JP2001034299A - Sound synthesis device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reproduce vocal data in high quality. SOLUTION: This device allots musical sound data, vocal data which is compression-encoded by an analysis-synthesis encoding system provided with a code book, and the code book used at the time of compression-encoding through a base station 2, and stores them in a storage means 13. When the compression-encoded vocal data is decoded in a sound synthesis part 15, the code book peculiar to the vocal is stored in the sound synthesis part 15 beforehand and decoded. Therefore, real and high quality vocal sound can be reproduced. Moreover, since musical sound is synthesized from the musical sound data by a musical sound synthesis part 16, and the vocal data is reproduced in synchronization with this musical sound and outputted with the musical sound, the musical sound with the high quality vocal sound can be reproduced.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice synthesizing apparatus suitable for application to a mobile telephone, a portable telephone, or the like.

[0002]

2. Description of the Related Art PDC (Personal Digital Cellular telecomm) known as a digital cellular system
unified system) and a simple mobile phone system (PHS: Personal Handyphone Syste)
In m), since the frequency band that can be occupied is limited, the audio signal is transmitted after being compressed and encoded with high efficiency. As one method of the high-efficiency speech compression encoding method, an analysis-synthesis encoding method using a speech synthesis model including a sound source model and a vocal tract model is known. Furthermore, this analysis / synthesis coding method includes MPC (Multi-Puls
There are a CELP (Code Excited LPC) method for performing vector quantization using a codebook and a CELP (Code Excited LPC) method, and the CELP method has been put to practical use in a kind of digital cellular method. Furthermore, for example, 51
VSELP (Vector Sum Exiceted Linear) that can set the number of code vectors in the two codebooks to 9
Prediction) method is also in practical use. The VSELP method, by determining the sum of the ^nine gives the sign of the ± a codevector example 9, are producing codevector 512 kinds. Therefore, the VSELP method has the advantages that the memory amount required for the codebook can be reduced as compared with the CELP method, the effects of code errors are small, and the calculation amount of the synthesis filter can be reduced.

FIG. 6 shows a typical configuration of a conventional CELP encoder which can compress and encode voice information with high efficiency.
The characteristics of the voice include the pitch L and noise component of the source voice generated from the vocal cords (hereinafter, referred to as “source voice feature parameters”), the vocal tract transmission characteristics when the voice passes through the throat, and the lips. (Hereinafter collectively referred to as “vocal tract feature parameters”). That is, a voice synthesis model can be represented by a vocal cord model that generates a source voice and a vocal tract model cascaded to the vocal cord model.

[0006] The CELP encoder shown in FIG. 6 encodes input speech based on this speech synthesis model, and the long-term predictor 101 converts the input speech into, for example, 20 m.
A frame having a length of about s is cut out and processed by a filter having a characteristic reverse to that of the throat filter to obtain a source voice. The cross-correlation between this source voice and the synthesized source voice of the previous frame is calculated for each frame, and a sample shift giving the maximum value of the cross-correlation is detected as a pitch. This pitch is output for each frame as a parameter of the pitch L substantially corresponding to the fundamental frequency of the input voice. Further, the reflection coefficient analysis unit 104 obtains a reflection coefficient γ by executing FLAT (fixed point covariance lattice type algorithm) using the autocorrelation of the input voice input, and uses the reflection coefficient γ as a vocal tract feature parameter. Output every frame. Further, by supplying the reflection coefficient γ as a vocal tract feature parameter to the throat approximation filter unit 105 to which the synthesized source speech is input, a synthesized speech is output from the throat approximation filter unit 105, and the input speech is subtracted from this synthesized speech. The subtractor 106 outputs a residual signal for each frame.

Then, a process of selecting one code vector from the source sound codebook 102 so that the synthesized speech is closest to the input speech (so that the residual signal is minimized) is performed for each frame. At this time, an index I for specifying the selected code vector is output. The dimension of this code vector is made equal to the number of samples for each frame, and an index I, which is an output code by vector quantization, and a pitch L are source speech feature parameters.
When the code vector is K-dimensional and the index I is B bits, the number of bits per sample b
Is b = B / K, and the compression ratio of vector quantization can be increased as the number of dimensions K is increased. Further, the code book is composed of code vectors of 2 ^B types of typical patterns. By the way, the pitch L from the long-term predictor 101
And a code vector indicated by the index I are synthesized by the synthesizing unit 103 to generate a synthesized source speech. The synthesized source speech is filtered by the throat approximation filter unit 105 based on the reflection coefficient γ. After being processed, the synthesized speech is obtained as described above.

