JPH0833742B2 - Speech synthesis method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial field of use] The present invention relates to a voice synthesis method of ADPCM system, and particularly to a waveform for synthesizing one representative waveform element and synthesizing voice by repeatedly using the representative waveform element. The present invention relates to a voice synthesis method of a unit synthesis format.

[Conventional technology]

Conventionally, there are known waveform coding methods such as the PCM method and the ADPCM method as the voice synthesis method. In particular, since the ADPCM method can synthesize a waveform with about half the data amount of the PCM method, the data amount is compressed. In that respect, it is a very effective method. The ADPCM method is a data compression method that utilizes the strength of correlation between adjacent audio samples. The ADPCM method encodes the difference in amplitude value at consecutive sampling points.
Noise is reduced by a method of quantization and by adaptively changing the quantization width. The quantization width in this case is determined by predicting the next quantization width using the current quantization width and the ADPCM code. An example of this prediction calculation is Δ _{n + 1} = Δ _n
× M (L _n ). Here, L _n is the nth ADPCM code data, Δ _n is the size of the quantization width for the nth sampling point, and M (L _n ) is a prediction coefficient, which is a value obtained by multiplying the current quantization width by the prediction coefficient. Is used as the quantization width for the next sampling point. When the voice level is low, the prediction coefficient is smaller than 1, and therefore the quantization width is small. On the contrary, the prediction coefficient is 1 when the voice level becomes high.
It becomes larger and thus the quantization width becomes larger. However,
On the other hand, if the quantization width becomes too large, noise will increase. Therefore, it is necessary to determine the upper limit of the quantization width. Further, if the value of the quantization width becomes 0, it will remain 0 even if any prediction coefficient is applied next time, so the lower limit of the quantization width is also necessary. For such processing including multiplication with the prediction coefficient, a method is generally adopted in which data of the quantization width is stored in a table in the ROM, and the prediction calculation is performed by reading and using the contents of the ROM. This method is also most suitable for LSI. The above quantization table is AD
The table of the read difference value data is created by designating the ROM address using the PCM code and the quantization width pointer.

If the difference value written to the ROM is increased at a constant rate according to the contents specified by the quantization width pointer, increasing the value of the quantization width pointer means that the quantization width is greater than 1 Means that multiplication of the prediction coefficient can be replaced with addition / subtraction for the quantization width pointer.

[Problems to be solved by the invention]

In general, the voice waveform is divided into the voice noise of the consonant part, the tone waveform of the vowel part, and the connecting part between them. Especially, the tone waveform of the vowel part has almost the same period and almost the same waveform. It is normal that they are lined up while deforming little by little. Since two or three consecutive waveforms have almost the same shape, the waveform segment synthesis method compresses the data amount of speech synthesis by selecting one waveform as a representative waveform and repeatedly using it. .

When trying to compress the amount of data for speech synthesis by repeating such waveforms in the ADPCM system, the speech is sampled at the Nyquist frequency and the difference value of the amplitude value of the speech waveform between adjacent sampling points is set to an appropriate quantum value. It is encoded by the width of conversion. Moreover, the quantization width is adaptively changed according to the magnitude of the difference value at each sampling point. Therefore, when synthesizing the representative waveform segment by the ADPCM method,
The quantization width is not constant within one waveform, and the quantization width at each sampling point depends on the immediately preceding ADPCM data.

Moreover, in the ADPCM system, a sampling frequency of 4 kHz to 8 kHz is usually used due to the relationship between the sound quality and the bit rate.

When synthesizing the representative waveform segment using the ADPCM method as described above, the pitch period of the actual original waveform and the pitch period of the synthesized waveform cannot be perfectly matched, and may be affected by the error of the pitch period. The amplitude of the last sampling point of the composite waveform is not zero.

Even when the pitch period of the original sound and the pitch period of the synthesized waveform completely match, the amplitude value at the last sampling point is the difference value determined by the quantization width and the ADPCM code in the amplitude value at the previous sampling point. Since it is added, there is a problem that the amplitude value of the last sampling point in one waveform cannot be set to 0 depending on the value of the quantization width at the immediately previous sampling point.

As described above, when synthesizing the representative waveform segment by using the conventional ADPCM method as it is, the amplitude value at the last sampling point of the representative waveform segment cannot be set to 0. If it is repeated a plurality of times by using it, there arises a problem that the amplitude center of the composite waveform changes.

This is because the ADPCM method is basically based on the basic condition of the differential encoding method in which the difference value is added to the amplitude value of each sampling point to obtain the amplitude value of the next sampling point. This is because if the value is not 0, the error accumulates and the amplitude center of the combined waveform changes. If the amplitude center of the synthesized waveform changes, the D / A converter that converts the waveform data to voice may overflow by the amount of the change in the amplitude center, and if it overflows, a normal voice waveform cannot be synthesized. There are some drawbacks.

The object of the present invention is to significantly improve the synthetic sound quality by eliminating the above-mentioned drawbacks and providing means for storing and reading the quantization width pointer and means for resetting the amplitude value register for accumulating and accumulating difference values to zero. To provide a possible speech synthesis method.

