JP4844623B2 - CHORAL SYNTHESIS DEVICE, CHORAL SYNTHESIS METHOD, AND PROGRAM - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To synthesize chorus voice capable of giving audience natural impression. <P>SOLUTION: A chorus synthesizer 100 includes a voice sample data base 110 for storing three voice sample data groups 110a, 110b and 110c which are respectively created based on different voice, and three song generation units 120, 121 and 122. When a chorus voice signal of a music piece composed of three parts is synthesized, the song generation units 120, 121 and 122 generate song voice signals according to lyrics information and melody information for each part, and synthesize each chorus voice signal, under control of a chorus control section 140. When the signal is generated, each of the song generation units 120, 121 and 122 uses a phoneme sample data included in the voice sample data groups 110a, 110b and 110c. <P>COPYRIGHT: (C)2009,JPO&amp;INPIT

Description

  The present invention relates to a chorus synthesizer that synthesizes a choral sound signal, a choral synthesis method, and a program for synthesizing a choral sound.

  Conventionally, a chorus synthesizer that synthesizes a singing sound signal and produces a singing voice based on lyrics information and melody information has been proposed. As devices for synthesizing a singing sound signal in this way, various devices such as a device applying a regular speech synthesis technique have been proposed. In a singing voice synthesizing apparatus to which a rule synthesis technique is applied, voice sample data in units of phonemes or phoneme chains including a plurality of phonemes is created from voices uttered by a speaker in advance and stored in a database. The voice signal data such as phonemes required according to the lyric information is read and connected to synthesize a singing sound signal.

  By the way, in the singing sound synthesizing apparatus that synthesizes the singing sound as described above, unlike a speech synthesizing apparatus such as a text-to-speech apparatus, a usage form of electronically outputting the singing sound at the time of chorusing such as singing or singing can be considered. Therefore, development of a chorus synthesizer having a function of synthesizing a singing sound (choral sound) at the time of chorus is also being performed.

  A choir synthesizing apparatus having a function of synthesizing a chorus sound signal at the time of chorus generates a chorus sound signal by reading and connecting audio sample data based on each of the plurality of parts. Then, by superimposing and outputting the singing sound signals generated for each part, the chorus sound can be electronically output.

  However, in the choral synthesizer having the function of synthesizing the conventional choral sound signal, the same voice sample data is used when generating the singing sound signal according to the lyric information and the melody information for each part. Although the singing sound generated for each part has a different melody, fine features (such as pitch fluctuation) of the generated speech waveform for each part are basically the same. Therefore, the chorus sound obtained by superimposing these sounds as an unnatural chorus sound for the listener. This is considered that the listener listens to the correlation between the parts (the fine features match), giving an unnatural impression.

  In addition, when synthesizing the chorus sound signal at the time of singing, the method of simply generating and superimposing the singing sound signal for each part as described above, the same singing sound is superimposed and output, As a result, an unnatural impression is given to the listener. Therefore, in the conventional chorus sound synthesizer, when synthesizing the chorus sound signal at the time of singing, the sound generation timing of the singing sound generated for each part (the content is the same) is slightly shifted or generated for each part. By slightly shifting the pitch of the singing sound, the same singing sound was prevented from being overlaid and pronounced. However, even when the sound generation timing and pitch are slightly shifted, the fine features (fluctuations and the like) of the speech waveform generated for each part as described above are basically the same. Therefore, the chorus sound obtained by superimposing these sounds like a chorus sound unnatural for the listener.

  Japanese Patent Application Laid-Open No. 7-146695 discloses an apparatus for generating a chorus sound signal. In this apparatus, when generating a singing sound signal for each part, the pitch fluctuations that differ for each part are disclosed. A singing sound signal with components added is generated. Thus, the correlation between each part can be made small by overlapping and outputting the singing sound signal which provided the fluctuation component of a different pitch for every part. However, in the apparatus described in this publication, the pitch component added to the singing sound signal for each part is not based on human voice, but is artificially created. Although the correlation between the two becomes small, the synthesized chorus may be heard unnaturally.

  The present invention has been made in consideration of the above circumstances, and provides a choral synthesizer, a choral synthesis method, and a program capable of synthesizing a choral sound that can give a listener a more natural impression. For the purpose.

  In order to solve the above-mentioned problems, a choral synthesizer according to the present invention is a choral synthesizer that synthesizes a choral sound signal according to music data, and includes voice sample data having a predetermined length of time created based on voice. A database to be stored; means for generating a singing sound signal according to the music data; and a plurality of singing sound generating means used for generating the singing sound signal by reading out the required voice sample data from the database; Corresponding to different times of the voice sample data stored in the database for each of the plurality of song generation means and the song synthesis means for synthesizing the chorus sound signal from the song sound signals generated by the plurality of song generation means Designating means for designating each part to be performed, the music data is composed of a plurality of parts, each of the plurality of song generation means When generating the singing sound signal corresponding to the part, each of the at least two singing generation means starts using the portion specified by the specifying means of the audio sample data read from the database, and A singing sound signal is generated, and the designation means designates each of the plurality of singing generation means when the voice sample data read from the database is voice sample data for a phoneme segment that is a phoneme chain. The use start time is concentrated on the first half of the audio sample data.

  According to this structure, when generating the song sound signal of the part to which each song generating unit corresponds, at least two song generating units start using the portions corresponding to different times of the voice sample data and generate them. It will be. Here, in the voice sample data having a certain length of time created based on the voice, the fine features (fluctuation of the voice waveform) are not constant during the time length, and the fine features and the like vary with time. For this reason, the singing sound signals output from the at least two singing generation means have different fine features. Therefore, since the singing sound having a unique characteristic is emitted as the singing sound corresponding to each part, a more natural impression can be given to the listener.

  Moreover, the choral synthesis method according to the present invention is a choral synthesis method for synthesizing a choral sound signal from a plurality of singing sound signals generated based on music data, and the plurality of the choral synthesizing methods according to the music data composed of a plurality of parts. When generating the singing sound signal corresponding to the part, the required voice sample data is read out from the database storing the voice sample data having a predetermined time length created based on the voice, and at least two of the voice sample data are read out. For the generation of the singing sound signal corresponding to the part, the voice sample data read from the database is generated by starting use from a part corresponding to a different time of the voice sample data read from the database and generating the singing sound signal. In the case of speech sample data for a speech unit that is a phoneme chain, the use start time of the speech sample data is set as the sound sample data. It is characterized by be concentrated on the first half of the sample data.

  In addition, the program according to the present invention reads out the required voice sample data from a database that stores voice sample data having a predetermined length of time created based on voice according to music data, and singing sound. A means for generating a signal, wherein the music data is composed of a plurality of parts, and when generating a singing sound signal corresponding to the plurality of parts, a singing sound signal corresponding to at least two of the parts is generated. In this case, the singing sound generating means for generating the singing sound signal by starting use from the part corresponding to the different time of the voice sample data read from the database, and the chorus sound signal from the generated singing sound signal. While functioning as a singing voice synthesizing means, the singing sound generation means reads the voice sample data read from the database. Data is in the case of speech sample data for voice segment is phoneme is characterized in that the functioning of the use start time of the audio sample data as a means to concentrate the first half of the audio sample data.