[0006] The original audio signal can be reproduced by providing the CELP decoder with the audio parameter comprising the index I, pitch L and reflection coefficient γ output from the encoder in this way. This CEL
FIG. 7 shows a typical configuration of the P-system decoder. In FIG. 7, the audio parameters for each frame input to the decoder are separated into respective parameters of an index I, a pitch L, and a reflection coefficient γ in a data processing unit 130, and the parameters of the pitch L are output to a pitch L generating unit 132. The parameter of the index I is distributed to the codebook 131, and the parameter of the reflection coefficient γ is distributed to the throat approximation filter 134. The code book 131 has the same contents as the source sound code book 102 in the encoder.
The contents are recorded in M (Read Only Memory).

The pitch L is determined based on the parameter of the pitch L.
The generator 132 generates a decoded signal of the audio frequency and supplies it to the source waveform reproducer 133. Source waveform reproducing unit 1
33 is supplied with the code vector data indicated by the index I read from the code book 131. By combining this data with the decoded signal having the pitch L, the source waveform reproducing unit 133 Is played. From the source waveform reproducing unit 133, a waveform similar to the waveform generated by the vibration of the human vocal cords is output as a composite source waveform. This synthesized source waveform is filtered by a throat approximation filter 134 in which the filter coefficient is controlled by the parameter of the reflection coefficient γ, to obtain a synthesized voice. The throat approximation filter 134 reproduces the transfer function of the human throat and mouth,
The reflection coefficient γ supplied from 0 is stored and supplied to each filter when necessary. The synthesized speech output from the throat approximation filter 134 is supplied to a spectral filter 135 to remove unnaturalness as speech.

Since the data amount of the reflection coefficient γ is large, it is conceivable to transmit the reflection coefficient γ by reducing the number of bits by compression coding such as vector quantization. For example, when the reflection coefficient γ is vector-quantized and transmitted, the encoder is provided with a reflection coefficient codebook including a plurality of representative vectors, and the reflection coefficient γ output from the reflection coefficient analysis unit 104 is used as one of the representative vectors. And the reflection coefficient index indicating the representative vector is transmitted. Further, the decoder may be provided with a reflection coefficient codebook having the same contents as the reflection coefficient codebook provided in the encoder, and may read out the reflection coefficient γ indicated by the transmitted reflection coefficient index.

[0009]

As a speech coding method in a digital cellular system, CEL is generally used.
Although the P-type high-efficiency voice compression system is adopted, the voice parameters to be transmitted are limited to the minimum parameter amount that can identify the identity and identify the conversation content due to the limited transmission capacity. Have been. In particular, V
In the SELP system, the number of code vectors stored in a code book is limited. For this reason, there is a problem that the sound quality when the transmitted sound data is decoded is not good.

Since the characteristics of the vocal cords and vocal tracts vary from person to person, it is necessary to change the parameters of the source voice characteristic parameters and the vocal tract characteristic parameters according to the speaker. Parameter amount, and the parameter amount cannot be changed according to the speaker. For this reason, depending on the speaker, when the transmitted voice data is decoded, only a sound quality that can barely recognize the other party can be obtained. Since only such sound quality can be obtained, the musical sound signal or the vocal sound for appreciation is compressed and encoded by a CELP encoder and transmitted, and the musical sound or vocal sound when decoding the transmitted data is used for appreciation. There was a problem that it was unbearable.

SUMMARY OF THE INVENTION In view of the above problems, the present invention can reproduce vocal voices with high sound quality or distribute vocal voices with call quality when music data or vocal data that has been subjected to high-efficiency voice compression encoding is distributed. It is an object of the present invention to provide a voice synthesizing apparatus which can reproduce a sound in synchronism with musical sound data.

[0012]

In order to achieve the above object, a speech synthesizer according to the present invention comprises a distributed vocal compression-encoded by an analysis-synthesis encoding system for performing vector quantization using a codebook. Storage means for storing at least data and a codebook used at the time of compression encoding, and the vocal data read from the storage means using the codebook stored in the storage means And a voice synthesizing means for synthesizing vocal voice by decoding the vocal voice.

