JPH1115489A - Singing sound synthesizing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To synthesize a more natural singing sound on the basis of text data. SOLUTION: Text data are read out corresponding to melody data stored in a music information storage part 5 and voice parameters consisting of phonematic formant data and phonematic articulation coupling control data corresponding to the vocal sound of the text data are read out of a voice quality control information storage part 6; and a voicing parameter supply control part 7 interpolates the respective parameters and supplies them to formant waveform generation parts 81 to 8m at specific intervals of time to synthesize and output a singing voice corresponding to the text. The speed of pitch variation at the time of a rise in interval is made slower than that at the time of a decrease. Further, the pitch is held when a voiceless sound is generated and begins to be varied when a voiced sound is generated. For consonant and vowel sounds, the target values of the formant frequencies of the vowels are varied.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a singing sound synthesizing apparatus which generates a corresponding phoneme based on lyrics data and sings the lyrics with a human voice.

[0002]

2. Description of the Related Art As one of the speech synthesis techniques, a speech synthesis system using formant synthesis is known. According to this method, storage means for storing parameter data relating to a formant that changes in time series over a plurality of steps, and reading out the parameter data from the storage means in a time series over a plurality of steps when a voice is to be uttered A readout unit; and a formant synthesizing unit that receives the read-out parameter data and synthesizes a tone signal having a formant characteristic determined according to the parameter data, and changes a formant of the audio signal in a time-series manner. Things.

Recently, a singing sound synthesizer (singing synthesizer) which synthesizes and outputs a natural singing sound based on lyrics data by applying such a voice synthesis technique to music has been proposed (Japanese Patent Laid-Open No. 9-50287). No.). The singing sound synthesizer stores lyrics data and melody data, reads the lyrics data in response to the reading of the melody data, and causes the lyric to be sung by generating a phoneme corresponding to the lyrics data. It is configured. Then, a plurality of syllable data can be pronounced within a time when one note is pronounced, and a consonant pronunciation time is set to a set time, and a vowel pronunciation time is changed according to a note length. It is designed to be able to synthesize and output natural singing sounds.

[0004]

However, in order to sound natural singing voice, there are other problems specific to singing in addition to the above. For example, in singing, the pitch changes, so that a change in pitch naturally appears. (A) of FIG.
FIG. 7 is a diagram showing a change in pitch frequency when a sound is generated while changing the pitch. As shown in this figure, the slope of the pitch change differs between when the pitch is raised from a low pitch to a high pitch and when it is lowered from a high pitch to a low pitch. That is, the gradient of the pitch change when the pitch rises to a high pitch is gentler than the gradient when the pitch falls to a low pitch. The pitch is raised and lowered by the tension and relaxation of the muscles in the vocal cords, which is considered to be due to its mechanical characteristics.

In the conventional singing sound synthesizer, as shown in FIG. 7 (b), the pitch is changed in the same manner both when the pitch is raised and when the pitch is lowered. Therefore, there was a feeling of discomfort that was not caused by natural utterances, especially when the pitch was lowered. Therefore, as shown in FIG. 7 (c), it is necessary to change the inclination of the pitch change when the pitch rises and when the pitch falls, in order to make the human sense of hearing uncomfortable.

FIG. 8 shows a voiced sound V
FIG. 7 is a diagram for explaining how the pitch changes when a syllable composed of an unvoiced sound U and a voiced sound V having different pitches is generated after. As shown in FIG. 3A, in such a case, pitch change is usually caused from the start of the sounding of the unvoiced sound U. However, since the unvoiced sound U has no pitch component, the sounding period of the unvoiced sound U is generated. Does not output a pitch change. Therefore, as shown in FIG. 8B, a vowel V with a sudden jump in pitch is generated after the vowel V of the syllable is in the sounding period. Therefore, the connection between the pitches results in a discontinuous and unnatural sound. Therefore, there is a demand for an audio output which does not have such unnaturalness and has a natural pitch connection as shown in FIG.