  According to the present invention, it is possible to synthesize a chorus sound that can give a listener a more natural impression.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
A. First embodiment A-1. Basic Configuration of First Embodiment First, FIG. 1 is a block diagram showing the basic configuration of a choral synthesizer according to the first embodiment of the present invention. As shown in the figure, the chorus synthesizer 100 includes a voice sample database 110, a plurality (three in the illustrated example) of song generators 120, 121, 122, a chorus control unit 140, and a song generator 120. , 121, and 122, adder 130 that adds and synthesizes and outputs the singing sound signals.

  The audio sample database 110 stores audio sample data created based on natural sounds uttered by a person. The speech sample database 110 stores speech sample data (hereinafter referred to as speech segment sample data) having a phoneme chain composed of a single phoneme or a plurality of phonemes as one unit.

  A large number of short-time long voice sample data is accumulated in a database, and these voice sample data are connected according to the lyrics etc., and in the voice synthesis processing technology, phoneme is basically used as a synthesis unit. . For this reason, speech unit sample data of only phoneme (about 30 to 50 types) units may be stored in the speech sample database 110 in the choral synthesizer 100, but the rules for combining phonemes are complex. For this reason, it is difficult to obtain good quality when speech sample data of only phoneme units is accumulated. Therefore, it is preferable that the speech sample database 110 stores speech unit sample data of a unit (phoneme chain) slightly larger than the phoneme in addition to the speech unit sample data of only the phoneme unit. Units larger than phonemes include units such as CV (consonant → vowel), VC (vowel → consonant), VCV (vowel → consonant → vowel), and CVC (consonant → vowel → consonant). It is conceivable to store all of the speech unit sample data of these units, but in the chorus synthesizer 100 that synthesizes the choral sound, one unit is an extended sound that is pronounced for a long time, such as a vowel frequently used in singing. Speech unit sample data, speech unit sample data with consonant to vowel (CV) and vowel to consonant (VC) as one unit, speech unit sample data with consonant to consonant as one unit, and vowel to vowel It is only necessary to store speech segment sample data with 1 as a unit.

  The speech sample database 110 stores speech segment sample data in units of phonemes or phoneme chains as described above. In the speech sample database 110, the same type of phonemes (for example, “a”) or Three speech segment sample data are stored for a phoneme chain (for example, “ai”). In other words, the speech sample database 110 stores three speech sample data groups 110a, 110b, and 110c composed of a predetermined number of unit speech segment sample data with a phoneme or phoneme chain as one unit.

  The three audio sample data groups 110a, 110b, and 110c stored in the audio sample database 110 are data created based on different sounds. Here, different voices do not only mean that the speakers are different, but even the same speaker may use voices uttered at different occasions or different utterance parts. As described above, the voice sample data groups 110a, 110b, and 110c are created based on voices uttered at different occasions or different voice parts even if they are different voicers or the same voicers. As described above, the data about the same phoneme (or phoneme chain) included in each of the audio sample data groups 110a, 110b, and 110c is different because the sound used as a basis for creating each data is different. Features (pitch fluctuations, etc.) are different.

  The voice sample database 110 stores the three voice sample data groups 110a, 110b, and 110c as described above, and each of the song generators 120, 121, and 122 generates the voice sample when generating the song signal. The speech segment sample data is read from the database 110 and used.

  Each of the singing generators 120, 121, and 122 reads out the required speech segment sample data from the speech sample database 110 according to the music information having the lyric information and the melody information, and uses the read speech segment sample data. A singing sound signal is generated.

  More specifically, each of the singing generators 120, 121, and 122 obtains a phoneme string according to the lyric information, determines speech segment sample data necessary to construct the phoneme string, and the speech sample database 110. Read from. Then, the read speech segment sample data is connected in time series, and the connected speech segment sample data is appropriately adjusted according to the pitch according to the melody information to generate a singing sound signal.

  The chorus synthesizer 100 according to the present embodiment includes three song generators 120, 121, and 122 that can generate a song sound signal according to lyrics information and melody information, and thus includes three parts. According to the music information (lyric information and melody information) of the choral music, it is possible to synthesize a chorus sound signal corresponding to the choral music.

  When the chorus control unit 140 synthesizes the chorus sound signal corresponding to the chorus sound based on the song information of the chorus music in the choir synthesizing apparatus 100, the song information is divided for each part and each song generator 120 is divided. , 121, 122. Thereby, when synthesizing a chorus sound signal according to music information consisting of three parts, each song generator 120, 121, 122 uses the lyrics information and melody information of each part supplied from the chorus control unit 140. Therefore, a singing sound signal is generated, and the singing sound signal generated by each of the singing generators 120, 121, 122 is output to the adder 130. Thereby, according to the music information of the choral music which consists of three parts from the adder 130, the chorus sound signal corresponding to this choral music can be synthesize | combined.

  In addition, when the chorus synthesizer 100 synthesizes the chorus sound signal as described above, the chorus control unit 140 stores each of the song generators 120, 121, and 122 in the voice sample database 110. Of the voice sample data groups 110a, 110b, and 110c, the designation information for designating which voice sample data is read out and used is output to the song generators 120, 121, and 122. Here, the chorus control unit 140 generates the singing sound signal by using the speech segment sample data included in the speech sample data groups 110a, 110b, and 110c different from each other. Designation information for designating different data groups is output to each song generator 120, 121, 122.

  Specifically, for the song generator 120, designation information for designating the voice sample data group 110a is output, for the song generator 121, designation information for designating the voice sample data group 110b is output, For example, designation information for designating the voice sample data group 110c is output to the song generator 122. When such designation information is supplied from the chorus control unit 140, the singing generator 120 reads out the voice segment sample data included in the voice sample data group 110a and uses it to generate a singing sound signal. The speech unit sample data included in the speech sample data group 110b is read out and used to generate a singing sound signal. The singing generator 122 generates a singing sound signal using the speech unit sample data included in the speech sample data group 110c. Will do.

  When the chorus synthesizer 100 synthesizes the singing sound signal of the chorus composed of three parts, as described above, the singing generators 120, 121, 122 are included in the audio sample data groups 110a, 110b, 110c different from each other. By using the speech segment sample data, it is possible to synthesize a chorus sound signal that can give the listener a more natural impression. That is, the voice sample data groups 110a, 110b, 110c stored in the voice sample database 110 are created based on different voices, and even if they are data on the same type of phonemes and phoneme chains, The fine features (pitch fluctuations, etc.) of the voice shown in the data included in each of the voice sample data groups 110a, 110b, 110c are different. Of the speech sample data groups 110a, 110b, and 110c including the speech segment sample data having different fine features as described above, the singing generators 120, 121, and 122 are included in different speech sample data groups. By generating the singing sound signal using the piece sample data, the singing sound signals generated by the singing generators 120, 121, and 122 have different fine features. Therefore, the chorus sound obtained by superimposing these has a unique characteristic with little correlation between the parts, and it is possible to reduce giving an unnatural impression to the listener.

