JP2560277B2 - Speech synthesis method - Google Patents

Description

The present invention relates to a speech synthesis system.

[Prior Art] Speech synthesis methods are roughly classified into a parameter method and a waveform coding method.

Typical examples of the former are formant, RARCOR, and LPC methods. The method using such a vocal tract parameter has an advantage of high data compression rate,
Due to the condition of the analysis process, the sound quality varies greatly, and the sound quality lacks naturalness.

The latter includes DM, ADM and ADPCM methods. Although these systems can obtain high quality synthetic sound quality, they have a drawback that a large amount of data is required to synthesize one second of voice as compared with the parameter system. Therefore, in combination with the waveform element synthesis method,
A method of compressing the amount of data is used.

The voice signal has a complicated waveform including various frequency components, but the waveform of the vowel part of the voice signal has a similar waveform repeatedly for each basic synchronization. Focusing on this point, the waveform segment synthesis method takes out one similar waveform that appears repeatedly (hereinafter referred to as a segment) as a representative (hereinafter, this segment is referred to as a representative segment).
By repeating this segment, the original similar waveform portion is reproduced. Thereby, it is not necessary to store the data of all the waveforms, and the data amount can be reduced.

In an actual voice signal, the envelope of the amplitude value of the fundamental cycle changes continuously. However, in this waveform segment synthesis method, since the representative segment is repeatedly used, the envelope of the repeated portion becomes constant. This is shown in FIG.

FIG. 2 is a waveform diagram in which a representative element shown in the range of a (hereinafter referred to as representative element a) is repeated three times and a representative element shown in the range of b (hereinafter referred to as representative element b) is repeated twice. Is. This is because the envelope does not change during the repetition of each of the representative pieces a and b.

Therefore, when the representative segment is repeated, the envelope change value ΔE, which is information about how much the amplitude increases and decreases relative to the previous repetitive waveform, is stored separately from the representative segment data, and the representative segment data is retained. At the time of one-sided repetition, this envelope change value ΔE is positive,
The method of giving to the waveform data on the negative side has been taken.

The envelope change value ΔE is given by multiplying the amplitude value of the representative segment by the envelope data E with the envelope change value ΔE as a parameter.

That is, for the first repetitive waveform, the envelope data is always multiplied by E = 1 and the amplitude value of the representative segment. Therefore, the amplitude value of the representative segment remains unchanged. With respect to the second repetitive waveform, the value of E = 1 + ΔE is used as the envelope data and multiplication with the amplitude value of the representative segment is performed. Similarly for the third and subsequent repetitive waveforms, E = 1 + 2ΔE
(3rd waveform), E = 1 + 3ΔE (4th waveform) ... Envelope data is used for multiplication with the amplitude value of the representative segment.

FIG. 3 shows a case where an envelope change value is added to the waveform shown in FIG. That is, FIG. 3 shows a case where an envelope change value of ΔEa is given when the representative segment a is repeated three times, and an envelope change value of ΔEb is given when the representative segment b is repeated twice. Is.

As a method of calculating the envelope change value, there is a method of obtaining it from the maximum amplitude value difference between the representative segment and the next representative segment. Equation (1)
Is an expression for obtaining the envelope change value ΔE from the maximum amplitude value difference between the representative segment and the next representative segment.

RPn: Number of repetitions of the representative segment Mn: Maximum amplitude value (absolute value) of the representative segment Mn + 1: Maximum amplitude value (absolute value) of the next representative segment Conventionally, the waveform segment synthesis method mainly uses onomatopoeia and melody. This signal is used for signals whose sine wave, cosine wave, claw wave, etc., and whose positive and negative amplitudes are almost symmetrical, and one envelope variation value ΔE is given to each representative segment. It was

[Problems to be solved by the invention]

In the above-described conventional waveform segment synthesis method, one envelope change value ΔE is given to each representative segment, so that one envelope segment when the positive and negative amplitudes of the representative segment are asymmetrical like a voice signal. When the change value ΔE is added to the positive-side waveform data and the negative-side waveform data, a sudden change in amplitude occurs between the last repeated element and the next first-repeated element, and this distortion There was a problem that a gritty, gritty sound was generated.

