JPS5949600B2 - Speech synthesis device for melody sound synthesis - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention is used by being connected to a replaceable control IC that stores voice messages for various purposes such as for clocks, alarms, and in-tab-on applications. In addition, it is possible to play various melodic sounds such as chime sounds for intab-ons and music box sounds for clocks at the same time as human voice messages, or to synthesize chords by playing multiple monotone scales with different scales at the same time. The present invention relates to a speech synthesizer that can also be used to synthesize melody sounds.

In general, parameters representing the characteristics of a sound include an amplitude parameter representing the magnitude of the sound, a pitch parameter representing the pitch of the sound, that is, the fundamental period, and a vector parameter representing the timbre of the sound, that is, the spectral distribution.

These various parameters are collectively referred to as feature parameters because they represent the characteristics of the voice, but since a voice signal can normally be regarded as a nearly stationary signal over a short period of 10 msec to 30 msec, conventionally this period has been A speech synthesis device has been developed that extracts one set of feature parameters from one frame (data update interval) and updates the data for each frame.
By the way, in such a speech synthesis device, when extracting a set of feature parameters from an original speech signal, first, the feature parameters related to the spectrum are extracted, and then the remaining waveform is normalized to extract the feature parameters related to the amplitude.
A process is adopted in which the residual waveform that remains at the end is input to an autocorrelator to extract characteristic parameters related to the fundamental period. As is clear from this principle, in a speech analysis system, it is not possible to extract multiple fundamental periods from one frame at the same time, and therefore only one fundamental period and its fundamental period can be extracted within one frame. only one set of amplitude and svector parameters corresponding to the period,
It is not possible to analyze complex sounds that are made up of many different sounds. Therefore, even in a speech synthesis system that resynthesizes the feature parameters analyzed and extracted in a speech analysis system, there is a drawback that only one type of sound can be reproduced within one frame. Of course, if such a speech synthesis device is used only for synthesizing human speech messages, there is no problem even if only one type of sound can be reproduced within one frame as described above. When synthesizing onomatopoeic sounds other than vocal sounds, such as chime sounds for intab-on and various melody sounds such as music box sounds for clocks, you may want to compose a chord by combining several types of single notes with different scales. In addition, in advanced home appliances that also have voice recognition functions, there are cases where it is desired to simultaneously play a human voice message and a melody sound, so conventional voice synthesis, which can only synthesize one type of sound in one frame, The device had the disadvantage of being non-luminous, which is not suitable for applications that require the synthesis of such melodic sounds.

Therefore, in order to perform pseudo-chord synthesis, the present inventors have conventionally developed a method in which feature parameters of a plurality of single notes extracted separately in a speech analysis system are played back while being alternately switched within one frame of the speech synthesis system. However, when such a method is used, the reproduced scale sounds less like a chord and more like a single note with intermediate scales between each note, making it sound less like the original chord. I couldn't get the beautiful sound that it had.

If a similar method is used to alternately switch between human voice messages and melody sounds, the melody sounds and voice messages can be heard at the same time, but the quality of the reproduced sound is extremely difficult to hear. However, it had the drawback that it could not be put to practical use. The present invention has been made in order to eliminate such drawbacks of the conventional example,
By reproducing multiple types of sounds that have been separately analyzed and extracted in the speech analysis system at the same time in the speech synthesis system, it is possible to reproduce the beautiful sound of pure chords, or to broadcast human voice messages at the same time as melody sounds. It is an object of the present invention to provide a speech synthesis device which can also be used for melody speech synthesis.

The configuration of the present invention will be described below with reference to the illustrated embodiment. As shown in FIGS. , a latch memory 1 that temporarily stores characteristic parameters related to the fundamental period.
A plurality of latch memories 13a, 13b are provided, and each latch memory 13a, 1
A multiplexer 34 is provided to sequentially switch and connect the output of the sound source forming means 30 to the sound source forming means 30, and the output of the sound source forming means 30 is passed through the digital filter 23 which is controlled by each characteristic parameter related to amplitude and spectrum, thereby producing the audible sound. In a speech synthesis device that also serves as melody sound synthesis and is configured to resynthesize signals, a demultiplexer 35 that operates in synchronization with the multiplexer 34 is connected to the output of the digital filter 23, and a data latch is connected to each output of the demultiplexer 35. A mixing circuit connects a plurality of DA converters 37a, 37b via memories 36a, 36b, and outputs the output of the DA converter 37b, which is connected to the digital filter 23 at the end by a demultiplexer 35 among the DA converters 37a, 37b. 38
The outputs of the other DA converters 37a are input to the mixing circuit 38 via the respective delay circuits 39, and the delay time of each delay circuit 39 is adjusted to all the DA converters 36a, 3.
7b is set to be input to the mixing circuit 38 at the same time.

