so as to operate based on the output from the waveform generation unit 960 and the output from the prosody analysis unit. The waveform superimposing unit 970 superimposes and outputs the waveform generated for each frame by the waveform generating unit 960 at the sample interval derived from the pitch information acquired by the prosody analyzing unit 920. A signal output from the waveform superimposing unit 970 is input to the amplifying unit 980.

  The amplifying unit 980 is connected to the waveform superimposing unit 970 so as to be operable based on the output from the waveform superimposing unit 970. Specifically, the amplifying unit 980 amplifies the waveform data output from the waveform superimposing unit 970 and outputs it as decoded audio data.

  With reference to FIG. 10, a waveform generation algorithm in another aspect of the present embodiment will be described. FIG. 10 is a flowchart representing an operation executed by waveform generation device 200 according to another aspect. The same steps as those shown in FIG. 5 are denoted by the same step numbers. Therefore, description thereof will not be repeated here.

  In step S1040, waveform generation apparatus 200 converts energy E changed by energy changing unit 210 into amplification factor G (G ← g (E)).

  Here, the concept of waveform amplification in the present embodiment will be described with reference to FIG. FIG. 11A to FIG. 11D are diagrams showing waveforms of three frames, respectively. FIG. 11A shows the original waveforms to be processed as frames 1111 to 1113. When the waveform is amplified twice for each frame, frames 1121 to 1123 are obtained as shown in FIG. Here, as is clear from the diagram representing the frame 1122, the amplitude exceeds the upper limit value and the lower limit value. In this case, if the range exceeding the upper limit value and the lower limit value is excluded from processing, the shape of each frame is the amplitude limit value (upper limit value and lower limit value) particularly in the frame 1132 as shown in FIG. Value). When sound is output based on such a signal, the sound related to the portion where the amplitude is saturated is distorted.

  On the other hand, when each frame shown in FIG. 11A is amplified using the waveform generation according to this embodiment, the amplified waveforms are shown as frames 1141 to 1143. As is clear from FIG. 11D, particularly in the frame 1142, the waveform is amplified within a predetermined range without exceeding the upper limit value or the lower limit value of the amplitude. Therefore, when a sound is output based on a signal having such a waveform, a sound with less distortion or no distortion is generated than when a sound is output based on the waveform shown in FIG. Will be output.

  Next, with reference to FIG. 12, a game apparatus 1200 capable of realizing the waveform generation function according to the present embodiment will be described. FIG. 12 is a block diagram showing a hardware configuration of game device 1200. The game device 1200 includes an operation button 1202, a data ROM 1204, a program ROM (Read Only Memory) 1206, a RAM (Random Access Memory) 1208, a CPU 1210, a digital-analog converter 1230, an amplifier 1240, a speaker 1250, A liquid crystal display 1260 and a card connector 1270. A game cartridge 1280 is attached to the card connector 1270. CPU 1210 includes an energy changing unit 1212, an energy correcting unit 1214, a decoding unit 1216, an amplifying unit 1218, and a waveform connecting unit 1220.

  The operation button 1202 receives an operation on the game device 1200 and outputs a signal corresponding to the operation to the CPU 1210. The data ROM 1204 stores control data created in advance for realizing the game apparatus 1200. The program ROM 1206 stores a program for executing processing predefined in the game device 1200. The RAM 1208 temporarily holds data generated during the operation of the game apparatus 1200 or data stored in the game cartridge 1280 read via the card connector 1270.

  The CPU 1210 operates on the operation buttons 1202, the data ROM 1204, the program ROM 1206, the RAM 1208, and the output signals from the card connector 1270, so that the CPU 1210 can operate. Connected to.

  The CPU 1210 uses the data stored in the game cartridge 1280 to execute processing for operating the game device 1200 as a device corresponding to the game cartridge 1280. Specifically, CPU 1210 reads video data 1282 and audio data 1284 stored in game cartridge 1280 and causes liquid crystal display 1260 to display manufacturing based on video data 1282. Further, the CPU 1210 uses the audio data 1284 to cause the speaker 1250 to output audio corresponding to the data. The voice data includes data obtained by analyzing voice generated by a person and text data generated by text-to-speech synthesis. Here, there is a case where the capacity of the speaker required for proper output of sound based on the sound data 1284 (output without causing distortion) does not match the capacity of the speaker 1250. For example, the output required by the audio data 1284 may exceed the output from the speaker 1250. In this case, the CPU 1210 performs the above-described waveform generation process using the audio data 1284, and executes the waveform generation process so that sound without distortion is output.