  The speech segment sample data included in the speech sample data groups 110a, 110b, and 110c stored in the speech sample database 110 is data created based on speech uttered by each person. Therefore, the difference in the fine features of the speech shown in the speech segment sample data included in each speech sample data group is not artificially created to give a pitch fluctuation prepared in advance. Therefore, it can reduce that the synthesized choral sound becomes unnatural.

A-2. Specific configuration of the choral synthesizer What has been described above is the basic configuration of the choral synthesizer 100 according to the present embodiment. In this chorus synthesizer 100, as the song generators 120, 121, and 122, the speech segment sample data is read from the speech sample database 110 according to the lyrics information and connected, and the speech connected according to the pitch according to the melody information. If it is a singing generator that adjusts the segment sample data and outputs a singing sound signal, various known singing generators such as a singing generator applying a regular speech synthesis technology, etc. can be used. The database 110 may store speech segment sample data corresponding to the song generator to be employed. In the following, a singing generator using a spectral modeling synthesis (SMS) technique proposed in US Pat. Nos. 5,029,509 and 2,906,970 is applied as the singing generator 120, 121, 122. As an example, the chorus synthesizer 100 will be specifically described.

  First, a method for creating the voice sample database 110 in the singing voice synthesizing apparatus 100 including the singing generators 120, 121, and 122 using the SMS technology will be described.

  As described above, the speech sample database 110 in the choir synthesizer 100 stores speech segment sample data created based on speech uttered by a speaker. The SMS technology is a technology for analyzing and synthesizing musical sounds using a model that represents an original sound with two components, ie, a harmonic component and a stochastic component. In the speech synthesis, data consisting of the harmonic component and the anharmonic component is used for speech synthesis as one unit speech unit sample data such as a phoneme or phoneme chain. Therefore, in the choral synthesizer 100 using the SMS technology, the speech sample database 110 includes data indicating harmonic components and inharmonic components obtained by performing SMS analysis on speech uttered by a speaker. Stored as sample data. Hereinafter, a method for creating the voice sample database 110 will be described with reference to FIG.

  As shown in the figure, the voice uttered by the speaker to create the voice sample database 110 is input to the SMS analysis unit 200 and subjected to SMS analysis in the SMS analysis unit 200. Here, since it is necessary to store voice sample data groups 110a, 110b, and 110c created based on different voices in the voice sample database 110, three different voices are input to the SMS analysis unit 200. become. In the drawing, three different sounds are shown to be input to the SMS analysis unit 200 in parallel. However, SMS analysis for each sound need not be performed in parallel at the same time. Also good.

  The SMS analysis unit 200 performs SMS analysis on the input voice and outputs SMS analysis data for each frame. More specifically, SMS analysis data for each frame is output by the following method.

  First, the input voice is divided into a series of frames. Here, the frame period used for the SMS analysis may be a fixed length, or may be a variable length period in which the period is changed according to the pitch of the input speech.

  Next, frequency analysis such as Fast Fourier Transform (FFT) is performed on the voice divided into frames. An amplitude spectrum and a phase spectrum are obtained from the frequency spectrum (complex spectrum) obtained by this frequency analysis, and a spectrum of a specific frequency corresponding to the peak of the amplitude spectrum is extracted as a line spectrum. At this time, a spectrum having a frequency in the vicinity of the fundamental frequency and an integral multiple of the fundamental frequency is defined as a line spectrum. The line spectrum extracted in this way corresponds to the harmonic component described above.

  Next, a line spectrum is extracted from the input speech as described above, and a residual spectrum is obtained by subtracting the extracted line spectrum from the input speech (waveform after FFT) of the frame. Alternatively, the time waveform data of the harmonic component synthesized from the extracted line spectrum is subtracted from the input speech waveform data of the frame to obtain the time waveform data of the residual component, and then subjected to frequency analysis such as FFT. Thus, a residual spectrum may be obtained. The residual spectrum obtained in this way corresponds to the above-described anharmonic component.

  The SMS analysis unit 200 outputs SMS analysis data for each frame including the line spectrum (harmonic component) and the residual spectrum (non-harmonic component) acquired as described above to the section cutout unit 201.

  The segment cutout unit 201 corresponds to the length of one unit (phoneme or phoneme chain) of the speech segment sample data to be stored in the speech sample database 110 for the SMS analysis data for each frame supplied from the SMS analysis unit 200. Cut out as you do. The section cutout unit 201 cuts out SMS analysis data so as to correspond to the unit length of each segment and stores it in the voice sample database 110.

  Here, the speech segment sample data stored in the speech sample database 110 is SMS data cut out for each phoneme or phoneme chain. For harmonic components, the spectral envelopes of all frames included in the phoneme or phoneme chain are included. (Line spectrum (overtone series) intensity (amplitude) and phase spectrum) are stored. Note that such a spectral envelope itself may be stored as a harmonic component, but the spectral envelope expressed by some function may be stored, or the harmonic component may be obtained by inverse FFT or the like. Alternatively, it may be stored as a time waveform. In the present embodiment, the anharmonic component is also stored as an intensity spectrum and a phase spectrum as in the harmonic component, but may be stored as a function or a time waveform as with the harmonic component.

  SMS analysis and segmentation for such speech are performed for each of three different input speech, and as a result, speech segment sample data created based on three different speech such as speech sample data groups 110a, 110b, and 110c. A group of (SMS analysis data for each phoneme or phoneme chain) is stored in the speech sample database 110.

  The above is the detail of the production method of the audio sample database 110 of the choral synthesizer 100 according to the present embodiment.

  Next, each singing generator 120, 121, which generates a singing sound signal using the voice sample database 110 storing the three voice sample data groups 110a, 110b, 110c created based on different voices as described above. 122 will be described. Since the song generators 120, 121, and 122 have the same configuration, the configuration of the song generator 120 will be described below with reference to FIG. 3 and the other song generators 121 and 122 will be described. Omit.

  As shown in the figure, the singing generator 120 includes a speech unit selection unit 301, a pitch determination unit 302, a duration length adjustment unit 303, a speech unit connection unit 304, and a harmonic component generation unit 305. , An addition unit 306, an inverse FFT (fast Fourier transform) unit 307, a windowing unit 308, and an overlap unit 309.

  The speech segment selection unit 301 reads necessary speech segment sample data from the speech sample database 110 in accordance with the lyrics information and designation information supplied from the chorus control unit 140 (see FIG. 1). More specifically, the supplied lyric information is converted into a phonetic symbol (phoneme or phoneme chain) sequence, and speech segment sample data is read from the speech sample database 110 according to the converted phonetic symbol sequence. For example, when a singing sound signal is generated according to lyric information such as “saita”, the lyric information is “#s”, “s”, “sa”, “a”, “ai”, “i”. ”,“ It ”,“ t ”,“ ta ”,“ a ”,“ a # ”and the like, and speech segment sample data corresponding to each of these phonetic symbols is read from the speech sample database 110. Will be.