The embodiment shown in FIG. 3 is for chord synthesis, and the fundamental period of the reproduced sound can be changed according to pitch parameters stored in latch memory 13a and latch memory 13b. Since only one set of parameters is stored in the parameter stack 22a, it is possible to synthesize a chord by simultaneously playing back two single notes that have exactly the same timbre and volume but differ only in scale. On the other hand, the embodiment shown in FIG. 4 shows an embodiment in which a human voice message and a melody sound can be played simultaneously. In this embodiment, when the pitch parameter related to the fundamental period is switched, the amplitude is changed accordingly. Parameters related to spectra can also be changed by switching the parameter stacks 22a and 22b, making it possible to simultaneously reproduce completely different types of sounds. The embodiment shown in FIG. 4 can also be used for chord synthesis; in this case, the number of latch memories 13a and 13b for storing the fundamental period is two, so it is different from the embodiment shown in FIG. Similarly, a chord consisting of two different single notes is synthesized. Of course, the latch memory 13 for basic period storage.
If a larger number of melody tones are provided, it is possible to simultaneously reproduce melody tones with different pitches equal to the number of latch memories 13, thereby obtaining a richly varied chord consisting of a set of a large number of single tones. In the embodiments shown in FIGS. 3 and 4, a demultiplexer 35 that operates in synchronization with a multiplexer 34 for switching the fundamental period is connected to the output side of the digital filter 23, and each output of the demultiplexer 35 has data. Latch memories 36a and 36b are connected to each other, and when data 1 is obtained from one output 35a of the demultiplexer 35 as shown in FIG.
6a, the data is held until the next data 1 is obtained.

Therefore, as soon as the demultiplexer 35 outputs data 1, the next sound source signal is output from the sound source forming means 30, and the digital filter 23 calculates data 2.
is calculated, and the other output 35b of the demultiplexer 35
Data 2 is input to the data latch memory 36b from the data latch memory 36b. In this case as well, data 2 is held by the data latch memory 36b until the next data 2 is obtained, so as soon as data 2 is held, the next sound source signal is output from the sound source forming means 30, and the digital filter 23 outputs the next sound source signal. After the calculation, data 1 is obtained again. Through the above process, the fifth
As shown in Figure a, data 1 and data 2 are obtained, but these timings are determined by the demultiplexer 35.
Since there is a shift of half a cycle, data 1 is delayed by a half cycle as shown in FIG. 5b. In this way, the two types of sounds synthesized in the mixing circuit 38 are not reproduced by pseudo-switching at high speed as in the past, but are completely continuous two types of sounds at the same time. It will be played, synthesize pure chords,
Or, it has become possible to reproduce a human voice message and a melody sound at exactly the same time. The embodiments shown in FIGS. 3 and 4 are of course also capable of synthesizing only human voice messages, or only synthesizing melody sounds consisting of only single tones. In this case, only the original latch memory 13a and the original parameter stack 22a are operated, and the other auxiliary latch memory 13b and parameter stack 22b are not used.

Such control is performed by the parameter code detection circuit 28 connected to the input terminal 1 of the speech synthesis LSI. In the embodiment shown in FIG. 3, 3a is a shift register for reading data, and is connected to various control LSIs 3l for use in combination with the voice synthesis LSI, such as for clocks, interrupt-ons, alarms, etc. Control L
A configuration is adopted in which a set of characteristic parameters relating to the amplitude, fundamental period, and spectrum of a voice or melody sound are serially input for each frame from the data storage unit 32 of the SI 3l, and then the data is transferred to the latch memories 13a, 13b, etc. This facilitates the connection between the speech synthesis LSI and its control LSI 3l.

In this embodiment, a sub-shift register 3b is connected in parallel to the data reading shift register 3a via a register switching circuit 2, and a parameter code detection circuit 28 connected to the input terminal 1 when synthesizing a complex sound. operates the register switching circuit 2, and in addition to the original pitch parameter, sub-pitch parameters required for compound sound synthesis are input in series to the sub-shift register 3b connected in parallel to the original shift register 3a. That's what I do. Therefore, when synthesizing human voice messages or synthesizing melody sounds consisting only of monophonic scales, neither the sub-shift register 3b nor the auxiliary latch memory 13b is in operation, and the original shift is not performed. Only the register 3a operates to read a set of feature parameters and transfer them to the latch memory 13a, but when synthesizing a complex sound such as a chord, only the pitch parameters related to the fundamental period need to be set in multiple numbers. This is what I am trying to load. The overall configuration of the speech synthesis device of the present invention will be described in further detail below.