Furthermore, in actual human pronunciation, when a vowel is sandwiched between consonants such as consonant C (consonant), vowel V (vowel), and consonant C, the original formant center frequency of the vowel Changes to the center frequency of the formant of the next consonant (so-called undershoot) without reaching. FIG. 9 is a diagram showing this state. For example, when the vowel u is generated alone, the first formant center frequency FF1 and the second formant center frequency FF2 as shown in FIG.
It has become. However, when the sound is produced as dud, as shown in FIG. 9B, the second formant center frequency FF2 of the vowel u becomes the FF shown in FIG.
The transition is made to the next d second formant frequency without reaching the frequency of 2. Thus, a vowel sandwiched between consonants is actually produced at a formant center frequency that is different from the formant center frequency when it is independently produced.

Accordingly, an object of the present invention is to provide a singing sound synthesizing apparatus capable of producing a more natural singing voice by taking in the features appearing in actual human pronunciation as described above.

[0009]

In order to achieve the above object, a singing sound synthesizer according to the present invention stores lyrics data and melody data, and reads out the lyrics data in response to the reading of the melody data. In a singing sound synthesizing device configured to cause a phonogram corresponding to the lyrics data to sound and sing the lyrics, the pitch change speed when increasing the pitch of the phonemes to be generated decreases the pitch of the phonemes to be generated. The speed is set to be lower than the speed at which the pitch is changed. As a result, the same pitch change as in the case of human utterance can be realized, and a natural singing sound can be uttered.

Further, another singing sound synthesizer of the present invention stores lyrics data and melody data, reads the lyrics data in response to the reading of the melody data, and generates a phoneme corresponding to the lyrics data. In the singing sound synthesizer configured to cause the lyrics to sing, the voiced sound, the unvoiced sound, when changing the pitch of the phoneme to be pronounced in the case of sounding in the order of the voiced sound, the sounding period of the unvoiced sound is preceding The voiced sound pitch data is held, and the change of the pitch starts from the start of the sounding of the following voiced sound. As a result, the connection between the pitches becomes smooth, and a natural sound is produced.

Still another singing sound synthesizer of the present invention stores lyric data and melody data, reads the lyric data in response to the reading of the melody data, and generates a phoneme corresponding to the lyric data. In a singing sound synthesizer configured to sing and sing the lyrics, in the case of consonants, vowels, and consonants, when the vowel is to be sounded, the formant center frequency causes the vowel to sound alone. Before reaching the formant center frequency in the case, the formant center frequency of the following consonant is restored. As a result, the same phoneme as that generated by a human can be uttered, and a natural singing sound can be synthesized.

[0012]

FIG. 1 is a diagram showing an example of the system configuration of a singing sound synthesizer according to the present invention. In FIG. 1, reference numeral 1 denotes a control unit which controls the entire apparatus, and includes a CPU, a RAM used as various flags, various variable areas, buffers, and the like. Reference numeral 2 denotes a program storage device PRGMEM in which a control program and the like are stored. Note that the program storage device 2 may be a program supply device that supplies a control program to the control unit 1 via a network or the like. Reference numeral 3 denotes an operation setting unit provided with a display unit for displaying the operation state of the device, input data, a message for the operator, and the like, and various setting operators such as operation knobs and operation buttons. Reference numeral 4 denotes a MIDI interface unit, which is connected to an external MIDI device, a network, or the like, and performs input / output of MIDI data. Reference numeral 5 denotes a song information storage unit SONGMEM for storing song information including melody data and lyric data and song information including accompaniment data. In addition to semiconductor memory, a floppy disk device, hard disk device, MO disk device, IC memory card, etc. Various media can be used.