  Specifically, in the CPU 1210, the energy changing unit 1212 is configured to function based on data read from the RAM 1208. Specifically, the energy changing unit 1212 realizes the same function as the energy changing unit 210 shown in FIG.

  The energy correction unit 1214 is configured to function based on the output from the data ROM 1204. Specifically, the energy correction unit 1214 realizes the same function as the energy correction unit 220.

  The decoding unit 1216 realizes the same function as the waveform decoding unit 230. Specifically, the decoding unit 1216 decodes the waveform based on the data sent from the card connector 1270, that is, the audio data stored in the game cartridge 1280.

  The amplifying unit 1218 is configured to function based on outputs from the energy correcting unit 1214 and the decoding unit 1216. Specifically, the amplifying unit 1218 realizes the same function as the amplifying unit 240.

  The waveform connection unit 1220 is configured to function based on the output from the amplification unit 1218. Waveform data output from the amplifying unit 1218 is input to the waveform connecting unit 1220. This waveform data is configured in units of frames in the audio signal. Therefore, the waveform connecting unit 1220 connects each waveform data to generate audio data.

  The audio data generated by the CPU 1210 is input to the digital / analog converter 1230. The digital-analog converter 1230 converts the audio data into an analog audio signal and outputs it. The audio signal is input to the amplifier 1240.

  The amplifier 1240 amplifies and outputs the audio signal based on the signal output from the operation button 1202 as a signal for adjusting the volume. The amplified audio signal is input to the speaker 1250. The speaker 1250 converts the sound signal into sound and outputs the sound.

  With the configuration as described above, the game device 1200 displays an image on the liquid crystal display 1260 based on the image data 1282 for the game stored in the game cartridge 1280. At this time, game device 1200 can output a sound based on the sound data whose waveform is adjusted so that distortion does not occur in CPU 1210 from speaker 1250. Therefore, the user of game device 1200 does not need to operate a volume adjustment dial (not shown) that is a part of operation button 1202 to adjust the distortion by adjusting the volume via amplifier 1240. .

  In this way, even if the audio data 1284 created in advance includes data with a low output level at the speaker 1250 and data with a high output level that exceeds the rated output of the speaker 1250, the game apparatus 1200. Outputs sound based on data with a low output level as it is, and outputs data with a high output level at a volume adjusted to suppress distortion. Thereby, the interest by the game apparatus 1200 is not hindered.

  Even when various game cartridges having different output levels are mounted on the card connector 1270, the volume is adjusted by the processing in the CPU 1210, so the speaker 1250 is changed to a speaker having a higher rated output standard. There is no need. Therefore, it is not necessary to change the game device 1200 itself according to the content of the game (specifically, the output level of the audio data 1284 stored in the game cartridge 1280). This eliminates the need to change the design of the hardware of the game apparatus 1200 and can prevent an increase in the cost of the game apparatus 1200.

  Note that the speech synthesizer according to the present embodiment can also be applied to a communication terminal having a text data reception function. The communication terminal is realized as, for example, a mobile phone or a PDA (Personal Digital Assistant).

  A mobile phone 1300 that implements the waveform generation function according to the present embodiment will be described with reference to FIG. FIG. 13 is a block diagram showing a hardware configuration of mobile phone 1300. A cellular phone 1300 includes an antenna 1302, a communication circuit 1304, an operation button 1306, a camera 1308, a CPU 1310, a flash memory 1312, a RAM 1314, a data ROM 1316, an analog-digital converter 1322, a microphone 1320, a digital An analog converter 1324, a speaker 1323, a display 1330, an LED (Light Emitting Diode) 1332, a data communication I / F (Interface) 1334, a vibrator 1336, and a memory card driving device 1340 are provided. A memory card 1342 can be attached to the memory card driving device 1340.

  The antenna 1302 and the communication circuit 1304 are electrically connected. The CPU 1310 includes a communication circuit 1304, an operation button 1306, a camera 1308, a flash memory 1312, a RAM 1314, a data ROM 1316, a memory card driving device 1340, an analog / digital converter 1322, a digital / analog converter 1324, and a display. 1330, LED 1332, data communication I / F 1334, and vibrator 1336 are electrically connected.