[0013] In the above-mentioned voice synthesizing apparatus of the present invention, the storage means further stores a distributed tone data, and further comprises a tone synthesis means for generating a tone from the tone data read from the storage means. The musical sound generated by the musical sound synthesizing means may be synchronized with the vocal sound. Further, in the above-mentioned speech synthesizing apparatus of the present invention, a second codebook is stored in the speech synthesizing means in advance, and the speech signal compressed and encoded is decoded using the second codebook. You may. Still further, in the above-described voice synthesizing apparatus of the present invention, a plurality of systems of vocal data compression-encoded by an analysis-synthesis coding method for performing vector quantization using a codebook unique to each vocal, A code book is distributed and stored in the storage means,
The voice synthesizing unit decodes the vocal data of the plurality of systems in parallel using each of the vocal-specific codebooks stored in the storage unit, whereby the musical sound synthesizing unit synthesizes the vocal data. A plurality of vocal sounds synchronized with the reproduced musical sound may be synthesized.

According to the present invention, the vocal data compression-encoded by the analysis-synthesis encoding method in which vector quantization is performed using the codebook, and the codebook used at the time of compression-encoding are used. Has been delivered. Thus, when decoding the compression-encoded vocal data, the vocal data can be decoded with reference to a codebook specific to the vocal, so that high-quality real vocal sound can be reproduced. When music data is also distributed, it is possible to synthesize a music sound from the transmitted music data, reproduce the vocal data in synchronization with the music sound, and output the vocal data together with the music sound. The accompanying musical sound can be reproduced. Further, since the call signal is decoded by using a codebook stored in advance, even if the codebook for vocal data is distributed, the sound quality does not deteriorate when the call signal is decoded. Furthermore, when the distributed vocal data is composed of multiple systems of vocal data, the multiple systems of vocal data can be decoded with high quality by distributing together multiple systems of codebooks. .

[0015]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a diagram for explaining an embodiment in which a speech synthesizer of the present invention is applied to a portable telephone.
Shown in In FIG. 1, 1 is a mobile phone, and 2 is a base station that manages each wireless zone. In a cellular system of a mobile phone, a small zone system is generally adopted, and a large number of wireless zones are arranged in a service area. The base station 2 manages each of these wireless zones. When the mobile phone 1 as a mobile station talks with a general telephone, the mobile phone 1 is connected to the exchange via the base station 2 and communicates therewith. Being connected to the general telephone network.

The mobile phone 1 has a generally retractable antenna 10, which is connected to a transceiver unit 11. The transceiver unit 11 includes the antenna 1
The signal received at 0 is demodulated and the signal to be transmitted is modulated and supplied to the antenna 10. Phone Function 1
Reference numeral 2 denotes a CELP-based encoder and decoder (speech synthesis unit) 15 for controlling the mobile phone 1 to function as a phone when talking with another phone and for coping with high-efficiency compression of speech. are doing. During a call, an audio signal input from the microphone 19 is compression-encoded by the encoder of the telephone function unit 12, modulated by the transceiver unit 11, and transmitted from the antenna 10. Also, antenna 1
0 is demodulated by the transmitter / receiver unit 11 and is transmitted to the voice synthesizer 1 of the telephone function unit 12.
The signal is decoded into an original audio signal at 5 and output from an output unit 20 including a speaker or the like. As described above, signals are exchanged between the transceiver unit 11 and the telephone function unit 12 during a call.

The storage means 13 stores MIDI-formatted tone data distributed as described later, vocal data which has been subjected to high-efficiency voice compression encoding by a CELP encoder having a codebook, and which encodes the vocal data. This is storage means for storing the code book used at that time. The MIDI decoder 14 stores the storage unit 13
This is a MIDI decoder that decodes the musical tone data in the MIDI format stored in the MIDI format and supplies it to the musical tone synthesizer 16 as performance data. Further, the vocal-specific code book stored in the storage unit 13 is stored in the voice synthesizing unit 15 in the telephone function unit 12 in the memory (RA) of the code book unit.
M), the vocal data read from the storage unit 13 is decoded by the voice synthesis unit 15 using a vocal-specific codebook, and vocal voice is synthesized. At this time, the musical sound synthesized by the musical sound synthesizing section 16 is regarded as an accompaniment, and the vocal sound is synthesized by the audio synthesizing section 15 in synchronization with the accompaniment. The accompaniment sound output from the musical sound synthesizer 16 and the vocal sound output from the voice synthesizer 15 are mixed by the mixer 17 and
After being amplified by, the signal is output from the output unit 20 and emitted.