In Fig. 3, 12 is an interpolation calculation circuit, and when the data is updated every frame, if the characteristic parameters change discontinuously at the connection point between each frame, the audio signal will be distorted and the clarity will deteriorate. Since it is easy to do so, approximately linear interpolation is performed at eight points within one frame so that the feature parameters change smoothly when updating data. Of course, when synthesizing melody sounds,
In order to accentuate the synthesized sound and produce a crisp sound, the interpolation calculation is stopped by the operation of the parameter code detection circuit 28 and the interpolation control signal generation circuit 33. The characteristic parameters of the voice and melody sound are transferred from the data storage section 32 of the control LSI 3l connected to the input terminal 1 to the shift register 3 via the register switching circuit 2.
A is stored in series. By the way, although the data read into the shift register 3a in this manner represents the characteristic parameters, they are not the characteristic parameters themselves, but are address signals of the reproduction ROM10 that stores the characteristic parameters.

Moreover, the address signal merely indicates a relative address within the reproduction ROMIO. Therefore, in order to reproduce the actual feature parameters from the read data, the first address stored in the index ROM 5 is extracted by the function of the address counter 4, and this first address is sent out from the reproduction control circuit 6. shift clock 41
Accordingly, an absolute address is created by adding it to the relative address taken out from the shift register 3a, and the reproducing ROM1O is accessed using this absolute address, and the reproducing ROM1O is accessed.
It is necessary to retrieve the feature parameters stored in MIO. In the figure, 8 is an adder circuit for calculating the above-mentioned absolute address, and 7, 9, and 11 are serial-parallel switching devices. By the way, the residual waveform that remains after extracting characteristic parameters related to amplitude and spectrum from an audible sound signal such as a human voice or a melody sound is white noise or an impulse train having a predetermined fundamental period. The individual impulse waveforms constituting an impulse train are slightly different when synthesizing human speech and when synthesizing melody sounds.

Therefore, in the speech synthesis device of the present invention, a sound source ROM that stores impulse waveforms used when synthesizing human speech is used.
16a and a sound source ROM 16b that can store impulse waveforms used when synthesizing melody sounds are provided separately. These sound source ROMs l6a and l6b are the same sound source ROM
It may be formed using different areas of l6. The sound source ROM 16 stores impulse waveform changes in address order, so that when the address counter of the sound source ROM 16 is sequentially incremented, the impulse waveform is reproduced. When the data in the address counter 15 matches the pitch parameter stored in the latch memory 13, the match circuit 14 operates and sends a set pulse to the address counter 15.
Therefore, the data in the address counter 15 is sequentially incremented from O to the value of the pitch parameter.
When the value of the pitch parameter is reached, it returns to O again and repeats the same operation. Therefore, when the address counter 15 accents the sound source ROM 16, an impulse train having a desired fundamental period is reproduced. In the present invention, one piece of data is stored in the address order of the sound source ROM 16 for each sampling period used when sampling the original sound, so one address counter 15a is incremented every sampling period, and the other The address counter 15b is incremented every sampling period at a timing delayed by 1/2 sampling period. That is, in this embodiment, the two address counters 15a and 15b operate completely independently, and therefore the digital filter 23 can be used completely independently in a time-sharing manner. A switching circuit 20 connected to the input side of the digital filter 23 switches between the voiced sound source 17 and the unvoiced sound source 19 under the control of the sound source control circuit 18.