  The speech segment selection unit 301 determines speech segment sample data to be read according to the lyrics information as described above, and is determined from the speech sample data group designated by the designation information supplied from the chorus control unit 140. Read speech segment sample data. For example, when the designation information designates the audio sample data group 110a, “#s”, “s”, “sa”, “a”, “a”, “included in the audio sample data group 110a of the audio sample database 110”. The speech segment sample data corresponding to “ai”, “i”, “it”, “t”, “ta”, “a”, “a #” is read out.

  The pitch determination unit 302 determines the pitch of the singing sound according to the melody information supplied from the chorus control unit 140 (see FIG. 1), and outputs the pitch information indicating the determined pitch to the harmonic component generation unit 305.

  The speech unit sample data (harmonic component and anharmonic component) read out by the speech unit selection unit 301 is supplied to the duration time adjustment unit 303. Here, the speech unit selection unit 301 may supply the read speech unit sample data as it is to the duration adjustment unit 303, but an appropriate correction process is performed according to the pitch or the like indicated by the melody information. May be supplied to the duration time adjustment unit 303.

  The duration time adjustment unit 303 changes the time length of each speech unit sample data supplied from the speech unit selection unit 301 according to the phoneme or phoneme chain duration determined by melody information or the like. I do. More specifically, when using certain speech unit sample data as a time shorter than the time length, a process of thinning out frames from the speech unit sample data is performed. On the other hand, when a certain speech segment sample data is continuously used for a time longer than the time length, a loop process for repeating the speech segment sample data for the length of time for which the speech segment sample data is used is performed. In this loop processing, when repeating certain speech segment sample data, the speech segment sample data j is first (0) after the data from the beginning to the end (0 to t) of the speech segment sample data. From the beginning to the end (0 to t), after the data from the beginning to the end (0 to t), the speech segment sample data is moved from the last (t) to the first part in time. The data may be connected and repeated.

  The duration adjustment unit 303 adjusts the duration of the speech unit sample data (harmonic component and anharmonic component) according to the sound duration of each speech unit as described above, and then adjusts the speech after the time adjustment. The segment sample data is output to the speech segment connection unit 304.

  The speech unit connection unit 304 connects the harmonic component data of the speech unit sample data supplied from the duration length adjustment unit 303 in time series, and connects the inharmonic component data in time series. In such connection, if the difference in the shape of the spectrum envelope between the two harmonic components to be connected is large, smoothing processing or the like may be performed. The speech element connection unit 304 outputs the connected harmonic component data to the harmonic component generation unit 305 and outputs the connected anharmonic component data to the addition unit 306.

  The harmonic component generation unit 305 is supplied with harmonic component data (spectrum envelope information) from the speech unit connection unit 304 and is supplied with pitch information according to melody information from the pitch determination unit 302. The harmonic component generation unit 305 generates a harmonic component corresponding to the pitch information from the pitch determination unit 302 while maintaining the spectrum envelope shape indicated by the spectrum envelope information from the speech unit connection unit 304.

  The adding unit 306 is supplied with the anharmonic component data from the speech unit connecting unit 304 and the harmonic component data from the harmonic component generating unit 305, and the adding unit 306 combines both of them to the inverse FFT unit 307. Output. The inverse FFT unit 307 converts the added frequency domain signal supplied from the addition unit 306 into a time domain waveform signal by performing inverse FFT, and outputs the converted waveform signal to the windowing unit 308. To do. The windowing unit 308 multiplies the time domain waveform signal by a window function corresponding to the frame length, and the overlap unit 309 generates a singing sound signal while overlapping the multiplied waveform signals. In this way, the singing sound signal is generated in the singing sound generator 120 in accordance with the lyric information and the melody information of the part having the music information supplied from the chorus control unit 140 (see FIG. 1). Is output to the adder 130 (see FIG. 1).

  The above is the detailed configuration of the song generator 120, and the music supplied from the chorus control unit 140 as described above from the other song generators 121 and 122 (same configuration as the song generator 120) shown in FIG. 1. A singing sound signal generated according to the lyrics information and melody information of the part with information is output. Here, as described above, when the singing generators 120, 121, 122 generate the singing sound signals corresponding to the parts distributed from the chorus control unit 140, the different voice sample data groups 110a, 110b, 110c, respectively. Since the voice segment sample data is read out from the voice and used for generation, the fine features (pitch fluctuations, etc.) of the singing sound signal generated by each of them are different.

  The adder 130 synthesizes and outputs the singing sound signals generated by the singing generators 120, 121, and 122 according to the parts of the music information of the choral music. A chorus sound signal obtained by synthesizing the singing sound signals of the three parts output from the adder 130 is converted into an analog voice waveform signal by a D / A (Digital to Analog) converter (not shown), and then an amplifier or the like. Through the speaker. Thereby, the listener can listen to the chorus sound according to the music information of the chorus composed of a plurality of parts. The chorus sound emitted from the chorus synthesizer 100 is different in the fine features of the singing sounds of each part (voice quality resulting from differences in pitch fluctuations, etc.) and gives a natural impression to the listener. It is possible to produce a choral sound that can be played.

B. Second Embodiment Next, a chorus synthesizer according to a second embodiment of the present invention will be described with reference to FIG. As shown in the figure, the voice sample database 110 of the chorus synthesizer 100 in the first embodiment stores three voice sample data groups 110a, 110b, and 110c, whereas the chorus according to the second embodiment. The speech sample database 110 in the synthesizer 400 is different in that only one type of speech segment sample data is stored for the same phoneme or phoneme chain. The chorus synthesizer 400 according to the second embodiment uses the speech sample database 110 that stores only one speech segment sample data for one phoneme or phoneme chain as described above, and thus is more natural than the first embodiment. It is possible to synthesize a chorus sound signal that can give a unique impression. Hereinafter, the configuration of the choral synthesizer 400 will be described focusing on differences from the choral synthesizer 100 according to the first embodiment.

  Each of the singing generators 120, 121, and 122 in the chorus synthesizer 400 is the same as that in the first embodiment, and a speech unit sample required from the speech sample database 110 according to music information having lyrics information and melody information. The data is read, and a singing sound signal is generated using the read speech segment sample data. In the second embodiment, since only one speech segment sample data is stored in the speech sample database 110 for one phoneme or phoneme chain, when generating a singing sound signal according to the music information of the choral tune. The song generators 120, 121, 122 may use the same speech segment sample data. As described above, when singing sound signals of a plurality of parts are generated using the same speech segment sample data, fine features (such as pitch fluctuation) are basically the same, which is unnatural to the listener. It gives an impression.