Next, the operation of the configuration shown in FIG. 1 will be described. When the portable telephone 1 is connected to a server connected to the exchange via a telephone line or the like, a desired database is stored in the server. Music data and vocal data can be downloaded to the mobile phone 1 via the exchange and the base station 2. This vocal data includes
A codebook specific to the vocals used when encoding was attached. At the time of downloading, the mobile phone 1 is connected to the base station 2 via a wireless line, and for example, the above-mentioned musical sound data, vocal data and codebook data transmitted from the antenna 2a of the base station 2 are The signal is received by the antenna 10 and demodulated by the transceiver unit 11.
At this time, since the mobile phone 1 is set to the communication mode connected to the server, the tone data demodulated by the transceiver unit 11 and the vocal data and codebook data are stored in the storage unit 13. Downloaded by

When the mobile phone 1 is in a communication mode for talking with another phone, a communication signal from the other phone transmitted from the antenna 2a of the base station 2 is received by the antenna 10 and The signal is demodulated by the transceiver unit 11. At this time, since the mobile phone 1 is set to the communication mode for talking with another telephone, the call signal demodulated by the transceiver unit 11 is supplied to the telephone function unit 12 and is used by the voice synthesis unit 15 for communication. , And is output as sound from the output unit 20 including a speaker or the like. Further, the voice input via the microphone 19 is encoded by an encoder in the telephone function unit 12, further modulated in the transceiver unit 11, and transmitted from the antenna 10. This transmission signal is transmitted to antenna 2
The data is received by the base station 2 via a, and is transmitted to another telephone via the exchange connected to the base station 2, so that a two-way communication is performed.

When the telephone number of the mobile phone 1 is transmitted in order to call the mobile phone 1 from another telephone, the exchange detects the base station 2 in which the mobile phone 1 is registered, and detects the detected base station. 2 that the call has been received by the mobile phone 1. The mobile phone 1 receives this, and emits a ringtone notifying that the user has received a call in the telephone function unit 12. At this time, the vocal data and the MIDI-formatted musical sound data set as ringtones in advance are read out from the storage means 13, and the voice synthesizing section 15 and the MIDI decoder 14 of the telephone function section 12 are read out.
To supply. The MIDI decoder 14 decodes the tone data and supplies the decoded tone data to the tone synthesizer 16. The tone synthesizer 16 generates a tone signal corresponding to the supplied tone data and supplies the tone signal to the mixer 17.

The vocal-specific code book stored in the storage means is set in the speech synthesizer 15 prior to decoding. Then, the voice synthesis unit 15 decodes the CELP-based compression-coded vocal data using a vocal-specific codebook.
The decoding of the vocal data in the voice synthesizer 15 is executed in synchronization with the tone signal generated by the tone synthesizer 16. Next, the vocal signal decoded by the voice synthesizer 15 and the tone signal generated by the tone synthesizer 16 are mixed in the mixer 17, amplified by the amplifier 18, and emitted from the output unit 20. That is, a vocal sound is emitted with a musical sound as an accompaniment. As a result, when an incoming call arrives at the mobile phone 1, a vocal sound accompanied by accompaniment (hereinafter referred to as “ringing melody”) is emitted from the output unit 20 as a ringing sound. Since the vocal sound in the ringtone melody is decoded using a codebook specific to the distributed vocal sound, a high-quality real vocal sound can be obtained from the sound synthesis unit 15.

The tone signal is generated by the tone synthesizer 16 and the vocal signal is decoded by the voice synthesizer 15 not only at the time of incoming call but also at the time of hold.
At this time, the vocal data and the MIDI-formatted musical sound data set in advance as the holding sounds are read out from the storage means 13 and supplied to the voice synthesizing section 15 and the MIDI decoder 14 of the telephone function section 12. MIDI
After decoding the tone data, the decoder 14 supplies the tone data to the tone synthesizer 16, and the tone synthesizer 16 generates a tone signal corresponding to the supplied tone data. Voice synthesis unit 15
Decodes the supplied vocal data using a vocal-specific codebook, and the decoded vocal sound is mixed with the tone signal in the mixer 17, amplified by the amplifier 18, and emitted from the output unit 20. You. Since the vocal sound in the hold sound is decoded by using a codebook specific to the distributed vocal sound, a high-quality real vocal sound can be obtained from the voice synthesizer 15.