The voiced sound source 17 simulates human vocal cord vibration,
This generates an impulse train that repeats every fundamental period. The unvoiced sound source 19 simulates fricative sounds caused by turbulence in the vocal tract, and generates white noise having a substantially uniform spectral distribution. The switching circuit 20 switches between the voiced sound source 17 when synthesizing a voiced sound that involves vocal cord vibration, such as a vowel, and the unvoiced sound source 19 when synthesizing an unvoiced sound that does not involve vocal cord vibration, such as a consonant. Switching connection to the digital filter 23 via
The digital filter 23 adds amplitude and spectrum information, the amplifier 24 amplifies the synthesized sound, and the speaker 25 reproduces the sound. By the way, in the speech synthesis device of the present invention, the amplitude parameter (
5 bits are assigned to the pitch parameter (P parameter) representing the fundamental period, and 6 bits are assigned to the pitch parameter (P parameter) representing the fundamental period.As shown in Figure 1, the spectral parameter representing the spectrum is the sample value Xt of the audio signal. The partial autocorrelation coefficient (PARCOR coefficient) Kp with the sample value Xt1p which is more than P points apart is used,
Quantization pits such as 7, 5, 4, 4, 4, 3, 3, 3, 3 are assigned to each K parameter of K, K2, .

The number of bit allocations for each parameter is stored in the index ROM 5, and the reproduction control circuit 6 sends the number of shift clocks 41 corresponding to the number of bit allocations to the shift register 3a.
,P,KlO,K,,K8,...,K2,K,
Each characteristic parameter input in series in the order of 1 is sent to the latch memory 13a and the parameter stack 22a. By the way, when synthesizing human speech, only a small number of pitch parameters representing the fundamental period are required.
Rather, a large number of parameters representing amplitude and spectrum are required, but when synthesizing melody sounds, a larger number of pitch parameters representing the fundamental period are required.
Therefore, in order to change the number of bits allocated to each characteristic parameter according to such various uses, in FIG. The configuration has a capacity in advance that allows reading all feature parameters. In this way, the bit allocation number of the pitch parameter can be freely changed by simply rewriting the bit allocation number stored in the index ROM 5. In addition, in the present invention, A parameter, P parameter, and K
, OK to K, and perform various calculations smoothly, an oscillation circuit 2T and a timing control circuit 26 are provided to read data and perform calculation operations based on a time allocation diagram as shown in FIG. I am doing it.

Data is read by 8 times each frame as shown in Figure 2.
This is performed in the first interpolation section D1 of the equally divided interpolation sections Dl, D2, . . . D8. Each section D, to D8 is divided into 25 equal parts, and divided into P1 to P25, respectively. A, P, K, O, K9...
The parameters of K, are all odd-numbered Pl, P3, P
5...They are arranged in series at P23,
P2 is a spare blank. Also, the even number P2,
P4, P6..., P2 are timings for performing interpolation calculations. Furthermore, each area 2 of P1 to P25
It is divided into two equal parts and becomes T,, T2..., T22.
Of these, T1 to T are control signal sections for activating the parameter code detection circuit 28, and actual data is read from T6 onwards. Next, FIG. 6 shows an embodiment of the combined invention.

In the embodiment shown in FIG. 6, the output of the digital filter 23 is directly input to the DA converter 3T without going through the demultiplexer 35, and the DA converter 3? A demultiplexer 35 is connected to the output side of the . That is, in the embodiment shown in FIGS. 3 and 4, since the demultiplexer 35 is placed before the DA converters 3Ta and 31b, two DA converters 31a and 37b are required, and more single tones are required. When synthesizing a chord consisting of In the merged invention, digital filter 23
By directly connecting the DA converter 3T to the output of the DA converter 37, one DA converter 37 is used to perform DA conversion on multiple types of sounds at once. DA converter 3
The time-division data converted from a digital signal to an analog signal at T is separated into separate sounds by a demultiplexer 35, and the timings of the sounds are matched by a delay circuit 39, and then by a mixing circuit 38. It is something that is mixed. In this case, data holding circuits 40a and 40b for holding analog data are required in place of the data latch memories 36a and 36b connected to the output of the demultiplexer 35 in FIGS. 3 and 4. Of course, as mentioned above, the demultiplexer 35 that switches the digital filter 23 operates with one cycle having an extremely short time equal to the sampling cycle of the original sound, so the data holding circuits 40a and 40b that hold analog data are Any device that can hold data for a period of time equal to the sampling period may be used. In the present combined invention, as in the embodiments of the present invention shown in FIGS. 3 and 4, the timing at which data that comes out earlier from the demultiplexer 35 enters the mixing circuit 38 is controlled by the delay circuit 39. By delaying the delay and inputting the data that comes out later from the demultiplexer 35 as it is to the mixing circuit 38, two or more sounds can be reproduced from the speakers 25 at exactly the same time. In FIGS. 4 and 6, the interpolation calculation circuit 12 is connected to the latch memory 13a and parameter stack 22a, which are mainly used for synthesizing human voice messages, but the other latch memories 13b and parameter stack 22b are The interpolation calculation circuit 12 is not connected because there is no need to perform interpolation calculations for these. As described above, in the present invention, a plurality of latch memories for temporarily storing characteristic parameters related to the fundamental period are provided, a multiplexer is provided to sequentially switch and connect the output of each latch memory to the sound source forming means, and the output of the sound source forming means is connected to the sound source forming means. In a speech synthesis device in which the signal passes through a digital filter that is controlled by characteristic parameters related to amplitude and spectrum, a demultiplexer that operates in synchronization with the multiplexer is connected to the output of the digital filter, and each output of the demultiplexer is connected to the output of the digital filter. A plurality of DA converters are connected to the DA converter via a data latch memory, and the output of the DA converter that is connected to the digital filter last by a demultiplexer is directly input to the mixing circuit, and the output of the DA converter is directly input to the mixing circuit, and the output of the DA converter that is connected to the digital filter at the end is inputted directly to the mixing circuit. The output of each DA converter is input to the mixing circuit via a delay circuit, and the delay time of each delay circuit is set so that the outputs of all DA converters are simultaneously input to the mixing circuit, so it is not as simple as the conventional example. Compared to pseudo-synthesizing chords by alternating the fundamental period, this has the advantage that multiple chopsticks with different fundamental periods can be played back at exactly the same time and continuously, making it possible to create pure chords. It has the advantage of being able to reproduce the beautiful sound that it has.