Reference numeral 6 denotes a voice quality control information storage unit PHONEMEM. The voice quality control information storage unit PHONEMEM 6 is provided with a voice parameter memory PMEM storing phoneme formant data PHFRMTDATA and phoneme tone combination control data PHCOMBDATA. ing. The voice parameter memory PMEM stores the type of voice to be generated, for example, male voice, female voice,
Or it is provided for each specific singer etc.,
The operator can select the voice parameter memory PMEM to be used according to the voice quality to be uttered. The phonological formant data PHFRMTDATA is
A parameter for synthesizing each phoneme in formant. Data (FShape 1 to m) that specifies the form of each formant to generate the phoneme, data that specifies the center frequency of each formant ( FF
req 1 to m), output level data of each formant (FLev
el 1 to m). The phonological tone combination control data PHCOMBDATA is various parameters for performing articulation coupling (particularly, formant frequency transition) when transitioning from one phonology to another phonology. Interpolation for each data in the phonological formant data PHFRMTDATA is performed. It consists of a rate and constants for correcting the interpolation rate.

Numeral 7 is a pronunciation voice parameter supply control unit which reads out the melody data and lyrics data of the music from the music information storage SONGMEM 5 and converts the phonological formant data PHFRMTDATA and the phonological tones combination control data PHCOMBDATA corresponding to the lyric data. The voice quality control information generation unit PHONEMEM reads out the voice parameter memory PMEM from the memory and performs an interpolation operation, and supplies it to the formant waveform generation units 81 to 8m at predetermined time intervals as sound generation control parameters.

Reference numerals 81 to 8 m denote formant waveform generation units FGen
1 to m, the sounding control parameters supplied from the sounding voice parameter supply control unit 7 of the control unit 1 every predetermined time, that is, formant center frequencies FFreq 1 to m,
Based on the formant amplitude levels FLevel 1 to m, the formant shape information FShape 1 to m, and the pitch information PITCH, corresponding formant waveforms are generated. As shown, each formant waveform generator FGen
Each of 1 to m is provided with an unvoiced sound source unit UNVOICED TG UNIT 9 for generating unvoiced sound and a voiced sound source unit VOICED TG UNIT 10 for generating voiced sound, and the outputs of the respective sound source units 9 and 10 are added. Then, the formant waveform generator FGen n (n = 1 to
m) is output as FORMANT_OUT n (n = 1 to m). Such a speech synthesizer has already been proposed by the present applicant (JP-A-3-200299). Further, each of the formant waveform generators 81 to 8m
Can also generate musical tones, and sound sources not assigned as sounding channels for voice can be allocated to generating musical tones.

Reference numeral 11 denotes a signal synthesizing unit, which is an output signal output from the m formant waveform generating units 81 to 8m.
FORMANT_OUT 1 to m are converted to a control signal MI supplied from the control unit 1.
This is a signal synthesis unit that adds and outputs according to XCONT.

The operation of the singing sound synthesizing apparatus according to the present invention thus configured will be described. FIG. 2 is an operation flowchart of a main routine executed in the control unit 1. When the operation is started, first, in step S10
In, the system is initialized. Subsequently, the process proceeds to an operation event detection process in step S20, and it is determined whether or not an operation event has occurred in the operation setting unit 3. Next, proceeding to step S30, if the operation event detected in the operation event detection process S20 is a selection event of a music piece and a song to be played, data (accompaniment data, melody data, Lyrics data) to the music information storage unit
Read from SONGMEM5. The melody data is the same as the MIDI sequence data, and includes KEYON, KEYOFF, pitch [PITCH], pitch [NOTELENGT
H], touch [TOUCH], etc. are included. Also,
The lyrics data is, for example, data in which phonemes corresponding to the lyrics are described, to which a delimiter indicating the delimitation of the notes has been added in order to associate the notes with the lyrics. further,
The accompaniment data is MIDI sequence data.