The vocal data and musical sound data distributed and stored in the storage means 13 can be reproduced for appreciation. In this case, the mobile phone 1
Is selected as the vocal data to be reproduced in the tone reproduction mode. At this time, the title of the vocal data stored in the storage means 13 is displayed and selected on a display unit (not shown). Thereafter, when the reproduction is instructed, the selected vocal data and musical sound data are read out from the storage means 13 and the voice synthesizing section 1 of the telephone function section 12 is read.
5 and the MIDI decoder 14. MIDI
After decoding the tone data, the decoder 14 supplies the tone data to the tone synthesizer 16, and the tone synthesizer 16 generates a tone signal corresponding to the supplied tone data. Voice synthesis unit 15
Decodes the supplied vocal data using a vocal-specific codebook, and the decoded vocal sound is mixed with the tone signal in the mixer 17, amplified by the amplifier 18, and emitted from the output unit 20. You. As a result, the vocal sound reproduced for appreciation is decoded using a codebook specific to the distributed vocal sound, so that a high-quality real vocal sound can be obtained from the sound synthesizer 15. .

Here, a vocal-specific code book will be described. Assuming that the vocal sound is to be compression-encoded by the encoder having the configuration shown in FIG. 6, a code vector having the smallest residual signal is selected from the source sound codebook 102. However, since the source codebook 102 includes a limited number of codevectors, there may be cases where there is no codevector with which the residual signal is very small. Therefore, a vocal sound is input as a training signal so as to be learned offline. In this learning, the code vector of the source codebook 102 is rewritten so that the residual signal is always small. As a result, the source sound codebook 102 adapted to the input vocal sound is generated. The source sound codebook 102 adapted to the vocal sound is defined as a vocal-specific codebook in this specification.

By the way, in the configuration shown in FIG. 1, the tone synthesis unit 16 and the voice synthesis unit 15 may be arranged in cascade as shown in FIG. In the configuration shown in FIG.
The output of the MIDI decoder 14 is supplied to a tone synthesizer 16 to generate a tone signal based on the tone data. The tone signal is supplied to the speech synthesizer 15 together with the vocal data passed through the tone synthesizer 16, and the vocal data is output. Is decoded using a distributed codebook specific to vocals, mixed with a musical sound signal, and output. Although the mixing signal is not shown, the amplification unit 18
And is output from the output unit 20.

It should be noted that a specific MIDI channel may be allocated to the vocal data and distributed together with the musical sound data in the MIDI format, or the vocal data may be distributed in the CELP format and the musical sound data in the MIDI format. As an example,
FIG. 3 shows a data format when the vocal data is inserted into MIDI sound data and transmitted. As shown in FIG. 3, flags are inserted before and after the audio parameter which is the vocal data to indicate the start and end of the audio parameter in the MIDI data. The voice parameter includes a pitch L, an index I, and a reflection coefficient γ, and a flag indicating the parameter is added to each parameter as a header as shown in the figure. At this time, since the data amount of the reflection coefficient γ is large, the bit may be reduced by vector quantization. When performing vector quantization of the reflection coefficient γ, the encoder is provided with a reflection coefficient codebook including a plurality of representative vectors, and the reflection coefficient γ output from the reflection coefficient analysis unit is replaced with one of the representative vectors. The reflection coefficient index indicating the representative vector is transmitted as a parameter. Also, the decoder-side speech synthesizer is provided with a reflection coefficient codebook having the same contents by distributing the reflection coefficient codebook provided in the encoder together with the vocal data, and is designated by the transmitted reflection coefficient index. If the reflection coefficient γ is read, the reflection coefficient γ can be reproduced. As a result, the data amount of the voice parameter can be reduced.

The vocal data and the codebook data unique to the vocal may be supplied collectively at the time of initial download. Is also good. Furthermore, when the required vocal voice sound quality is about the same as the telephone sound quality, when it is desired to have a better sound quality than the telephone sound quality, and when it is desired to have a high-fidelity sound quality, the amount of the voice parameter of the vocal data to be distributed is determined. You may make it select suitably and distribute.

Next, FIG. 4 shows an example of the configuration of the speech synthesizer according to the embodiment of the present invention. The configuration shown in FIG.
It can be used as the voice synthesizing unit 15 in the mobile phone 1 shown in FIG. 1 and can decode vocal data, but can also decode voice data during a call. In FIG. 4, the voice parameters read from the storage unit 13 are stored in the data processing unit 3.
At 0, the parameters of the index I, the pitch L, and the reflection coefficient γ are separated, and the parameter of the pitch L is approximated to the pitch L generator 32, the parameter of the index I is approximated to the codebook 31, and the parameter of the reflection coefficient γ is approximated by the throat. Filtered to filter 34. Note that the ROM of the codebook 31 stores a codebook for a call that is common to the mobile phone 1 used for decoding voice data, and the RAM of the codebook 31 stores a vocal-specific code used for decoding vocal data. Stored when the codebook is distributed.