That is, in the present invention, by connecting a data latch memory to the output of the demultiplexer, it is possible to convert the intermittent sound that is output while sequentially switching in a time-division manner into a continuous sound. The advantage of this is that the timing at which multiple sounds that have been converted into sounds enter the mixing circuit is matched by a delay circuit, so multiple sounds that have been analyzed and extracted separately can be played back completely simultaneously in the audio analysis system. It has the following. As shown in the embodiment of FIG. 4, if the structure is configured so that not only the fundamental period of the synthesized sound but also the amplitude and spectrum of the synthesized sound can be switched in synchronization with the switching of the fundamental period, completely different kinds of sounds can be obtained. For example, it is possible to reproduce a human voice message and a melody sound at the same time, and if applied to various home appliances, it is possible to obtain a richer variety of reproduced sounds than when only voice messages are notified.

In addition, in a combined invention of the present invention, the output of the digital filter is directly converted to DA without going through a demultiplexer.
input to the converter, connect a demultiplexer to the output side of the DA converter, and among the outputs of the demultiplexer, the output that is last connected to the digital filter is directly input to the mixing circuit via the data holding circuit, and the demultiplexer The other outputs of the demultiplexer are input to the mixing circuit via a data holding circuit and a delay circuit, respectively, and the delay time of each delay circuit is set so that all outputs of the demultiplexer are simultaneously input to the mixing circuit. One DA for multiple types of sounds output in a time-sharing manner from
It is possible to perform DA conversion all at once using a converter, so even if a large number of sounds are to be output at the same time, the number of DA converters will not increase, and the circuit configuration will be significantly simplified. There are advantages.

In other words, the present combined invention takes advantage of the fact that the data output from the digital filter is originally output in a time-division manner, and passes the output of the digital filter through the DA converter before passing it through the demultiplexer. This is intended to reduce the number of circuits and simplify the circuit configuration. It should be noted that the present invention and its combined inventions do not refer to the digital filter itself, but are intended to share the digital filter in order to obtain a composite sound output.

Therefore, any configuration may be used as a digital filter, but since a normal digital filter is based on the synthesis of a single voice, one filter operation is completed in one sampling period, and this It is designed to output one piece of audio data within a period. Therefore, when two or more pieces of audio data are time-divisionally output within one sampling period as in the present invention and its combined invention, the digital filter is required to operate at a higher speed. Furthermore, when using a so-called PARCOR type (partial autocorrelation coefficient type) filter as a digital filter, it is necessary to store data in the previous sampling period, so two or more pieces of audio data can be stored. If this is desired, it is necessary to provide delay storage means for storing data in a delayed manner until the next filter operation is performed, corresponding to the number of individual sounds constituting the complex sound.