  The radio wave received by the antenna 1302 is subjected to processing prescribed in advance by the communication circuit 1304 and then transmitted to the CPU 1310 as a digital signal. The radio waves include radio waves for calls and radio waves for data communication. The CPU 1310 internally processes the digital signal and transfers the processed signal to the digital / analog converter 1324.

  The digital-analog converter 1324 converts a digital signal output from the CPU 1310 into an analog signal and sends the analog signal to the speaker 1326. The speaker 1326 outputs voice (that is, a telephone that receives an incoming call) based on the analog signal.

  Microphone 1320 accepts an utterance to mobile phone 1300 and outputs an electrical signal corresponding to the utterance. The analog-digital converter 1322 performs a digital conversion process on the signal output from the microphone 1320 and sends it to the CPU 1310. CPU 1310 converts the signal into a signal for transmission and sends the signal to communication circuit 1304. Communication circuit 1304 wirelessly transmits the signal via antenna 1302. The user of the mobile phone 1300 can talk with other parties in this way.

  The operation button 1306 is realized as a button for accepting input of characters or numbers. Alternatively, in another aspect, as a configuration for receiving the input, a jog dial, a touch panel, or other operation unit may be realized instead of the operation button 1306. The operation button 1306 receives an operation on the mobile phone 1300 and sends a signal corresponding to the operation to the CPU 1310. The operation on the operation button 1306 includes an operation for the user of the mobile phone 1300 to input characters.

  The camera 1308 shoots a subject based on an operation on the operation button 1306, and sends a signal acquired by the shooting to the CPU 1310. The camera 1308 can shoot a subject as a still image or a moving image. The CPU 1310 temporarily holds the signal and stores it in the flash memory 1312 in response to a save instruction for the operation button 1306.

  The RAM 1314 temporarily holds data generated by the CPU 1310 based on an operation performed on the operation button 1306. Alternatively, the RAM 1314 temporarily holds data included in the radio wave received by the antenna 1302. The data ROM 1316 stores data or application programs for causing the mobile phone 1300 to perform operations specified in advance. The CPU 1310 reads the data or application program from the data ROM 1316 and causes the mobile phone 1300 to execute a predetermined process.

  Display 1330 displays an image or video defined by the data based on the data output from CPU 1310. For example, when the CPU 1310 reads moving image data stored in the flash memory 1312 or the memory card 1342, the display 1330 displays an image corresponding to the data.

  The LED 1332 realizes a light emitting operation defined in advance based on a signal output from the CPU 1310. For example, when the LED 1332 can display a plurality of colors, the LED 1332 emits light in a color associated with the data based on data included in a signal output from the CPU 1310.

  The data communication I / F 1334 accepts attachment of a communication cable from the outside. The data communication I / F 1334 sends a signal output from the CPU 1310 to the cable. Alternatively, the data communication I / F 1334 sends a signal received via a cable to the CPU 1310.

  Vibrator 1336 performs a vibration operation at a predetermined frequency based on a signal output from CPU 1310.

  The memory card driving device 1340 accepts the mounting of the memory card 1342. The memory card drive device 1340 reads data stored in the memory card 1342 and sends it to the CPU 1310. Conversely, the memory card driving device 1340 stores the data output by the CPU 1310 in the data storage area of the memory card 1342. The memory card 1342 uses, for example, a flash memory as a storage medium, but other cards may be used.

  In such a configuration, the CPU 1310 can function as the speech synthesizer 600 shown in FIG. That is, character information (for example, electronic mail) received by the communication circuit 1304 or a character string input by operating the operation button 1306 is input to the CPU 1310 as text data. In response to the input of the voice output instruction to the operation button 1306, the CPU 1310 synthesizes voice data from the text data. The mobile phone 1300 outputs a sound based on the sound data via the speaker 1326.

  In this case, the CPU 1310 functions as the text analysis unit 610, volume setting unit 630, prosody generation unit 640, waveform data selection unit 650, waveform generation unit 660, waveform superposition unit 670, and amplification unit 680 shown in FIG. The waveform dictionary storage unit 620 is realized by the flash memory 1312, for example. The data stored in the waveform dictionary storage unit 620 may be written at the time of manufacture by the manufacturer of the mobile phone 1300, or by the user of the mobile phone 1300 via the data communication I / F 1334 or the memory card driving device. It may be input via 1340. Software for executing specific processing for waveform generation is also stored in a storage device such as the flash memory 1312 and the data ROM 1316.