Then, the process proceeds to a step S40, where the performance processing of the accompaniment data of the selected music read out in the step S30 is performed. This is the same process as the normal tone generation process, and the accompaniment tone is generated by the tone generator 7.
Generated by Subsequently, the process proceeds to step S50, where a singing sound control process for the music is performed. With this process,
Singing sounds corresponding to the melody data and the lyrics data are generated in the formant waveform generating units FGen1 to FGen, and are synthesized and output in the signal synthesizing unit 11. The singing sound control processing S50 will be described later in detail. Subsequently, the process proceeds to step S60, where a process corresponding to the operation when the process detected in step S20 is other than the music / singing selection process and a display control process are performed. Then, the process returns to step S20, and the above-described processing is repeated and executed again.

Next, the singing sound control processing in S50 will be described. FIG. 3 is an operation flowchart of the singing sound control processing. In this flowchart, the control of the amplitude of the generated singing sound is omitted for the sake of simplicity. When the singing sound control process is started, first, in step S301, it is determined whether or not the singing sound generation process is being performed. Whether or not the singing sound generation process is being performed is determined based on whether or not a flag SINGING_START set on the RAM in the control unit 1 is “1”. This flag SINGING_START is "0"
In the case of, it is determined that the singing sound generation process is not being performed, and the process proceeds to step S302. Then, it is determined whether or not the singing sound generation start instruction event has occurred in the operation setting unit 3. As a result, when the singing sound generation start instruction event has not occurred, the singing sound control processing S50 ends, and the flow proceeds to the other processing of the step S60.

When a singing sound generation start instruction event has occurred, the result of the determination in S302 is YES, a singing sound generation processing initialization processing S303 is performed, and a sequence pointer i of a phoneme to be read from the lyrics data.
Is set to the initial value "1". Next, step S30
Proceeding to 4, the flag SINGING_START is set to "1", and this singing sound control processing is ended.

When the singing sound control processing S50 is started in a state where the flag SINGING_START is set to "1" (a state in which the singing sound generation processing is being performed), the determination result in the step S301 becomes NO. Step S305
Proceed to. Then, in this step S305, it is determined whether or not a singing sound generation end instruction event has occurred.
When the result of this determination is YES, the flag SINGING_START is reset to "0" in step S306, singing sound generation processing end processing is performed, and the singing sound control processing ends.

On the other hand, if the decision result in the step S305 is NO, the process advances to a step S307 to determine whether or not it is time to start generating the i-th phoneme, that is, whether the note-on event of the melody data has been started. It is determined whether or not the timing has come. This step S3
When the result of the determination in 07 is YES, a preparation process for generating this phoneme i is executed. First, step S3
At 08, a melody and lyrics event analysis is performed. Specifically, the pitch information KCPITCH included in the note-on event of the melody data, the currently sounding phoneme (i-1), its pitch (i-1), and its sounding time (i-1) Then, the phoneme i, the pitch i, and the sounding time i, which are the sounding start timing, are analyzed.
Here, the pitch i of the phoneme i is a frequency corresponding to the pitch information KC PITCH of the note, regardless of voiced sound or unvoiced sound. The sounding time i is determined by the sounding time setting of the phoneme and the KEYON and KEYOFF information of the melody event. Generally, in Japanese singing, the sounding time i is determined so that the preceding consonant is mainly pronounced for a predetermined time and the succeeding vowel continues until the next melody note key-on.

Next, the process proceeds to step S309, where the voice parameter memory PM of the voice quality control information storage unit PHONEMEM6 is stored.
From the EM, read out the phoneme / tone combination control data PHCOMBDATAx corresponding to the preceding phoneme (i-1) and the subsequent phoneme i.
Next, in step S310, phoneme formant data PHFRMNTDATAy corresponding to phoneme i to start sounding is read from the voice parameter memory PMEM. Then, the process proceeds to step S311 to set a sound-in-progress processing flag of phoneme i and perform a sound-generation start processing of phoneme i. Also, if necessary, a process of ending the pronunciation of the preceding phoneme (i-1) is performed. After the step S311 ends, the singing sound control process of this time ends. Details of the sounding start processing of the phoneme i will be described later.