When decoding, the supplied pitch L
The decoded signal of the frequency of the sound is generated from the pitch L generating unit 32 based on the above parameters, and is supplied to the source waveform reproducing unit 33. The source waveform reproducing unit 33 is supplied with the data of the code vector indicated by the index I read from the codebook 31, and synthesizes this data with the decoded signal of the pitch L to obtain the source waveform reproducing unit 33. At, the combined source waveform is reproduced. From the source waveform reproducing unit 33, a waveform similar to the waveform generated by the vibration of the human vocal cords is output as a composite source waveform. This synthesized source waveform is filtered by a throat approximation filter 34 in which the filter coefficient is controlled by the parameter of the reflection coefficient γ to be a synthesized voice. Throat approximation filter 34
Reproduces the transfer function of the human throat and mouth, stores the reflection coefficient γ supplied from the data processing unit 30 in advance, and supplies it to each filter when necessary. The synthesized speech output from the throat approximation filter 34 is supplied to a spectral filter 35 to remove unnaturalness as speech.

In this case, when decoding the vocal data, since the code vector is read out from the vocal-specific code book stored in the RAM by the index I, a real high-quality vocal sound is output from the spectral filter 35. Will be done. Also, when decoding the voice data of a call, the index I
As a result, the code vector is read from the common code book stored in the ROM, so that normal-quality sound is output from the spectral filter 35. The voice data of the call is supplied from the transceiver unit 11 directly to the data processing unit 30 without passing through the storage unit 13, and is decoded in real time.

The reflection coefficient γ in the vocal data
Is vector-quantized and the reflection coefficient index is used as a parameter, the code book for the reflection coefficient used for vector-quantizing the reflection coefficient γ is also distributed along with the vocal data. Then, the reflection coefficient codebook delivered to the speech synthesizer is stored, and the reflection coefficient γ indicated by the reflection coefficient index separated by the data processing unit 30 is read out from the reflection coefficient codebook to obtain the reflection coefficient γ. Reproduce. Reconstructed reflection coefficient γ
Is supplied to the throat approximation filter 34 so that the above-described decoding operation is performed.

Elements that determine the sound quality include the number of dimensions of the code vector and the number of types of the code vector at the time of vector quantization, and the order of the throat approximation filter, that is, the number of parameters of the reflection coefficient γ. By increasing the number of types of code vectors in the codebook (the number of bits of the index I is increased) and increasing the number of parameters of the reflection coefficient γ, more realistic high-quality vocal sound can be reproduced. In addition, since the above-mentioned parameters differ depending on the nature of voice, if high-quality reproduction can be performed when the number of code vectors is increased, the index I is increased, and when the order of the throat approximation filter is increased, the index is increased. If the quality can be reproduced, the memory for storing the index I and the reflection coefficient γ may be shared and the memory may be used efficiently so as to increase the number of parameters of the reflection coefficient γ.

By the way, when a call is made with another telephone, a code book specific to one's voice is transmitted prior to the call to be downloaded to the other telephone, and the other telephone transmitted from the other telephone is transmitted to the own telephone. Download a codebook specific to your voice. That is, each telephone has a first codebook for vector quantizing its own voice and a second codebook for decoding the other party's vector quantized audio data. In this way, each audio data can be decoded using a codebook specific to the audio, so that a real high-quality call can be made. Instead of transmitting a codebook specific to one's own voice, a codebook specific to a famous singer distributed and stored in a RAM may be selected and transmitted, and downloaded to a telephone of the other party. According to this, the user's voice can be heard as a famous singer's voice on the other party's telephone, so that it is possible to enjoy a playful call.