  With such a configuration, the mobile phone 1300 can also output voice using text data. As described with respect to the game device 1200, by using the waveform generation according to the embodiment of the present invention, it is possible to output a sound that is easy to hear regardless of the volume level without changing the capability of the speaker 1326. . Further, in the case where sound can be output through an earphone (not shown) in addition to the speaker 1326, the volume of sound output through the earphone can also be adjusted. As a result, for example, even when the vicinity of the user of the mobile phone 1300 is noisy, it is possible to output while suppressing the volume level of the sound set to a large volume from the beginning while increasing the small volume. As a result, a voice that is easy to hear is output.

  Note that the basic operation of the mobile phone 1300 can be easily understood by those skilled in the art. Therefore, detailed description thereof will not be repeated here.

  As described above, according to the waveform generation device 200 according to the present embodiment, the sound for suppressing the sound data while preventing the distortion of the sound data for outputting the large sound volume and for outputting the sound volume not too large. Data, that is, data that outputs a sound volume that does not exceed the output capability of the speaker, is increased to a level that does not exceed that capability. As a result, it is possible to output a sound whose volume is increased as a whole.

  Specifically, the speech synthesizer including the waveform generation device 200 performs correction by changing energy information calculated in advance when performing speech synthesis. Accordingly, it is not necessary to calculate the waveform energy as compared with the case where the output volume is adjusted using the data after the speech synthesis. Moreover, since the calculation of the amplification factor is easily realized, the addition of the processor is also suppressed. As a result, the response in the volume adjustment of the output sound becomes faster.

<Second Embodiment>
Hereinafter, a second embodiment of the present invention will be described. The waveform generation apparatus 1400 according to the present embodiment is different from the waveform generation apparatus 200 described above in that it has a function of deforming an output waveform so as to be gradually saturated.

  With reference to FIG. 14, a waveform generation apparatus 1400 according to the present embodiment will be described. FIG. 14 is a block diagram showing a configuration of functions realized by waveform generation device 1400 according to the present embodiment. The waveform generation device 1400 includes a saturation processing unit 1450 in addition to the configuration shown in FIG.

  The saturation processing unit 1450 is connected to the amplification unit 240 so as to be operable based on the output from the amplification unit 240. More specifically, the saturation processing unit 1450 receives an input of a waveform output from the amplification unit 240. The saturation processing unit 1450 deforms the waveform so as to be saturated nonlinearly with respect to the waveform. Specifically, the saturation processing unit 1450 uses a function in which the relationship between the amplitude of the input waveform and the output waveform is defined as a non-linear relationship, or equivalent map data. Correct the shape.

  Therefore, with reference to FIG. 15 and FIG. 16, the waveform amplitude saturation characteristic of the waveform generation device 1400 according to the present embodiment will be described. FIG. 15 is a diagram showing the relationship between the amplitude of the input waveform and the amplitude of the output waveform in waveform generating apparatus 200 shown in FIG. FIG. 16 is a diagram showing the relationship between the amplitude of the input waveform and the amplitude of the output waveform in waveform generating apparatus 1400 according to the present embodiment shown in FIG.

  Referring to FIG. 15, the amplitude of the input waveform (hereinafter, input amplitude) (Ain) is proportional to the amplitude of the output waveform (hereinafter, output amplitude) (Aout) in the range (Xmin−Xmax). There is a relationship. Here, since the output amplitude value can take only a value within the range (Ymin−Ymax), an amplitude smaller than the lower limit amplitude value Ymin and an amplitude exceeding the upper limit amplitude value Ymax are not output. Therefore, for example, even when the value of the input amplitude is larger than the upper limit value Xmax, the output amplitude value remains at Ymax, so that distortion may remain in the output audio.

  On the other hand, referring to FIG. 16, in waveform generation device 1400 according to the present embodiment, the output amplitude range maintains the same range as that described above, and the input amplitude range provides each output amplitude. Has been changed. That is, X (1) is defined as a value lower than the lower limit input amplitude Xmin, and the output amplitude is also linearly output up to the input amplitude X (2). Similarly, in the vicinity of the upper limit value of the input amplitude, the output amplitude is linearly output in the range of the input amplitude X (3) below the upper limit value Xmax and the value X (4) above the upper limit value Xmax. The characteristics have been corrected. As a result, the value of the output amplitude can be between a predetermined range (lower limit value Ymin to upper limit value Ymax) from X (1) to X (4). The input amplitude that can give a value exceeding the upper limit value Ymax of the output amplitude is limited to only a small region exceeding the upper limit value X (4). In this way, distortion does not occur abruptly with respect to the output of the speech, so that a larger amplification factor than that of the speech synthesizer having the waveform generation device 200 shown in FIG. 2 can be obtained.