When i = 1, that is, when the sound generation process of the first phoneme of the lyrics is performed, there is no preceding phoneme (i-1) in step S308. Further, in this embodiment, a silent section is generated in the case of a rest or a breath occurring during singing, and these are also treated as phonemes. Therefore, after a rest or a breath, the preceding phoneme (i-1) is performed in the same manner as in the case of the pronunciation of the first phoneme.
Does not exist. In order to cope with such a case, data indicating the transition state of the beginning of singing or the rising of the pronunciation is stored in the phonological and articulated tone combination control data PHCOMBDATA, and this data is stored in the phonological (i-1) and pitch. (I-1).

If the result of the determination in step S307 is NO, instead of the sounding start timing of the phoneme i, the flow advances to step S312 to determine whether the sounding process of the phoneme i is in progress. If the phoneme i sounding processing flag is set and the sounding processing of phoneme i is being performed, the determination result is YES, and the sounding voice parameter generation control processing of step S313 is performed. This process is a process of outputting a sound control parameter to the formant waveform generators 81 to 8m at predetermined time intervals, whereby the formant waveform generators 81 to 8m are output.
Is a process of actually outputting a singing sound. Details of this processing will be described later.

Next, the flow advances to step S314 to check the sounding time of the phoneme i (sounding time i), and the next phoneme (i +
It is determined whether the sound generation start timing of 1) has been reached (step S315). If the result of this determination is NO, this singing sound control processing S50 is ended, and
It proceeds to other processing of 0. On the other hand, when it is time to start generating the next phoneme (i + 1), step S316 is executed.
Then, the process proceeds to step S308, where (i + 1) is set as the phoneme sequence number i during sound generation. Hereinafter, steps S308 to S311 are executed for a new phoneme sequence number i. As a result, a preparation process for the pronunciation of the phoneme i is performed.

FIG. 4 is a flowchart for explaining the phoneme i sound generation start processing in S311. This phoneme i
When the pronunciation start processing is started, first, in step S401
In, it is determined whether or not this phoneme i is a phoneme sounded from a silent state. If the result of this determination is YES, there is no pitch of the immediately preceding phoneme, so step S402
To the melody pitch KCPITC corresponding to the phoneme i
H i is set to the pitch data PITCH i of the phoneme i and the pitch data PITCH i-1 of the phoneme (i−1) preceding the phoneme i. At this time, the phoneme i is started from the pitch i. Instead of setting the pitch data as in step S402, for example, 0 or another predetermined value can be set as PITCH i-1.
In this case, the set value is set as an initial value and rises to the pitch i of the phoneme i.

If there is a phoneme to be sounded earlier and the result of the determination in step S401 is NO,
Proceeding to step S407, it is determined whether or not the phoneme i is an unvoiced sound. If the phoneme i is an unvoiced sound and the determination result is YES, the process proceeds to step S408, where the pitch data PITCH i-1 of the preceding phoneme (i-1) is set as the pitch data PITCH i of the phoneme i which is the unvoiced sound. As a result, the pitch PITCH of the preceding phoneme can be held for an unvoiced sound. When the preceding phoneme (i-1) is not an unvoiced sound, the pitch i of the phoneme i is directly used as PITCH i.

After the pitch data PITCH i and PITCH i-1 of the phoneme i and (i-1) are set as described above, in step S403, PITCH i and PITCH i are set.
CH i-1 is compared. As a result, PITCH i> PITCH i-1
If the pitch is i, that is, if the pitch i of the phoneme i is higher than the pitch (i-1) of the phoneme (i-1) that is generated earlier, the process proceeds to step S404. This step S404
In the above, the interpolation rate Rpitch i of the pitch of the phoneme i included in the phoneme formant data PHFRMNTDATA stored in the voice parameter memory PMEM is multiplied by a coefficient Ku, and the pitch interpolation used in the interpolation calculation is performed. Let the rate be R'pitch i. Also, PITCH i ≤ PITC
When the pitch decreases at Hi-1, the determination result in S403 is NO, and in step S409, the interpolation rate Rpitchi of the phoneme i is multiplied by the coefficient Kd to obtain a pitch interpolation rate R'pitchi.