Next, FIG. 5 shows another configuration example of the voice synthesizing apparatus according to the embodiment of the present invention in the mobile phone 1 shown in FIG. The configuration shown in FIG. 5 is a configuration of a speech synthesis unit for two vocals, and has a parallel configuration so that respective vocal data can be decoded in parallel and simultaneously. In FIG. 5, the voice parameters for two vocals read from the storage unit 13 are stored in the data processing unit 30.
, Index Ia, Ib, pitch La, Lb
And the parameters of the pitch La are separated into parameters of the reflection coefficient γa and γb.
The parameter of the pitch Lb is the pitch L generating unit 32b
The parameters of the indexes Ia and Ib are distributed to the codebook 31, the parameters of the reflection coefficient γa are distributed to the approximate throat filter 34a, and the parameters of the reflection coefficient γb are distributed to the approximate throat filter 34b. The code book 31
The ROM stores a codebook for a call that is common to the mobile phone 1 used for decoding voice data, and the RAM of the codebook 31 stores a code for the two vocals used for decoding the vocal data for two vocals. Codebook is stored when it is distributed.

When decoding 2 vocal data,
The pitch L is determined based on the supplied parameters of the pitch La.
The generator 32a generates a decoded signal of the first vocal frequency and supplies it to the source waveform reproducer 33a. At the same time, a decoded signal of the second vocal frequency is generated from the pitch L generating unit 32b based on the supplied parameter of the pitch Lb, and is supplied to the source waveform reproducing unit 33b. The source waveform reproducing unit 33a reads and supplies the code vector data indicated by the index Ia from the codebook unique to the first vocal stored in the RAM of the codebook 31, and supplies this data to the pitch. By combining with the La decoded signal, the combined source waveform of the first vocal is reproduced in the source waveform reproducing unit 33a. At the same time, the source waveform reproducing unit 33b
The data of the code vector indicated by the index Ib is read out from the codebook specific to the second vocal stored in M and supplied, and this data is sent to the pitch L
b by combining with the decoded signal of the
At b, the composite source waveform of the second vocal is reproduced.

The first vocal synthesized source waveform is filtered by a throat approximation filter 34a whose filter coefficient is controlled by the parameter of the reflection coefficient γa.
The synthesized voice of the first vocal is used. At the same time, the synthesized source waveform of the second vocal is filtered by a throat approximation filter 34b whose filter coefficient is controlled by the parameter of the reflection coefficient γb, to obtain a synthesized voice of the second vocal. The throat approximation filter 34a (34b) reproduces the transfer function of the human throat and mouth, stores the reflection coefficient γa (γb) supplied from the data processing unit 30 in advance, and stores Supplying to filter. The synthesized voice of the first vocal output from the throat approximation filter 34a is supplied to the spectral filter 35a,
The unnaturalness of the voice of the first vocal is removed. At the same time, the synthesized voice of the second vocal output from the throat approximation filter 34b is
b to remove the unnaturalness of the second vocal sound.

As described above, when decoding the first and second vocal data, respectively, the indexes Ia and I
Since the code vector is read from the vocal-specific code book for each of the two vocals stored in the RAM by b, real high-quality two-vocal sounds are output from the spectral filters 35a and 35b, respectively. When decoding the voice data of the call, the code vector is read out from the codebook for the call stored in the ROM by the index I and decoded using only the configuration of one of the series arranged in parallel. . The quality of the sound at this time is such that the sound of normal quality is output from the spectral filter 35a (35b). The voice data of the call is transmitted directly from the transceiver unit 11 to the data processing unit 3 without passing through the storage unit 13.
0 is supplied with audio data and decoded in real time.

When the reflection coefficients γa and γb in the two-vocal data are vector-quantized and the respective reflection coefficient indexes are used as parameters, the respective reflection coefficients used for vector-quantizing the reflection coefficients γa and γb are used. The coefficient code book is also distributed along with the vocal data. Then, the respective codebooks for reflection coefficients distributed to the speech synthesizer are stored, and the reflection coefficients γa and γb indicated by the respective reflection coefficient indexes separated by the data processing unit 30 are respectively used for the corresponding reflection coefficients. Read from the code book and reflectivity γ
a and γb are reproduced. Reconstructed reflection coefficients γa, γb
Is supplied to the throat approximation filters 34a and 34b so that the above-described decoding operation is performed. At this time,
The codebook for each reflection coefficient for 2 vocals is
It is a codebook for reflection coefficient applied to each vocal.

When two vocals are set, the first vocal can be set as the lead vocal and the second vocal can be set as the back chorus. The number of vocals is not limited to two, but may be three or more.
In this case, a series configuration corresponding to the number of vocals is arranged in parallel. Instead of distributing a codebook unique to each vocal, a common codebook for chorus may be distributed and a plurality of vocals may be decoded using the codebook for chorus.