  As described above in detail, according to the waveform generation apparatuses according to the first and second embodiments of the present invention, the sound waveform is adjusted before the output level of the sound data is amplified. This adjustment is realized by software that processes audio data. Therefore, the adjustment can be applied only to data that may exceed the upper limit value defined by the speaker, audio output terminal, or other audio output device standards. With such a configuration, level adjustment is performed on audio data that truly requires waveform adjustment (data whose volume level is too high), and adjustment on audio data that does not require adjustment is not performed.

  As a result, audio that does not exceed the upper limit of the output level regardless of what output level is defined for the audio data without changing the hardware configuration of the audio output device or the like, that is, audio that does not cause distortion Can be output.

  Note that the waveform generation device and the speech synthesizer according to the embodiment of the present invention can also be realized by a computer system having a speech output function. A computer system 1700 that functions as a speech synthesizer will be described with reference to FIG. FIG. 17 is a block diagram showing a hardware configuration of computer system 1700.

  The computer system 1700 includes, as hardware, a CPU 1710, a mouse 1720 and a keyboard 1730 that receive input of instructions from a user of the computer system 1700, data generated by execution of a program by the CPU 1710, or a mouse 1720 or a keyboard 1730. RAM 1740 for storing the input data in a volatile manner, hard disk 1750 for storing the data in a nonvolatile manner, CD (Compact Disk) -ROM drive 1760, and a sound card for generating and outputting an audio signal from the audio data 1770, a speaker 1772 that outputs sound based on a signal output from sound card 1770, a monitor 1780, and a communication IF 1790. Each component is connected to each other by a data bus. A CD-ROM 1762 is mounted on the CD-ROM drive 1760.

  Information processing in the computer system 1700 is realized by hardware and software executed by the CPU 1710. Such software may be stored in the hard disk 1750 in advance. The software may be stored in a CD-ROM 1762 or other storage medium and distributed as a program product. Alternatively, the software may be provided as a program product that can be downloaded by an information provider connected to the so-called Internet. Such software is read from the storage medium by the CD-ROM drive 1760 or other reading device, or downloaded via the communication IF 1790 and then temporarily stored in the hard disk 1750. The software is read from the hard disk 1750 by the CPU 1710 and stored in the RAM 1740 in the form of an executable program. CPU 1710 executes the program.

  Each hardware constituting the computer system 1700 shown in FIG. 17 is general. Therefore, it can be said that the essential part of the present invention is software stored in the RAM 1740, the hard disk 1750, the CD-ROM 1762 or other storage media, or software downloadable via a network. Since the operation of each hardware of computer system 1700 is well known, detailed description will not be repeated.

  The recording media are not limited to CD-ROM, FD, and hard disk, but are magnetic tape, cassette tape, optical disk (MO (Magnetic Optical Disc) / MD (Mini Disc) / DVD (Digital Versatile Disc)), IC (Integrated). Circuit (including memory card), optical card, mask ROM, EPROM (Erasable Programmable Read Only Memory), EEPROM (Electronically Erasable Programmable Read Only Memory), semiconductor memory such as flash ROM, etc. It may be a medium.

  The program here includes not only a program directly executable by the CPU but also a program in a source program format, a compressed program, an encrypted program, and the like.

  The embodiment disclosed this time should be considered as illustrative in all points and not restrictive. The scope of the present invention is defined by the terms of the claims, rather than the description above, and is intended to include any modifications within the scope and meaning equivalent to the terms of the claims.

  The present invention is also applicable to a device having a sound output interface such as a speaker, a sound output terminal, etc., such as a mobile phone or other communication device, a PC or other information processing device, or a game device.