  Therefore, in this chorus synthesizer 400, the chorus control unit 140 divides the lyrics information and melody information of the music information of the choral music into parts and outputs them to the song generators 120, 121, 122, and the voice sample database. Designation information for designating at which time the speech segment sample data stored in 110 is to be used is output to each song generator 120, 121, 122.

  As described in the first embodiment, the speech segment sample data stored in the speech sample database 110 is created based on the speech uttered by the speaker, and has a predetermined time length (1 frame). This is data created based on a speech waveform of ~ several frames. That is, it is data created based on the speech waveform represented by the relationship between time and amplitude within the predetermined time. Therefore, even when speech segment sample data is stored as frequency domain data as in the first embodiment, the data is obtained by performing FFT or the like on the time domain speech waveform. The chorus control unit 140 provides the singing generators 120, 121, and 122 with designation information that designates from which part the voice segment sample data, which is information that changes with time, is used. Supply.

  Here, the chorus control unit 140 generates each singing sound signal so that each singing generator 120, 121, 122 starts using the speech segment sample data from portions corresponding to different times and generates a singing sound signal. Designation information for designating different use start times is output to the generators 120, 121, and 122.

  Each singing generator 120, 121, 122 reads out the speech unit sample data required based on the lyrics information of each part supplied from the chorus control unit 140 from the speech sample database 110 and reads out the speech unit sample that has been read out. The use of the data is started from the part corresponding to the time specified in the specification information from the chorus control unit 140, and the singing sound signal is generated.

  Hereinafter, the speech unit sample data read according to the lyrics information of the three parts is “a” of the vowel, and the speech unit sample data “a” is composed of 13 frames (time 0 to T) such as F0 to F13. The case where each song generator 120, 121, 122 generates a singing sound signal by using the length of 13 frames of the speech segment sample data is specifically illustrated with reference to FIGS. To explain.

  In the example shown in FIG. 5, designation information is supplied to the song generator 120 so as to start use from the first frame F0, and the song generator 121 starts use from the frame F3. The designation information that designates to be used is supplied, and the designation information that designates the song generator 122 to start use from the frame F6 is supplied. In the drawing, for convenience of explanation, the speech segment sample data is shown as a speech waveform in the time domain, but the data stored in the speech sample database 110 is expressed in the frequency domain as in the first embodiment. It may be in the form of a harmonic component (line spectrum) and an anharmonic component (residual spectrum).

  When such designation information is supplied, as shown in FIG. 6, the singing generator 120 uses the order of frames F0, F1, F2,... F13, that is, the speech segment sample data “a” as it is. Used to generate a singing sound signal. The singing generator 121 generates a singing sound signal by using the speech segment sample data “a” in order of frames F3, F4, F5... F13, F0, F1, F2, F3. Further, the singing generator 122 generates a singing sound signal by using the speech segment sample data “a” in the order of frames F6, F7... F13, F0, F1.

  In this way, the chorus control unit 140 starts using the portions corresponding to different times and outputs the designation information so as to generate the singing sound signal, so that the same phoneme “a” has the same time length (0 to T). Even when the singing sound signal is generated using only the singing), the data actually used by the singing generators 120, 121, 122 are different. That is, the fine features (pitch fluctuations, etc.) shown in the speech segment sample data actually used by each singing generator 120, 121, 122 are different, and one kind of speech element is obtained for one phoneme or phoneme chain. Using the single sample data, each singing generator 120, 121, 122 can generate singing sound signals with different fine features.

  By the way, when using speech segment sample data for a single phoneme like the phoneme “a”, it is generated for each part by the method of simply shifting the use start time in the data as described above. It is possible to synthesize a more natural chorus sound by changing the minute characteristics of the singing sound, but in the case of speech segment sample data for a phoneme chain consisting of a plurality of phonemes, the use start time in the data is simply It may be inconvenient to simply shift the position. For example, in the case of speech sample data for a phoneme chain such as “ai”, the first half in the time domain is data that more strongly reflects the phoneme of “a”, and the second half is data that more strongly reflects the phoneme of “i”. It is. Therefore, in order to generate the singing sound signal of the phoneme chain “ai”, when the use is started from the latter half portion where the influence of the phoneme “i” is strong, data having a tendency similar to that of the phoneme chain “ia” is used. In this case, the signal for the phoneme chain “ai” to be generated cannot be generated accurately.

  Therefore, in this embodiment, when using speech segment sample data corresponding to a plurality of phoneme chains, the chorus control unit 140 sends designation information as shown in FIG. 7 to each song generator 120, 121, 122. I am trying to output. In the example shown in the figure, the singing generator 120 is supplied with designation information that specifies that the use is started from the first frame F0, and the singing generator 121 is used from the frame F2. The designation information that designates to be used is supplied, and the specification information that designates the song generator 122 to start use from the frame F4 is supplied. That is, when compared with the designation information for the single phoneme, the use start times designated for the song generators 120, 121, 122 are concentrated in the first half (the influence of “a” is strong). Thus, by concentrating the use start time of each part on the first half of the data, it is possible to prevent the data actually used from being similar to the phoneme chain “ia” as described above.

  Further, when the designation information is supplied as described above, the singing generator 121 performs the speech segment sample data in the order of frames F2, F3, F4... F13, F0, F1, F2, that is, in one direction. Is used to generate the singing sound signal, the frames F0 to F2 that are strongly influenced by the phoneme “a” are used as the data in the latter half that should be strongly influenced by the “i”. Therefore, in this embodiment, when speech segment sample data for a phoneme chain composed of a plurality of phonemes is used, frames F12, F11,... Are not returned to the frame F1 after the last frame (F13). Thus, the frames are used in the order of returning in the reverse direction. Therefore, when the use start frame is designated as shown in FIG. 7, the song generator 120 performs the order of frames F0, F1, F2,... F13, that is, the voice sample database 110 as shown in FIG. Is used as it is to generate a singing sound signal. Further, the singing generator 121 generates a singing sound signal using the speech segment sample data in the order of frames F2, F3, F4... F13, F12, F11. Further, the song generator 122 generates a song sound signal using the speech segment sample data in the order of frames F4, F5, F6... F13, F12, F11, F10, F9. Note that when frames are used in the reverse order, such as from frame F13 to frame F12, there is a risk of noise or the like occurring at the connection between the two, so amplitude adjustment processing or crossfade processing is performed at the connection between the frames. Etc. may be applied.

  In the case where the speech unit sample data of a phoneme chain composed of a plurality of phonemes is used in each song generator 120, 121, 122, a phoneme chain can be generated more accurately as described above. The fine features of the singing sound signals output from the singing generators 120, 121, 122 are different.

  As explained above, in the chorus synthesizer 400 according to the second embodiment, even if only one speech segment sample data is stored for one phoneme or phoneme chain, one speech segment sample data is used. As in the first embodiment, it is possible to synthesize a choral sound signal that can give a more natural impression. That is, it is possible to synthesize a chorus sound signal that can give a more natural impression while suppressing the amount of data stored in the audio sample database 110.