[Means for solving problems]

The method of the present invention uses a quantization width pointer having an upper limit and a lower limit for the difference value required for speech synthesis and ADPC.
In the ADPCM voice synthesizing method that synthesizes the original voice while reading the difference value data based on the amplitude information of the original voice after storing it as the difference value data selected by the M data, based on the amplitude information of the original voice. Quantization width pointer storing / reading means for storing and reading the quantization width pointer designating the difference value data to be read, and an amplitude value to be accumulated by adding the difference value data read by the quantization width pointer. Amplitude value register reset means for resetting the register to zero at the initial timing of the repetitive waveform.

〔Example〕

Next, the present invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the present invention,
L register 101, decoder 102, limiter 103, A register 10
4, A'register 105, difference value data table ROM 106, X register 107, and adders 108 and 109 are provided. The ADPCM code data is latched in the L register 101 and the address of the difference value data table ROM 106 is designated by the content of the L register. Quantization of the next sampling point obtained by adding the value obtained by decoding the content of the L register 101 by the decoder 102 and the value of the A register 104 in which the current value of the quantization width pointer is registered by the adder 108. The value of the width pointer is registered in the A register 104. The purpose of the decoder 102 is to decode the contents of the L register and determine the movement of the quantization width pointer. The address of the difference value data table ROM 106 is designated by the contents of the A register 104. Note that the limiter 103 is inserted so that the value of the quantization width pointer does not become a value other than the designated value.

A register 10 in which the value of the quantization width pointer is registered
The contents of 4 are transferred to and stored in the A'register 105 so that the contents of the A'register can be read out to the A register when necessary. The difference value data is output from the difference value data table ROM 106, and the content of the X register 107 in which the previous waveform amplitude value is registered for each sampling point and the difference value data are added by the adder 109 to obtain the next waveform. The amplitude value is registered in the X register 107. The contents of the X register 107 can be set to 0 by a reset signal.

FIG. 2 is a flow chart showing the contents of voice synthesis in the embodiment of FIG. When voice synthesis is started, first, an initial value set 1001 for setting an initial value 0 in the X register 107 and the A register 104 is performed, and then ADPCM code data is latched in the L register 101 at each sampling point, Register 101 and quantizer width pointer A register 1
Specify the ROM address with 04 and the difference value data table ROM106
Output and the contents of the X register 107 are added, the result is stored in the X register, and the contents of the X register are output from the D / A converter.
Output 1002 from the A converter. At this time, the value obtained by decoding the contents of the L register 101 and the contents of the A register 104 are added, and the A register 104 is used as the next quantization width pointer.
To be stored.

The above is the flow when synthesizing speech in the normal ADPCM system.
Explaining the flow at the time of repeating waveform synthesis, as shown in the initial value set 1003 of the first repeating waveform, the X register 107 is reset to 0 at the beginning of the first repeating waveform and A
The first sampling point S1 is calculated after storing the contents of the register 104 in the A'register 105. Thereafter, the normal arithmetic processing is performed for each sampling point, and the value of the X register 107 is output from the D / A converter.

At the beginning of the second repeated waveform, the X register 107 is reset to 0, the contents of the A'register 105 are read to the A register 104, and then the normal arithmetic processing is performed at each sampling point to set the value of the X register 107 to D / Output from A converter. These processing flows are shown in steps 1004 to 1006.

FIG. 3 is an explanatory diagram of sampling points of the composite waveform in the embodiment of FIG. As described with reference to FIG. 2, if the same flow is repeated as many times as necessary, the sampling data of S1 to S8 as shown in FIG. 3 can be repeatedly output.

〔The invention's effect〕

As described above, according to the present invention, in the ADPCM voice synthesis method, the amplitude 0 is used as a reference as the difference value of the first sampling point of the repetitive waveform, and the quantization of the first sampling point of the second and subsequent repetitive waveforms is performed. The value of the width pointer is set to the same value as the initial value of the repeated waveform using the first value, and the ADP of the representative waveform segment is set by the ADPCM method.
By using the CM code data as it is and providing a means for repeatedly synthesizing the same waveform, there is an effect that it is possible to realize a voice synthesizing method by the ADPCM method that can significantly improve the sound quality and significantly reduce the amount of synthesized data.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of the speech synthesis system of the present invention, FIG. 2 is a flow chart showing the contents of speech synthesis in the embodiment of FIG. 1, and FIG. 3 is a diagram showing the embodiment of FIG. It is a sampling point explanatory view of a synthetic waveform. 101 ... L register, 102 ... decoder, 103 ... limiter, 104 ... A register, 105 ... A 'register, 106 ...
… Differential value data table ROM, 107 …… X register, 108,
109 ... Adder.

Claims (1)

[Claims] 1. A speech synthesis method for designating difference value data by using a quantization width pointer and ADPCM data, and synthesizing a plurality of repetitive waveforms by using difference value data for one waveform of original speech. A quantization width pointer that specifies the first difference value data of the waveform to be specified at the beginning of each repetitive waveform, and an amplitude value register for cumulatively adding the difference value data read by the quantization width pointer are provided. A voice synthesis method characterized in that the amplitude value register in which the amplitude value data to be cumulatively added is stored at the first timing of the repetitive waveform is reset to zero.