It is a block diagram showing the structure of the function implement | achieved by the conventional waveform generation apparatus. It is a block diagram showing the structure of the function implement | achieved by the waveform generation apparatus which concerns on the 1st Embodiment of this invention. It is a figure showing the relationship between the energy information E before correction | amendment, and the energy information f (E) after correction | amendment. It is a figure showing the relationship between energy information and an amplification factor. It is a flowchart showing the operation | movement which the waveform generation apparatus which concerns on the 1st Embodiment of this invention performs. It is a block diagram showing the structure of the function implement | achieved by the speech synthesizer using the structure of the waveform generation which concerns on this Embodiment. FIG. 5 is a diagram (part 1) conceptually showing one aspect of data storage in a waveform dictionary storage unit. FIG. 5 is a diagram (part 1) conceptually showing one aspect of data storage in a waveform dictionary storage unit. It is a block diagram showing the structure of the function implement | achieved by the audio | voice coding / decoding apparatus which has the structure of the waveform generation which concerns on this Embodiment. It is a flowchart showing the operation | movement which the waveform generation apparatus according to the other situation of this Embodiment performs. It is a figure showing the waveform of a flame | frame. It is a block diagram showing the hardware constitutions of the game device which can implement | achieve the function of the waveform generation which concerns on embodiment of this invention. It is a block diagram showing the hardware constitutions of the mobile telephone which implement | achieves the function of the waveform generation which concerns on embodiment of this invention. It is a block diagram showing the structure of the function implement | achieved by the waveform generation apparatus which concerns on the 2nd Embodiment of this invention. It is a figure showing the relationship between the amplitude of the waveform input into the waveform generation apparatus which concerns on the 1st Embodiment of this invention, and the amplitude of the output waveform. It is a figure showing the relationship between the amplitude of the waveform input into the waveform generation apparatus which concerns on the 2nd Embodiment of this invention, and the amplitude of the output waveform. FIG. 11 is a block diagram showing a hardware configuration of a computer system 1700.

Explanation of symbols

  100, 200, 1400 Waveform generation device, 600 speech synthesis device, 900 speech coding / decoding device, 910 coding device, 940 decoding device, 1200 game device, 1210 CPU, 1280 game cartridge, 1302 antenna, 1342 memory card.

Claims (22)