C. Modifications Note that the present invention is not limited to the first and second embodiments described above, and various modifications as exemplified below are possible.

(Modification 1)
In each of the embodiments described above, unit speech unit sample data such as phonemes or phoneme chains are connected to generate a singing sound signal, but there is a singing expression method called vibrato, You may make it add the function which adds the singing expression by this vibrato to a chorus synthesizer.

  Conventionally, as a method for generating a singing sound signal for electronically pronouncing a singing sound by vibrato, as in each of the above-described embodiments, the speech unit sample data of phonemes or phoneme chain units is connected, and the connection There has been known a method of applying a frequency modulation of about 6 Hz to a waveform represented by the speech segment sample data. A configuration for carrying out such a method may be added to the choir synthesizer in each of the above embodiments, but as a method of generating a vibrato singing sound signal that can give a natural impression to the listener, There is a method of using vibrato voice sample data created based on the voice when a person sings with the vibrato singing method, and it is preferable to add a configuration for carrying out this method to the chorus synthesizer according to each of the above embodiments. .

  Hereinafter, with reference to FIG. 9, an example in which a function of generating a singing sound signal using vibrato sound sample data created based on a speaker's vibrato singing sound is added to the chorus synthesizer in the first embodiment will be described. Will be described.

  As shown in the figure, in the voice sample database 110 in the chorus synthesizer 100 ′, in addition to the voice unit sample data in units of phonemes or phoneme chains such as the voice sample data groups 110a, 110b, 110c, the vibrato song Vibrato voice sample data created based on the singing voice of the time is stored. Here, in the audio sample database 110, three vibrato audio sample data BDa, BDb, and BDc created based on different sounds are stored.

  Under this configuration, the chorus control unit 140 designates the lyric information and melody information of each part and the voice sample data group to be used for each singing generator 120, 121, 122, as in the first embodiment described above. In addition to the information, second designation information for designating which of the three vibrato sound sample data BDa, BDb, and BDc is to be used is supplied. Here, the 2nd designation | designated information supplied to each song generator 120,121,122 is information which designates using different vibrato audio | voice sample data. By supplying such second designation information to each song generator 120, 121, 122, each song generator 120, 121, 122 receives different vibrato sound sample data when generating a vibrato song sound signal. The waveform represented by the read vibrato speech sample data is superimposed on the speech waveform represented by the speech unit sample data read out and connected in the same manner as in the above embodiment, and the superimposed waveform signal is output as a singing sound signal.

  In this way, when generating the vibrato singing sound signal, each singing generator 120, 121, 122 uses the three vibrato sound sample data BDa, BDb, BDc created based on different sounds for each part. The fine characteristics of the vibrato singing sound signal to be generated (such as frequency fluctuation during vibrato) are also different for each part. In this way, there is almost no correlation between each part of the vibrato singing sound, and each part has a unique characteristic, so that the singing sound based on the choral sound signal synthesized by the choral synthesizer 100 ' It becomes possible to give a more natural impression to the listener who has listened to the vibrato portion.

  By the way, the fact that the characteristics of the vibrato part of each part are basically the same in the chorus sound gives the listener a more unnatural impression than when the characteristics of the other parts are the same. Therefore, there may be a demand for an apparatus in which only the vibrato portion is given unique characteristics for each part. In such a case, as in each of the above embodiments, the voice sample data about the phoneme or the phoneme chain uses the same sound as it is in each part to generate a singing sound signal, and the generated singing sound signal A vibrato effect may be given by adding waveforms expressed by different vibrato audio sample data for each part.

(Modification 2)
Moreover, as shown in FIG. 9, you may make it use three vibrato audio | voice sample data corresponding to the number of each song generator 120,121,122, but of choral song synthesis apparatus 400 'shown in FIG. As described above, the song generators 120, 121, and 122 may generate the song sound signal of the vibrato portion using the same vibrato sound sample data.

  As described in the above-described embodiment, the song generators 120, 121, and 122 can generate song sound signals having different unique characteristics by properly using the audio sample data groups 110a, 110b, and 110c. Even if the waveform expressed by the same vibrato sound sample data is added to the singing sound signal generated in this way, the singing sound signal of the vibrato portion output from each of the singing generators 120, 121, and 122 has unique characteristics. It will have. Therefore, each song generator 120, 121, 122 may simply use one vibrato sound sample data. However, each song generator 120, 121, 122 in the second embodiment is also used for vibrato sound sample data. As described as the method of using the speech segment sample data by 122, each song generator 120, 121, 122 may start using the portion corresponding to different time of the same vibrato speech sample data. . In this case, what is necessary is just to make it supply to each song generator 120, 121, 122 the designation | designated information which designates from which part the chorus control part 140 starts using from. By doing in this way, the actual audio sample data which each song generator 120, 121, 122 uses for vibrato provision has a different characteristic. Therefore, the singing sound signal of the vibrato part of each part has a unique characteristic, and the listener who has listened to the vibrato part of the singing sound based on the choral sound signal synthesized by the chorus synthesizer 400 ′. It becomes possible to give a more natural impression.

(Modification 3)
In the above modification, the vibrato voice sample data is stored in the voice sample database 110 in order to give the vibrato effect to the singing sound signal to be generated. In order to electronically emit the singing sound of the singing method, the voice sample data created based on the singing voice of the tremolo part and the singing voice of the portamento part by the speaker is stored in the voice sample database 110. It may be. Also in this case, as with the vibrato audio sample data in the above-described modification, audio sample data is prepared for each part, or even the same audio sample data is used from a part corresponding to a different time. By doing so, it is possible to give unique characteristics to the tremolo of each part and the singing sound signal of the portamento part.

(Modification 4)
Further, in the first embodiment described above, the three audio sample data groups 110a, 110b, and 110c are stored in the audio sample database 110. However, based on audio having different pitches such as high, medium, and low sounds. Thus, speech segment sample data included in each speech sample data group 110a, 110b, 110c may be created. For example, speech segment sample data created based on high-pitched speech is included in the speech sample data group 110a, and speech segment sample data created based on medium speech is included in the speech sample data group 110b. In this way, speech segment sample data created based on low-pitched speech may be included in the speech sample data group 110c.

  When the audio sample database 110 that stores the audio sample data groups 110a, 110b, and 110c created for each sound range is used in this way, the chorus control unit 140 has a high frequency range among a plurality of parts included in the music information. Designation information for designating the use of the voice sample data group 110a created based on the high-pitched sound is output to the singing generator responsible for generating the singing sound signal of the part composed of the melody. In addition, for the singing generator responsible for generating the singing sound signal of the part composed of the melody in the middle range, the designation information for designating the use of the voice sample data group 110b created based on the middle tone voice is output. Further, designation information for designating the use of the voice sample data group 110c created based on the low-pitched sound is output to the singing generator responsible for generating the singing sound signal of the part composed of the melody in the low-pitched range. Thereby, each singing generator 120, 121, 122 can use voice segment sample data more suitable for generation of the singing sound signal of the part each is in charge, and can generate a higher quality singing sound signal. it can.