That is, as shown in FIG. 3, when the envelope change value ΔE of the representative segment a is added to the positive-side waveform data and the negative-side waveform data, the third amplitude of the representative segment a and A sudden change in amplitude occurs between the representative element b and the first amplitude, and a harsh sound is generated.

An object of the present invention is to provide a speech synthesis system that can obtain speech with less distortion.

[Means for solving problems]

The speech synthesis method of the present invention is a speech synthesis method using a waveform segment synthesis method that assigns envelope information relating to an increase / decrease of an amplitude value of waveform data to a plurality of representative segments each having a repeated similar waveform. The increase / decrease rate between the maximum value on the positive side (and the negative side) of the first representative element and the maximum value on the positive side (and the negative side) of the second representative element is calculated, and this increase / decrease rate A value obtained by dividing the number of repetitions of one representative segment is set as envelope information, and the envelope information obtained at the time of speech synthesis is separately given to the positive side waveform data and the negative side waveform data.

〔Example〕

Next, embodiments of the present invention will be described with reference to the drawings.

The following expressions (2) and (3) are examples of expressions for obtaining the envelope change value ΔEPn given to the positive side waveform data and the envelope change value ΔEMn given to the negative side waveform data, respectively.

MAXn: Maximum value on the positive side of representative segment data MAXn ₊₁ : Maximum value on the positive side of the next representative segment data MINn: Minimum value on the negative side of the representative segment data MINn ₊₁ : Next representative segment data Negative minimum value of RPn: Number of iterations In this case, the parameters for synthesis are representative segment data, number of iterations, and envelope data ΔEPn, ΔEM
You can give n.

FIG. 1 is a composite waveform diagram for explaining one embodiment of the present invention, in which the vertical axis represents the amplitude value and the horizontal axis represents the time.

The waveform of FIG. 1 is obtained by independently adding the envelope change value to the waveform data of the positive side and the waveform data of the negative side with respect to the waveform of FIG.

That is, the envelope change values ΔEPa and ΔEP to be added to the waveform data on the positive side for the representative segment a and the representative segment b, respectively.
b, Envelope change value ΔEMa, ΔE given to the waveform data on the negative side
Mb is given and these are added when the representative element is repeated.

That is, when the representative segment a is repeated for the first time, a and 1
When the amplitude value of the representative segment a is positive, it is multiplied by 1 + ΔEPn, and when the amplitude value is negative, it is multiplied by 1 + ΔEMn. Similarly, for the third time, if the amplitude value is positive, 1 + 2ΔEPn is multiplied, and if the amplitude value is negative, 1 + 2ΔEMn is multiplied.

In this way, by adding the envelope change values to the positive-side waveform data and the negative-side waveform data separately, it is possible to reduce the change in amplitude between different representative units and synthesize speech with less distortion. You can

〔The invention's effect〕

As described above, according to the present invention, in the waveform segment synthesizing method, the envelope change value given to the positive-side waveform data and the negative-side waveform data is held as envelope data, and when the representative segment is repeated, Envelope data can be added to the positive-side waveform data and the negative-side waveform data independently of each other, so that the envelope can be continuously changed when the representative unit is repeated. Is obtained.

[Brief description of drawings]

FIG. 1 is a synthetic waveform diagram for explaining one embodiment of the present invention,
FIG. 2 is a synthesized waveform diagram for explaining the waveform segment synthesis method.
FIG. 3 is a synthesized waveform diagram for explaining the conventional speech synthesis method.

Claims (1)

(57) [Claims] 1. A voice synthesizing method using a waveform element synthesizing method for assigning envelope information relating to an increase / decrease of an amplitude value of waveform data to a plurality of representative elements each having a repeated similar waveform. Maximum value on the positive side (and negative side) of the representative segment and the positive side (and negative side) on the second representative segment
Of the maximum value of, and the value obtained by dividing the increase / decrease rate by the number of repetitions of the first representative segment is the envelope information, and the envelope information obtained at the time of speech synthesis is the waveform data on the positive side and A voice synthesis method characterized in that it is given separately to the waveform data on the negative side.