Here, 0 <Ku <Kd, and the pitch interpolation rate at the time of pitch rise is set smaller than the interpolation rate at the time of pitch decrease. The coefficient K
The interpolation rate Rpitch included in the phonological formant data PHFRMNTDATA is set as d = 1, as an interpolation rate when there is no change in pitch and when the pitch falls.
i may be used as it is. In this way, the pitch transition rate can be changed between when the pitch rises and when it falls.

Note that the coefficients Ku and Kd may be set to different values depending on the pitch difference or pitch. For example, the interpolation rate may be different when changing from a high sound to a higher sound. By doing so, a more natural singing sound can be obtained. Next, step S
Proceeding to 405, a correction calculation process of the formant center frequency is executed, and then, in step S406, a sounding start instruction process of the phoneme i is performed. This processing is based on the phoneme i
This is a process of setting a flag during sounding processing.

FIG. 5 is a flowchart of the FFreq correction calculation processing in step S405. This process is a process for controlling the formant center frequency of a vowel pronounced between consonants when the pronounced phoneme is a consonant (C) vowel (V) consonant (C). When this process is started, first, in step S501, the lyrics data is checked, and the preceding phoneme (i-1), the phoneme i that has become the sound generation start timing, and the subsequent phoneme (i + 1) are vowels (V). Or it is checked which of the consonants (C). As a result, when the CVC is continuously sounded, the determination result of step S502 is YES, and
Proceed to step S503. In this step S503, the initial value "1" is substituted for a variable j indicating the order of the formant, and the following steps S504 to S505 and S50
7, the target value of each formant center frequency is corrected by the loop of S506 and S509.

That is, first, in step S504, the target center frequency FFreq of the j-th formant of the phoneme i
ij is compared with the target center frequency FFreq (i-1) j of the j-th formant of the preceding phoneme (i-1). As a result, if the target center frequency FFreq ij of the j-th formant of the phoneme i is lower than the target center frequency FFreq (i-1) j of the j-th formant of the preceding phoneme (i-1), the process proceeds to step S505. , The target center frequency FFreq ij of the j-th formant of the phoneme i is multiplied by a coefficient Hj1 (0 <Hj1 <1) to obtain a new target center frequency FFreq ′ ij of the j-th formant set lower.

On the other hand, when the target center frequency FFreq ij of the j-th formant of the phoneme i is higher than the target center frequency FFreq (i-1) j of the j-th formant of the preceding phoneme (i-1), the step In S507, the phoneme i
To the target center frequency FFreq ij of the j-th formant of
j2 (1 ≦ Hj2) to obtain a new high j-th
The target center frequency of the formant is FFreq'ij. Note that the target center frequency FFreq of the j-th formant of phoneme i
When ij matches the target center frequency FFreq (i-1) j of the j-th formant of the preceding phoneme (i-1), the above-mentioned Hj2 = 1 is set so that the target center frequency is not corrected. I do.

Next, the process proceeds to step S506, in which it is determined whether or not processing for all formants has been completed.
If not, the variable j is incremented (S509), and the processing from step S504 is repeated. When the processing for all the formants is completed, the FFreq correction operation processing is completed.

If the determination in step S502 is NO
, That is, when the CVC is not continuously sounded, it is not necessary to change the target center frequency of the formant. Therefore, the target center frequency [FFreq] i of each formant corresponding to the phoneme i is set as the target. The center frequency [FFreq ′] i is set, and the FFreq correction operation processing is terminated. The above is the contents of the phoneme i pronunciation start processing in step S311. In this way, the sound generation preparation processing at the time of starting sound generation of the phoneme i in FIG. 3 ends.