  The voice sample data groups 110a, 110b, and 110c used by the song generators 120, 121, and 122 may be fixed at the time of generating a song sound signal corresponding to a certain piece of music as described above. It is also conceivable that the pitch level of each part determined by the melody information changes every time. In this case, when generating the singing sound signal of a certain piece of music, the chorus control unit 140 selects the singing generators 120, 120 according to the pitch of each part determined by the melody information for each part. It is also possible to output designation information such that the audio sample data group designated for 121, 122 is sequentially changed during the music.

(Modification 5)
In the above-described modification, the voice sample data groups 110a, 110b, and 110c created based on voices at different pitches are stored in the voice sample database 110. However, the same phoneme is uttered when singing. The pitch may fluctuate greatly in between. Therefore, speech unit sample data is created in the speech sample database 110 based on speech generated by changing the pitch (pitch) while speaking the same phoneme, for example, “a”. The sample data may be stored in the voice sample database 110. As described above, in the speech sample database 110, speech segment sample data is created in consideration of not only phonemes or phoneme chains having the same pitch described in the above-described embodiments, but also various pitch fluctuations that may occur during singing. You may make it leave.

(Modification 6)
In the first embodiment described above, the voice sample data groups 110a, 110b, and 110c stored in the voice sample database 110 are used by the singing generators 120, 121, and 122, and in the second embodiment, the same. By using the voice segment sample data from the part corresponding to different times, a chorus sound signal capable of giving a more natural impression was synthesized. In the choir synthesizer according to the first and second embodiments, the singing generators 120 and 121 assign some value (that is, a parameter that determines the sound) indicated in the speech segment sample data read from the speech sample database 110 to each of the song generators 120 and 121. , 122 may be provided to provide parameter changing means for supplying after changing. FIG. 11 shows a configuration when such parameter changing means is added to the chorus synthesizer according to the first embodiment.

  As shown in the figure, in addition to the configuration of the choral synthesizer 100 in the first embodiment, the choral synthesizer 100 ″ includes a parameter changing unit 220 provided corresponding to each singing generator 120, 121, 122. Under this configuration, the chorus control unit 140, in the same way as the first embodiment described above, provides the song generators 120, 121, and 122 with the lyrics information and melody information of each part, and which audio sample. While outputting the designation information which designates whether a data group is used, the change information which shows the change content of a parameter is output with respect to each of parameter change part 220,221,222. Change information that changes different contents of the speech segment sample data read from the speech sample database 110 is changed to each parameter change. And outputs it to the parts 220, 221, 222.

  The parameter changing units 220, 221, and 222 read out the speech unit sample data required by the corresponding song generators 120, 121, and 122 from the speech sample data group indicated by the designated information, and supply it from the chorus control unit 140. The speech segment sample data read out is changed in accordance with the change information. Then, the changed speech segment sample data is supplied to the corresponding song generators 120, 121, 122. And each song generator 120,121,122 produces | generates a song sound signal using the audio | voice unit sample data after a change.

  Here, the content of the changing process performed on the speech segment sample data read from the speech sample database 110 by the parameter changing unit 220, 221, 222 is a process of changing the timbre etc. to such an extent that the phonological property is not impaired. Various modification processes can be applied. For example, the voice formant structure expressed by a certain voice segment sample data read from the voice sample database 110 is modeled, and the tone color is changed by changing the band width of the formant by several percent or shifting the center frequency of the band by about 10 Hz. There is a way to change it slightly. In this case, by changing the ratio of the bandwidth of the formant to be changed and the amount by which the center frequency of the band is shifted to each of the parameter changing units 220, 221, and 222, the parameter changing units 220, 221, and 222 The timbre of the voice shown in the read voice segment sample data is slightly different.

(Modification 7)
Moreover, in each embodiment mentioned above, in order for the chorus sound produced | generated by each song generator 120,121,122 to give a listener a more natural impression, the song sound by the song sound signal produced | generated for every part The sound generation timing may be shifted. In this case, the chorus control unit 140 supplies timing designation information for designating how much the sound generation timing is shifted for each part. At this time, the chorus control unit 140 supplies the singing generators 120, 121, and 122 with timing designation information such that the sound generation timings of the singing generators 120, 121, and 122 are slightly shifted. For example, for the singing generator 120, the singing sound signal generated according to the lyrics information and the melody information supplied from the chorus control unit 140 is output to the adder 130 without delay, and the singing generator 121 is output. If the singing sound signal is output to the adder 130 with a delay of 10 msec, and the singing sound signal is output to the adder 130 with a delay of 20 msec for the singing generator 122, the singing sound of each part can be obtained. It is pronounced with a slight shift, giving a more natural impression to the listener.

  In addition, when generating the singing sound signal of one piece of music as described above, the correlation between the sound generation timings of the singing generators 120, 121, and 122 may be fixed. Even in the middle, the correlation of the sound generation timings of the song generators 120, 121, 122 may be changed. For example, the first half of the music is sounded in the order of song generators 120, 121, 122 as in the above example, and the second half of the music is sounded in the order of song generators 122, 121, 120. Good.

(Modification 8)
Moreover, in 1st Embodiment mentioned above, although the audio | voice sample database 110 memorize | stores the audio | voice sample data group of the kind according to the number (three) of the singing generators 120, 121, 122, a singing generator A larger number of types of audio sample data groups may be stored.

  Further, in the case where three song generators such as the song generators 120, 121, and 122 are provided, and only two voice sample data groups 110a and 110b are stored in the voice sample database 110, at least two song songs are stored. What is necessary is just to make it produce | generate a song sound signal using the audio | voice sample data group 110a, 110b from which a generator differs. In this case, the song generator 120 uses the voice sample data group 110a, the song generator 121 uses the voice sample data group 110b, and the song generator 122 uses one of the voice sample data groups 110a and 110b as a song generator. If the use is started from a portion corresponding to a time different from 120, 121, the three song generators 120, 121, 122 will actually generate a song signal using different speech segment sample data, As in the above embodiments, a choral sound signal that can give a natural impression can be synthesized.

(Modification 9)
The chorus synthesizer in each of the embodiments and modifications described above may be configured by a dedicated hardware circuit, but may be configured by software by a computer system as shown in FIG. As shown in the figure, this computer system includes a CPU (Central Processing Unit) 320 for controlling the entire apparatus, a ROM (Read Only Memory) 321 for storing various control data and program groups, and a RAM (Random) used as a work area. (Access Memory) 322, an external storage device 323 such as a hard disk or CD-ROM (Compact Disc Read Only Memory) drive for storing music information and program groups, an operation unit 324 such as a keyboard and a mouse, and a display for displaying various information to the user A unit 325, a D / A converter 326, an amplifier 327, and a speaker 328.