A speech synthesizer that synthesizes and outputs speech,
Obtaining means for obtaining data for generating output sound and designation information for designating a volume of the output sound;
Between each waveform information for generating a waveform of each part obtained by dividing speech into predetermined units, each energy information of each said part, and between energy before and after correction to correct each energy information Storage means for storing correction data prepared according to a prescribed relationship ;
Changing means for changing each of the energy information to energy information corresponding to the specified information;
By applying each said energy information after change by pre Symbol changing means to said correction data, a correction means for deriving the energy information after correction,
Decoding means for decoding each waveform of the output speech based on each waveform information;
Amplifying means for amplifying each waveform decoded by the decoding means based on each energy information after the correction;
A speech synthesizer comprising: output means for outputting speech based on each waveform amplified by the amplification means.   The speech synthesizer according to claim 1, wherein the correcting unit corrects the energy information changed by the changing unit so as to gradually approach an upper limit value defined in advance for the energy information.   The speech synthesizer according to claim 2, wherein the correction unit corrects the energy information nonlinearly.   The speech synthesis apparatus according to claim 2, wherein the correction unit continuously corrects the energy information.   The speech synthesizer according to claim 1, wherein the correction data includes energy information before correction and energy information after correction of the energy information before correction. The output means includes
Waveform connecting means for connecting the waveforms amplified by the amplifying means to generate a combined waveform;
The speech synthesis apparatus according to claim 1, further comprising speech output means for outputting speech based on the synthesized waveform. The output means further includes waveform saturation means for adjusting the composite waveform output from the waveform connection means so as to gradually approach a predetermined upper limit value,
The speech synthesis apparatus according to claim 6, wherein the speech output unit outputs speech based on the synthesized waveform adjusted by the waveform saturation unit. The recording medium further includes a drive unit that is mounted with a removable recording medium and drives the recording medium, the recording medium storing audio data and the designation information associated with the audio data,
The speech synthesis apparatus according to claim 1, wherein the acquisition unit includes a reading unit that reads out the voice data and the designation information from the recording medium attached to the driving unit. The acquisition means includes
Receiving means for receiving a signal including character information;
The speech synthesis apparatus according to claim 1, further comprising an input unit that receives an input of the designation information. The acquisition means includes
A microphone that receives an utterance and outputs an audio signal corresponding to the utterance;
Waveform information analyzing means for analyzing the voice signal and outputting waveform information corresponding to the utterance;
The speech synthesis apparatus according to claim 1, further comprising: prosodic analysis means for outputting prosodic information including energy information corresponding to the utterance. A speech synthesizer that synthesizes and outputs speech,
Between each waveform information for generating a waveform of each part obtained by dividing speech into predetermined units, each energy information of each said part, and between energy before and after correction to correct each energy information Correction data prepared according to the prescribed relationship , memory for storing the program,
A processor for executing the program,
The processor is
An acquisition step of acquiring data for generating output sound and designation information for designating a volume of the output sound;
A change step of changing each of the energy information to energy information corresponding to the designated information;
By applying each said energy information after the change before Symbol changing step on the correction data, the correction deriving energy information after correction,
A decoding step for decoding each waveform of the output speech based on each waveform information;
Performing the amplification step of amplifying each waveform decoded by the decoding step based on each energy information after the correction ,
The speech synthesizer further includes an output unit that outputs speech based on each amplified waveform. The speech synthesizer according to claim 11, wherein the correcting step includes a step of correcting the energy information changed by the changing step so as to gradually approach an upper limit value defined in advance for the energy information. Wherein the correction step includes a step of correcting the energy information in a non-linear, the speech synthesis apparatus according to claim 12. Wherein the correction step includes a step of continuously correcting the energy information, the speech synthesis apparatus according to claim 12.   The speech synthesizer according to claim 11, wherein the correction data includes energy information before correction and energy information after correction of the energy information before correction. The processor further performs a waveform connection step of connecting the waveforms amplified by the amplification step to generate a composite waveform ,
The speech synthesis apparatus according to claim 11, wherein the output unit outputs speech based on the synthesized waveform. The processor further executes a waveform saturation step for adjusting the combined waveform generated in the waveform connection step so as to approach the predetermined upper limit value ,
The speech synthesis apparatus according to claim 16, wherein the output unit outputs speech based on the synthesized waveform after adjustment in the waveform saturation step. The speech synthesizer further includes a drive device that is mounted with a detachable recording medium and drives the recording medium, and the recording medium stores audio data and the designation information associated with the audio data. And
The speech synthesizing device according to claim 11, wherein the obtaining step includes a reading step of reading out the speech data and the designation information from the recording medium mounted on the driving device. The obtaining step includes
A receiving step for receiving a signal including character information;
The speech synthesizer according to claim 11, further comprising an input step of receiving input of the designation information. The speech synthesizer further includes a microphone that receives an utterance and outputs an audio signal corresponding to the utterance,
The obtaining step includes
Analyzing the audio signal and outputting waveform information corresponding to the utterance;
The speech synthesis apparatus according to claim 11, further comprising: outputting prosodic information including energy information corresponding to the utterance. A speech synthesis method for synthesizing and outputting speech,
Obtaining data for generating output sound and designation information for designating a volume of the output sound; and
Between each waveform information for generating a waveform of each part obtained by dividing speech into predetermined units, each energy information of each said part, and between energy before and after correction to correct each energy information Loading correction data prepared according to a defined relationship ;
Changing each of the energy information to energy information corresponding to the designation information;
By applying each said energy information after the change in the step of pre-Symbol changed to the correction data, deriving energy information corrected,
Decoding each waveform of the output speech based on each waveform information;
Amplifying each waveform decoded by the decoding step based on each energy information after the correction;
Outputting speech based on each waveform amplified by the amplifying step. A program for causing a computer comprising a memory and a processor to implement a speech synthesis method, wherein the program is
Before Symbol memory, and data for generating a sound output, a step of acquiring a designation information for designating the volume of sound to be the output,
Before Symbol memory, and each waveform information for generating a waveform of each part obtained by dividing a unit that is pre-defined speech, and energy information of each of the said portions, before and after correction for correcting each of said energy information Reading correction data prepared according to a predefined relationship between the energies of
Each said energy information, and changing the respective energy information corresponding to the designated information,
By applying each said energy information after the change at the next on the correction data, the correction deriving energy information after correction,
Decoding each waveform of the output speech based on each waveform information;
Based on each said energy information after previous SL correction, the step of amplifying each waveform decoded by said step of decoding,
Based on each of said waveforms amplified, and a step of outputting an audio signal, the program.

Images