JPS5952439B2 - sound generator - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a sound generation device including a sound source and a digital filter, and in particular, by providing several types of periodic sound sources as periodic sound sources in the middle of the sound source circuit. ,
It is designed to alternately generate multiple types of voices with different characteristics, or to improve the quality of a single speaker's voice.

In general, it is necessary to use different sound sources, especially periodic sound sources, for men and women, sound effects and human voices, etc., in order to fully incorporate the characteristics of each and perform high-quality speech synthesis.

Also, if you consider the voice of a specific person, the voice is 10
Divided into units of ~20 milliseconds, each unit of speech can be roughly divided into voiced sounds, which can be considered to be driven by one type of periodic sound source, and voiced sounds, which can be considered to be driven by one type of noisy sound source. It is divided into voiceless sounds. However, from the perspective of aiming to improve quality, it is better to use multiple types of periodic sound sources as sound sources to drive voiced sounds, classify the voiced sounds, and drive each with the optimal sound source. Are known. According to the present invention, the periodic sound source signal is automatically switched by including the periodic sound source specifying signal in the audio parameter output from the parameter generating section and extracting the periodic sound source specifying signal by the speech synthesis section as necessary. The present invention aims to provide such a sound generating device.

Hereinafter, the present invention will be explained based on illustrated embodiments. FIG. 1 is a schematic diagram of the main parts of an embodiment of the present invention. In the figure, 1 is a voiced sound source ROM (read only memory) which is a sound source that generates a plurality of periodic sound source signals, and 2 is its address decoder, which is one element constituting the selection means. 3 is a noise signal generator which is a sound source that generates a noise signal, and 4 is the RO for the source sound source.
A sound source switching unit selects either the output of Ml or the output of the noise signal generator 3, 5 is a digital filter, 6 is a DA (digital-analog) 5 converter, 7 is a pitch counter, and 8 is a pitch counter 7 The reset control signal generating circuit 9 and 10 for controlling reset are latch circuits, and the above-mentioned sections constitute a combining section.

The pitch counter 7 is selected together with the address decoder 2. constitutes a means. 11 is a synthesis control unit, and 12 is a phrase start signal.

Further, 13 is a parameter ROM, 14 is a decoding ROM, 15 is an interpolation circuit, and 16 is a parameter register, which sequentially input audio parameters to the synthesis section in chronological order. It constitutes the input parameter generation section.
The voiced sound source is a periodic sound source, and several types of its patterns are stored in the voiced sound source ROM1.

In this example, we will consider below assuming that two patterns are stored. Which pattern is to be read out is specified by the output of the latch circuit 10, and is decoded by the address decoder 2. Which word in one pattern is to be read is repeatedly designated by the pitch counter 7 in pitch cycles.

The unvoiced sound source is generated by a noise signal generator 3, which may be constant regardless of the type of sound. In addition, as the digital filter 5, for example, PARCOR (partial autocorrelation coefficient)
Latisse-type filters based on this method are often commonly used. Further, in the parameter 13, three types of parameters, amplitude, pitch, and reflection coefficient in the PARCOS system are stored, one set for each frame length of 20 milliseconds, for example. These parameters are taken out in chronological order and input into the decoding ROM 14 to be decoded.
In addition, for the purpose of improving sound quality, the interpolation circuit 15 interpolates the values of parameters of the same type between adjacent frames, for example, every 2.5 milliseconds, and inputs the values to the pitch counter 7 as pitch parameters. It is transferred to the register 16 as amplitude and reflection coefficient parameters. FIG. 2 shows the relationship between the breakdown of audio parameters including a periodic sound source designation signal and an example of bit length in the PARCOR system.

In the figure, bit strings are read out from the parameter ROM 13 for each frame in the order of amplitude to K1. KlO
-K1 is the PARCOR reflection coefficient or K parameter. RP and rk are a pitch repetition signal and a K parameter repetition signal, respectively. When this bit becomes 1, the values of the previous frame are used repeatedly for the pitch and K parameters, so in the current frame, those pitch and K parameters are There is no need to store it in the parameter ROM13. E
is a periodic sound source signal designation signal and is used only for voiced sounds. Since unvoiced sounds use a noise signal as a driving sound source, they do not have a pitch. Therefore, unvoiced sounds are identified by setting the pitch to 00000.

Furthermore, for unvoiced sounds, sufficient sound quality can be ensured by using only K1 to K4 as reflection coefficients. When the amplitude parameter is 00000, it is assumed that there is no sound, and other parameters are unnecessary. Also, all amplitude parameters are 11111
In the case of , the speech synthesis is terminated as it indicates the end of the series of speech parameters that started with the phrase start signal, and other parameters are unnecessary. As a summary of the above, the following table shows an example of a pattern for one set of audio parameters. The E bit in FIG. 2 and the table above is latched by latch circuit 9 in FIG.

The output signal of the synthesis control section 11 is used as the load signal for the latch circuit 9. The synthesis control unit 11 receives the phrase start signal 12
is used as input signal 2 to control the operation of the synthesis section and parameter generation section. When the speech synthesis period starts with the input of the phrase start signal 12, during that period, when the E bit extraction signal is input from the synthesis control section 11 to the parameter ROM 13, the E bit 5 is extracted from the parameter ROM 13 and latched into the latch circuit 9. be done. The E bit extraction signal from the synthesis control unit 11 to the parameter ROM 13 may be generated once in the minimum unit of audio parameter update, that is, once per frame period, and the pitch parameter is all O bits. , and therefore occurs when the frame is a voiced sound frame. The output of latch circuit 9 is connected to another latch circuit 10.
The reason why it is more desirable to add this latch circuit 10 will be described below.