  The CPU 320 constructs the audio sample database 110 in the RAM 322 or the external storage device 323 according to a program group stored in the ROM 321 or an external storage device 323 such as a hard disk, and uses the audio sample database 110 to change the above-described embodiments and modifications. The singing sound signal synthesis process for each part is performed as in the example. Then, CPU 320 adds the generated singing sound signals for each part, and then outputs the added chorus sound signals to D / A converter 326. In the D / A converter 326, the choral sound signal is converted into an analog signal, amplified by the analog signal amplifier 327 of the choral sound, and then emitted from the speaker 328.

  As described above, the chorus synthesizer in each of the above embodiments and modifications can be configured by software using a computer system, and a program for causing a computer system to execute the chorus sound synthesizing process similar to each of the above embodiments and the like. You may make it provide to a user with the form. As a method of providing such a program, there are a method of providing it by storing it in various recording media such as a CD-ROM and a floppy disk, a method of providing it via a communication line such as the Internet, and the like.

It is a block diagram which shows the basic composition of the choral synthesizer which concerns on 1st Embodiment of this invention. It is a figure for demonstrating the production method of the audio | voice sample database which is a component of the said choral synthesizer. It is a block diagram which shows the function structure of the song generator which is a component of the said chorus synthesis apparatus. It is a block diagram which shows the basic composition of the choral synthesizer which concerns on 2nd Embodiment of this invention. It is a figure for demonstrating the singing sound signal generation method by the said chorus synthesizer which concerns on 2nd Embodiment. It is a figure for demonstrating the singing sound signal generation method by the said chorus synthesizer which concerns on 2nd Embodiment. It is a figure for demonstrating the singing sound signal generation method by the said chorus synthesizer which concerns on 2nd Embodiment. It is a figure for demonstrating the singing sound signal generation method by the said chorus synthesizer which concerns on 2nd Embodiment. It is a block diagram which shows the basic composition of the modification of the said chorus synthesizer which concerns on 1st Embodiment. It is a block diagram which shows the basic composition of the modification of the said chorus synthesizer which concerns on 2nd Embodiment. It is a block diagram which shows the basic composition of the other modification of the said choral synthesizing | combining apparatus which concerns on 1st Embodiment. It is a block diagram which shows the structure of the computer system for implement | achieving the function by the said choral synthesizer by software.

Explanation of symbols

100, 100 ', 100 "... chorus synthesizer, 110 ... voice sample database, 110a, 110b, 110c ... voice sample data group, 120 ... song generator, 121 ... song generator, 122 ... song Generator, 130... Adder, 140... Choral control unit, 200... SMS analysis unit, 201... Segment extraction unit, 220, 221, 222. 302 ...... Pitch determination unit 303 ... Duration duration adjustment unit 304 ... Voice unit connection unit 305 ...... Harmonic component generation unit 306 ...... Addition unit 307 …… Inverse FFT unit 308 …… Window Hanging part, 309... Overlapping part, 400, 400 ′.

Claims (8)

A chorus synthesizer that synthesizes a choral sound signal according to music data,
A database for storing audio sample data having a predetermined length of time created based on audio;
A means for generating a singing sound signal according to the music data, wherein a plurality of singing sound generating means used for generating the singing sound signal by reading out the required voice sample data from the database;
Singing synthesis means for synthesizing a chorus sound signal from the singing sound signals generated by the plurality of song generation means;
For each of the plurality of song generation means, comprising: designation means for designating portions corresponding to different times of the voice sample data stored in the database,
When the music data is composed of a plurality of parts and each of the plurality of song generation means generates a song sound signal corresponding to each of the parts, at least two of the song generation means are read from the database. Start using from the portion specified by the specifying means of the audio sample data to generate the singing sound signal,
When the voice sample data to be read from the database is voice sample data for a phoneme segment that is a phoneme chain, the designation means specifies a use start time designated for each of the plurality of song generation means, A choral synthesizer that concentrates on the first half of audio sample data. The database stores audio sample data having the predetermined time length for a phoneme or a phoneme segment that is a phoneme chain that is a connection of two or more phonemes;
The singing generation means reads out and connects voice sample data corresponding to the lyrics shown in the music data from the database, adjusts the connected voice sample data according to the pitch shown in the music data, and sings sound signal The choral synthesizer according to claim 1, wherein: The database stores vibrato audio sample data having the predetermined time length indicating the characteristics of the vibrato portion of the audio;
3. The choral synthesizer according to claim 1, wherein the singing generation unit reads out and uses the vibrato voice sample data stored in the database when generating a singing sound signal of a vibrato portion. 4. . When the music data consists of a plurality of parts, and each of the plurality of song generation means generates a song sound signal corresponding to each of the parts, the song sound signal generated by each of at least two of the song generation means The choir synthesizing apparatus according to claim 1, further comprising timing adjusting means for shifting the output timing of the chord synthesizing apparatus. Provided corresponding to each of the plurality of singing generation means, comprising timbre changing means for changing data relating to the timbre included in the voice sample data read from the database;
Each of the timbre changing means corresponding to at least two of the song generation means when the music data is composed of a plurality of parts and each of the plurality of song generation means generates a song sound signal corresponding to each of the parts. The choir synthesizing apparatus according to claim 1, wherein data relating to the timbre is changed by different methods. The database is a voice sample data group composed of a plurality of voice sample data, and each voice sample data group created based on a plurality of different voices is stored for each sound range,
When the music data is composed of a plurality of parts, and each of the plurality of song generation means generates a song sound signal corresponding to each of the parts, each of the at least two song generation means includes: 6. The voice sample data included in a voice sample data group corresponding to a range corresponding to each part is read from the database and used for generation of the singing sound signal. 6. Choral synthesizer. A chorus synthesis method for synthesizing a choral sound signal from a plurality of singing sound signals generated based on music data,
When generating a singing sound signal corresponding to the plurality of parts according to the music data consisting of a plurality of parts, it is necessary from a database storing sound sample data having a predetermined time length created based on sound. Read the voice sample data
For generation of the singing sound signal corresponding to at least two of the parts, the use is started from a part corresponding to a different time of the audio sample data read from the database, and the singing sound signal is generated,
When the voice sample data read from the database is voice sample data for a phoneme unit that is a phoneme chain, the use start time of the voice sample data is concentrated on the first half of the voice sample data. Chorus synthesis method. Computer
A means for reading out the voice sample data required from a database for storing voice sample data having a predetermined time length created based on voice according to the music data and generating a singing sound signal, the music data Is composed of a plurality of parts, and when generating the singing sound signals corresponding to the plurality of parts, the sound sample data read from the database when generating the singing sound signals corresponding to at least two of the parts Singing sound generation means for generating the singing sound signal by starting use from a portion corresponding to a different time;
While functioning as a singing synthesis means for synthesizing a chorus sound signal from the generated singing sound signal,
When the voice sample data to be read from the database is voice sample data for a voice unit that is a phoneme chain, the use time of the voice sample data is set to the first half of the voice sample data. A program to function as a means to concentrate.

Images