[Brief explanation of the drawing]

Fig. 1 is a principle diagram of the PARCOR type speech synthesis method used in the present invention, Fig. 2 is a time allocation diagram of the same as above, Fig. 3 is an overall block diagram of an embodiment of the present invention, and Fig. 4 is a diagram of the same as above. 5A and 5B are operation explanatory diagrams showing the same operation as above, and FIG. 6 is also a partial block diagram. FIG. 2 is a block diagram of a main part of an embodiment of the invention. 13 is a latch memory, 23 is a digital filter, 30 is a sound source forming means, 34 is a multiplexer, 5 is a demultiplexer, 36 is a data latch, 3T is a DA converter, 38 is a mixing path, 39 is a delay circuit, 40 is a data holding circuit Out.

Claims (1)

[Claims] 1. A plurality of latch memories for temporarily storing characteristic parameters related to the fundamental period among characteristic parameters related to the amplitude, fundamental period, and spectrum constituting an audible sound signal such as voice or melody sound. A multiplexer is provided to sequentially switch and connect the output of each latch memory to the sound source forming means, and the output of the sound source forming means is passed through a digital filter controlled by each characteristic parameter related to amplitude and spectrum, thereby producing the audible sound. In a speech synthesis device that also serves as melody sound synthesis and is configured to resynthesize signals, a demultiplexer that operates in synchronization with the multiplexer is connected to the output of the digital filter, and a data latch memory is connected to each output of the demultiplexer. A plurality of DA converters are connected, and the output of the DA converter that is last connected to the digital filter by a demultiplexer is directly input to the mixing circuit, and the output of the other DA converters is input to the delay circuit, respectively. 1. A speech synthesis device for melody tone synthesis, characterized in that the delay time of each delay circuit is set so that the outputs of all DA converters are simultaneously input to the mixing circuit. 2. Provide a shift register into which a set of characteristic parameters relating to amplitude, fundamental period, and spectrum are input in series at each data update interval, and a plurality of sub-shift registers into which only characteristic parameters relating to the fundamental period are input in series, A transfer means that connects the input and output terminals of each shift register in parallel with each other via a register switching circuit, and transfers characteristic parameters related to different fundamental periods input to each shift register to each latch memory via the register switching circuit. 2. A speech synthesis device for both melody sound synthesis and melody sound synthesis as claimed in claim 1. 3. A shift register into which a set of feature parameters related to amplitude, fundamental period, and spectrum are input in series at each data update interval, and bit allocation of feature parameters related to at least the fundamental period among each feature parameter input to the shift register. An index ROM stores a number in advance, and a number of shift clocks corresponding to the number of bit allocations stored in the index ROM are generated to read the characteristic parameters related to the fundamental period from among the characteristic parameters in the shift register. 2. A voice synthesis device for melody tone synthesis and melody tone synthesis as claimed in claim 1, further comprising a playback control circuit for transferring data to a latch memory. 4 Parameter stacks for temporarily storing characteristic parameters related to amplitude and spectrum are provided in the same number as a plurality of latch memories temporarily storing characteristic parameters related to fundamental period, and a parameter stack is provided between each latch memory and the sound source forming means. The melody sound synthesis system as claimed in claim 1, characterized in that another multiplexer operating in synchronization with the multiplexer interposed in the digital filter is interposed between each parameter stack and the digital filter. Speech synthesizer. 5. A plurality of latch memories are provided to temporarily store characteristic parameters related to the fundamental period among the characteristic parameters related to the amplitude, fundamental period, and spectrum constituting an audible sound signal such as voice or melody sound, and each latch memory is A multiplexer is provided that sequentially switches and connects the output to the sound source forming means, and the output of the sound source forming means is passed through digital filters each controlled by characteristic parameters regarding amplitude and spectrum, thereby resynthesizing the audible sound signal. A demultiplexer that operates in synchronization with the multiplexer is connected to the output of a DA converter connected to the output of the digital filter, and the last of each output of the demultiplexer is The output connected to the digital filter is directly input to the mixing circuit via the data holding circuit, the other outputs of the demultiplexer are input to the mixing circuit via the data holding circuit and the delay circuit, respectively, and the delay of each delay circuit is 1. A speech synthesis device for melody sound synthesis, characterized in that the time is set so that all outputs of a demultiplexer are simultaneously input to a mixing circuit. 6 Parameter stacks for temporarily storing characteristic parameters related to amplitude and spectrum are provided in the same number as a plurality of latch memories temporarily storing characteristic parameters related to fundamental period, and a stack is provided between each latch memory and the sound source forming means. A melody sound synthesis system according to claim 5, characterized in that another multiplexer operating in synchronization with the multiplexer installed in the digital filter is interposed between each parameter stack and the digital filter. Speech synthesizer.