Next, the sounding voice parameter generation control processing in step S313 will be described with reference to the flowchart of FIG. As described above, this step S313 is a process executed during the sounding process of the phoneme i, and supplies each parameter to the formant waveform generators 81 to 8m at predetermined time intervals to generate a predetermined phoneme. Processing.

In this process, first, at step S6
At 01, it is determined whether the preceding phoneme (i-1) is a voiced sound. If the preceding phoneme (i-1) is a voiced sound, the result of this determination is YES, and it is next determined in step S602 whether the phoneme i in the pronunciation process is a voiced sound. If the preceding phoneme (i-1) is unvoiced, the process proceeds to step S606, and the phoneme i
Is determined to be a voiceless sound.

When the result of the determination in step S602 is YES, that is, when the preceding phoneme (i-1) is a voiced sound, the phoneme i is a voiced sound, and the voiced sound continues to be generated, and If the decision result in the step S606 is NO, that is, if the preceding phoneme (i-1) is an unvoiced sound and the phoneme i is a voiced sound, the step S603 is performed.
Is performed. In step S603, the pitch between pitch PITCH i-1 of the preceding phoneme (i-1) and pitch PITCH i of the currently sounding phoneme i is set in step S40.
4 or the interpolation rate R'pitc set in S409
hi, and interpolate the result for a predetermined time (for example,
c) Each time the formant waveform generators 81 to 8m
This is a process of outputting as TCH data. Thereby, the pitch of the phonemes generated in each of the formant waveform generators 81 to 8m is changed at the set rate.

When the result of the determination in step S602 is NO, that is, when the preceding phoneme (i-1) is a voiced sound and the succeeding phoneme i is an unvoiced sound, and the result of the determination in step S606 is YES. If the preceding phoneme (i-1) is an unvoiced sound and the subsequent phoneme i is an unvoiced sound, step S607 is executed. In this step S607, the pitch PITCH at that time is held, that is, without performing the interpolation calculation processing,
It is output as PITCH data to the formant waveform generators 81 to 8m at predetermined time intervals. As a result, the pitch PITCH is held for an unvoiced sound.

After the pitch data is transmitted in step S603 or S604, step S603 is executed.
Proceeding to 604, control data of each formant frequency is transmitted. In this processing, the preceding phoneme (i
-1) and the phonological articulation combination control data PH
Interpolation is performed between [FFreq] i−1 to [FFreq ′] i for each formant frequency with an articulation coupling characteristic according to COMBDATAx, and an interpolated value is output at predetermined time intervals. Where [FFre
q] i−1 is the center frequency of each formant of the preceding phoneme (i−1), and [FFreq ′] i is the target center frequency of each formant of the phoneme i calculated in the FFreq correction calculation processing S405. It is.

Next, the process proceeds to step S605, where interpolation calculation processing is similarly performed for other sounding voice parameters, FShape, FLevel, and the like, and output to the formant waveform generators 81 to 8m at predetermined time intervals. In this way,
At predetermined time intervals (for example, several msec), the sounding voice parameters are transmitted to the formant waveform generators 81 to 8m, and the formant waveform generators 81 to 8m generate the corresponding formant waveforms of the phoneme as described above. Is done. The output corresponding to each formant output from each of the formant waveform generators 81 to 8m is the signal synthesizer 1
1 and the synthesized phoneme corresponding to the lyrics is pronounced.

In the above description, the parameters such as the formant center frequency, formant level, formant bandwidth and pitch frequency are transmitted from the control unit 1 at predetermined time intervals (for example, at intervals of about several milliseconds). The control is performed by sequentially transmitting the signals. However, the time interval is set longer, and each of the formant waveform generators 81 to 81 is controlled.
The parameters may be sequentially controlled by an envelope generator included in 8 m.