The load signal of the latch circuit 10 is the output signal of the reset control signal generation circuit 8.

That is, the load signal of the latch circuit 10 is generated when the pitch counter 7 is being cleared or reset to zero.

Therefore, even if the sound source switching bit changes, the pattern of the voiced sound source does not change immediately, and the pitch counter 7
It will change after becoming 9. Due to this,
The connection between synthesized waveforms from different voiced sound sources becomes smoother, and it is possible to notice an improvement in voice quality. To summarize the above from the perspective of load signal generation timing, the third
The result will be as shown in the figure.

FIG. 4 shows the main block configuration of another embodiment of the present invention. 1 to 7, 9, and 11 to 16 in the figure are the same as those shown in FIG. The example of the audio parameter breakdown and bit length including the periodic sound source designation signal shown in FIG. 2 can be applied as is to the embodiment shown in FIG. The difference from the embodiment shown in FIG. 1 is that the parameter ROM13 is
This is the timing at which the E bit extraction signal to the latch circuit 9 and the load signal from the synthesis control section 11 to the latch circuit 9 are output. In FIG. 4, when the phrase start signal 12 is input, the signal is output from the synthesis control section 12 to the parameter 13 and the latch circuit 9 only once at the beginning frame portion of the speech synthesis period.

As a result, the periodic sound source designation signal during the speech synthesis period is latched in the latch circuit 9, and the address decoder 2
The output signal of the Heratsu circuit 9 is input, and one pattern of the voiced sound source corresponding to the periodic sound source designation signal is selected. FIG. 5 summarizes the above-mentioned information in the form of load signal generation timing.

By configuring as in the embodiment shown in FIG. 4, if E bits are used for one frame for one speech synthesis period, a periodic sound source can be specified.
The number of parameter ROM bits can be saved compared to the embodiment shown in the figure.

Such a method is reasonable if the periodic sound source signal can be fixed to either one during one speech synthesis period. As is clear from the above description, the present invention provides a parameter ROM
The periodic sound source designation signal inherent in
By specifying periodic sound sources in the latch circuit, the quality of the voiced sound for one speaker can be improved in the former two cases, and in the latter case, the quality of the voiced sound can be improved when the speaker changes every speech synthesis period. It can create the convenience of automatically switching the voice sound source,
The effects are very large.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of the main part of an embodiment of the present invention, FIG. 2 is a pattern diagram showing a breakdown of audio parameters including a periodic sound source designation signal and an example of bit length, and FIG. 3 is a timing chart in FIG. 1. 4 is a block diagram of a main part of another embodiment of the present invention, and FIG. 5 is a timing chart in FIG. 4. 1...ROM for voiced sound source, 2...Address recorder, 3...Noise signal generator, 4...
...Sound source switching sound 5...Digital input recorder, 7...Pitch counter, 8...
...Reset control signal generation circuit, 9, 10...
Latch circuit, 11...Synthesis control section, 12...
...Phrase start signal, 13...Parameter ROM
, 14...decoding ROM, 15...
...Interpolation circuit, 16...Parameter register.

Claims (1)

[Scope of Claims] 1. A first sound source that generates a plurality of periodic sound source signals, a second sound source that generates a noisy sound source signal, and a first and second sound source that generates a plurality of periodic sound source signals.
A synthesis section includes a sound source switching section that can select the output of either one of the sound sources, a filter section that amplifies the output of the sound source switching section and performs frequency filtering in a time-varying manner, and a periodic filter for the synthesis section. a parameter generation section that sequentially supplies audio parameters including a sound source designation signal in time series; and a synthesis control section that receives a start signal and controls operations of the parameter generation section and the synthesis section, and the synthesis section a selection means for selecting any one of the plurality of periodic sound source signals and supplying the selected one to the sound source switching section; a periodic sound source designation signal inputted from the parameter generation section; A sound generating device characterized in that it is configured to include a latch circuit that receives an output signal as a load signal and supplies the output signal to the selection means. 2. In the description of claim 1, the output signal of the synthesis control section, which becomes the load signal of the latch circuit, is configured to be generated once in synchronization with the generation of the start signal. A sound generating device. 3. In the statement of claim 1, the parameter generating section sequentially inputs one set of audio parameters including a periodic sound source designation signal to the synthesizing section in time series for each predetermined period, and The output signal of the synthesis control unit, which is a load signal of the circuit, is transmitted once every fixed period in which audio parameters are sent from the parameter generation unit, and when the first sound source is selected by the sound source switching unit. A sound generating device characterized in that it is configured to generate sound only when the sound is being played. 4. In claim 3, the selection means includes a pitch counter that repeats counting at the pitch period of the voice in order to sequentially supply the sampling values of the selected periodic sound source signal to the sound source switching unit. another latch circuit is connected between the latch circuit and the selection means, and a reset control signal for the pitch counter is used as a load signal for the another latch circuit. Device.