The tone generator of the tone synthesizer 16 incorporated in the mobile phone 1 can be an FM tone generator, a waveform memory tone generator (PCM tone generator), a physical model tone generator, or the like. May be a hardware tone generator using a DSP or the like, or a software tone generator that executes a tone generator program. Further, although the musical sound is output as the accompaniment of the vocal sound, it is also possible to reproduce only the vocal sound without reproducing the musical sound. Furthermore, although it has been described that the music data is distributed as MIDI data, the present invention is not limited to this, and the music data may be distributed in another data format.

[0041]

As described above, according to the present invention, the vocal data compression-encoded by the analysis-synthesis encoding method in which vector quantization is performed using the codebook, and the codebook used in the compression encoding And has been delivered. Thus, when decoding the compression-encoded vocal data, the vocal data can be decoded with reference to a codebook specific to the vocal, so that high-quality real vocal sound can be reproduced. When music data is also distributed, it is possible to synthesize a music sound from the transmitted music data, reproduce the vocal data in synchronization with the music sound, and output the vocal data together with the music sound. The accompanying musical sound can be reproduced. Further, since the call signal is decoded by using a codebook stored in advance, even if the codebook for vocal data is distributed, the sound quality does not deteriorate when the call signal is decoded. Furthermore, when the distributed vocal data is composed of multiple systems of vocal data, the multiple systems of vocal data can be decoded with high quality by distributing together multiple systems of codebooks. .

[Brief description of the drawings]

FIG. 1 is a diagram for describing an embodiment when a speech synthesizer of the present invention is applied to a mobile phone.

FIG. 2 is a diagram showing a modification of the configuration when the speech synthesis device of the present invention is applied to a mobile phone.

FIG. 3 is a diagram showing an example of a data format when vocal data is inserted into MIDI-formatted musical sound data and transmitted in the voice synthesizer according to the embodiment of the present invention;

FIG. 4 is a diagram illustrating an example of a configuration of a speech synthesizer according to the embodiment of the present invention;

FIG. 5 is a diagram showing another configuration of the speech synthesizer according to the embodiment of the present invention;

FIG. 6 is a diagram illustrating a typical configuration of a conventional CELP encoder that can compress and encode audio information with high efficiency.

FIG. 7 is a diagram illustrating a typical configuration of a conventional CELP decoder that decodes audio information that has been compressed and encoded with high efficiency.

[Explanation of symbols]

1 mobile phone, 2 base station, 2a antenna, 10
Antenna, 11 transceiver unit, 12 telephone function unit, 13
Storage means, 14 MIDI decoder, 15 voice synthesizer, 16 tone synthesizer, 17 mixer, 18 amplifier,
19 microphones, 20 output units, 30 data processing units, 3
1 codebook, 32, 32a, 32b pitch L generating section, 33, 33a, 33b source waveform reproducing section, 34, 34
a, 34b throat approximation filter, 35, 35a, 35b
Spectral filter, 101 long-term predictor, 102
Source codebook, 103 synthesis unit, 104 reflection coefficient analysis unit, 105 throat approximation filter unit, 106 subtractor, 130 data processing unit, 131 codebook, 1
32 pitch L generator, 133 source waveform reproducer, 134
Throat approximation filter, 135 spectral filter

Claims (4)

[Claims] At least, distributed vocal data compressed and encoded by an analysis-synthesis encoding method for performing vector quantization using a codebook and a codebook used at the time of compression-encoding are stored. Storage means, and speech synthesis means for synthesizing vocal sound by decoding the vocal data read from the storage means using the codebook stored in the storage means. Characteristic speech synthesizer. 2. The storage means also stores distributed tone data, further comprising a tone synthesis means for generating a tone from the tone data read from the storage means, wherein the tone generated by the tone synthesis means is provided. The voice synthesizing apparatus according to claim 1, wherein the vocal voice and the vocal voice are synchronized. 3. The speech synthesizer according to claim 2, wherein a second codebook is stored in advance, and the second codebook is used to decode a compression-coded speech signal. The speech synthesizer according to claim 1. 4. A vocal data of a plurality of systems compressed and encoded by an analysis-synthesis coding method for performing vector quantization using a codebook unique to each vocal, and the codebook unique to each vocal are distributed. Using the respective vocal-specific codebooks stored in the storage means to decode the vocal data of the plurality of systems in parallel in the speech synthesis means. 2. The voice synthesizing apparatus according to claim 1, wherein a plurality of vocal voices are synthesized.