Further, in the above, a plurality of types of phoneme formant data PHFRMNTDATA and phoneme tone combination control data PHCOMBDAT are stored in the voice quality control information storage unit PHONEMEM6.
A is stored and selected according to the voice quality desired to be generated, but multiple types of stored data (for example, male voice, female voice, or individual voice obtained by analyzing individual voice) Phonological formant data or phonological tonal connection control data having an intermediate characteristic or a new characteristic by selecting some of them, and combining them arbitrarily and performing an interpolation process. The sound generation can be controlled using the data. For example, by generating phoneme formant data from two types of data, male and female, and using the data for pronunciation control, it is possible to generate phonemes with intermediate voice quality between two men and women.

Particularly suitable examples of the application field of the singing sound synthesizing device of the present invention include an electronic musical instrument, a computer system, a voice response device, a game machine and an amusement device such as a karaoke capable of outputting a singing sound. Can be Further, the singing sound synthesizer of the present invention can be implemented in the form of software of a computer system represented by a personal computer. At that time, the processing up to the synthesis of the audio waveform may be executed by the CPU,
Alternatively, a separate sound source may be provided as shown in FIG.

[0046]

As described above, according to the singing sound synthesizing apparatus of the present invention, the pitch changing speed when the pitch changes is made slower when the pitch is rising and faster when the pitch is falling. Even when the pitch changes, a natural singing sound can be uttered. Further, the connection of the pitches when the pitch changes is smoothed, resulting in a natural sound. Furthermore, when producing a consonant, a vowel, and a consonant in this order, the formant center frequency of the vowel is returned to the formant center frequency of the following consonant before reaching the formant center frequency of the case of producing the sound alone. Thus, it is possible to output a singing sound without a sense of incongruity.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an example of a system configuration of a singing sound synthesizer of the present invention.

FIG. 2 is a flowchart showing a main routine in a singing sound synthesizing measure of the present invention.

FIG. 3 is a flowchart illustrating a singing sound control process.

FIG. 4 is a flowchart illustrating a phoneme i pronunciation start process.

FIG. 5 is a flowchart illustrating an FFReq correction calculation process.

FIG. 6 is a flowchart illustrating a sound generation voice parameter generation control process.

FIG. 7 is a diagram for explaining a state when a pitch rises and falls.

FIG. 8 is a diagram for explaining unvoiced sound and voiced sound output when the pitch changes.

FIG. 9 is a diagram for explaining a change in formant frequency when a consonant vowel consonant is generated.

[Explanation of symbols]

1 control section, 2 program storage device, 3 operation setting section, 4 MIDI interface, 5 music information storage section, 6 voice quality control information storage section, 7 sounding voice parameter supply control section, 81-8 m formant waveform generation section,
9 unvoiced sound source section, 10 voiced sound source section, 11 signal synthesis section, 12 accompaniment sound source section

Claims (3)

[Claims] The present invention is configured to store lyrics data and melody data, read the lyrics data in response to the reading of the melody data, generate a phoneme corresponding to the lyrics data, and sing the lyrics. Singing sound, wherein the pitch change speed when increasing the pitch of the phoneme to be pronounced is set to be lower than the pitch changing speed when decreasing the pitch of the phoneme to be pronounced. Synthesizer. 2. The system is configured to store lyrics data and melody data, read the lyrics data in response to the reading of the melody data, generate a phoneme corresponding to the lyrics data, and sing the lyrics. In the singing sound synthesizer, when changing the pitch of the phoneme to be pronounced in the case of voiced sound, unvoiced sound, and voiced sound in order, the pitching period of the unvoiced sound retains the preceding voiced sound pitch data and follows. The singing sound synthesizer, wherein the change of the pitch is started from the start of the voiced sound. 3. The system is configured to store lyric data and melody data, read the lyric data in response to the reading of the melody data, generate a phoneme corresponding to the lyric data, and sing the lyrics. In the singing sound synthesizer, when a consonant, a vowel, and a consonant are to be pronounced in order to produce the vowel, the formant center frequency does not reach the formant center frequency in the case where the vowel is solely produced. The singing sound synthesizer is adapted to return to the formant center frequency of the consonant.