JPH0690635B2 - Pitchiera-correction method - Google Patents

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a pitch error correction method that can be used for voice processing such as voice synthesis and voice recognition.

2. Description of the Related Art In recent years, with the development of computers, the development of voice processing techniques such as voice synthesis and voice recognition has been rapidly advanced.
In particular, pitch cycle extraction is an indispensable technique when performing voice processing, and in addition, pitch cycle error correction is required to be more accurate.

An example of the conventional pitch period error correction method described above will be described below with reference to the drawings. FIG. 5 shows a conventional pitch period error correction method.

The pitch cycle is often doubled or halved, and as shown in FIG. 5, when the flow of the extracted pitch cycle data is seen, when it greatly deviates from the values of the preceding and following frames,
Replace the previous and next values or the previous and next values with interpolated values to correct them.

Further, since the judgment of voiced sound and unvoiced sound is made by using a totally empirical discrimination function, it may be erroneous depending on the voice quality of voice data. Therefore, if the unvoiced sound is mistaken for the voiced sound, a random pitch period will be extracted. In addition, the vowel sounds "a" and "i" may be mistaken as unvoiced sounds. This modification displays the speech waveform and makes a visual judgment. (For example, edited by Kazuo Nakata and Toru Sanbe, “Actual Use of Voice Devices”, Ohmsha).

Problems to be Solved by the Invention However, although the above method is effective in a process of extracting a pitch period in advance such as voice synthesis, the result of the pitch period extraction is displayed and used in a learning device. In this case, there is a problem that incorrect information is given to the learner.

In view of the above problems, the present invention provides a pitch error correction method that can be easily processed by a learning device using a personal computer and that is accurate.

Means for Solving the Problems In order to solve the above problems, the pitch error correction method of the present invention is based on a predetermined threshold value when comparing the extracted pitch period with the preceding and following pitch periods. Whether or not an error has occurred is determined, and if it is determined that an error has occurred, pitch error correction is performed such that it is replaced with the preceding and following pitch periods or 0.

Effect The present invention can quantitatively and easily correct the pitch period error by the above-described method, so that it can be easily processed by a personal computer in a short time.

Embodiment A pitch error correcting method according to an embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 shows an example of pitch extraction results for explaining a pitch error correction method in an embodiment of the present invention.

FIG. 2 is a flow chart for explaining the pitch error correction method in the embodiment of the present invention.

The pitch error correction method of the present invention will be described below with reference to FIGS. 1 and 2.

Assuming that one section is one frame in which the pitch period is extracted, the pitch period of one frame is represented by one dot in FIG. Since the pitch period of 4 dots or more, that is, 4 frames or more is extracted, the following pitch error correction is performed. First, the threshold value is determined in advance. Let this value be c. The value of the pitch period of the first frame is p (1), the second
If the value of the pitch period of the frame is p (2), the value of the pitch period of the third frame is p (3), and the value of the pitch period of the fourth frame is p (4), then (p (3) -p ( The absolute value of 4)) is calculated, and this is designated as d (3). If d (3)> c, replace p (3) with p (4). d (3) ≦
If c, the value of p (3) is left unchanged. further,
The absolute value of (p (2) -p (4)) is calculated, and this is d
(2) If d (2) <c, p (2) is changed to p
Replace with (3). If the value of p (3) has already been replaced, that value is used. If d (2) ≦ c, then p
The value of (2) is left unchanged. Finally, (p (1) -p
The absolute value of (4)) is calculated, and this is designated as d (1). d
If (1)> c, p (1) is replaced with p (2). If the value of p (2) has already been replaced, that value is used. If d (1) ≦ c, the value of p (1) remains unchanged. These operations are performed for all sections in which the pitch cycle is continuously extracted for four frames or more.

As described above, according to the present embodiment, the error correction is performed only by comparing and replacing the pitch periods, so that the process is simple,
Moreover, it is very effective because it corrects the beginning of voiced sounds that are prone to errors.

A second embodiment of the present invention will be described below with reference to the drawings.

FIG. 3 shows an example of pitch extraction results for explaining the pitch error correction method in the second embodiment of the present invention.

FIG. 4 is a flow chart for explaining the pitch error correction method in the second embodiment of the present invention.

The pitch error correction method of the present invention will be described below with reference to FIGS. 3 and 4.

As in the case of the first embodiment, assuming that one section in which the pitch period is extracted is one frame, the pitch period of one frame is represented by one dot in FIG. Since the pitch period of 5 dots or more, that is, 5 frames or more is extracted, the following pitch error correction is performed. The value of the pitch period of the last frame is p (n), the value of the pitch period one frame before the last frame is p (n-1), and the value of the pitch period two frames before the last frame is p (n-2). And (P
The absolute value of (n) -p (n-1)) is calculated, and this is d
(N), the absolute value of (p (n-1) -p (n-2)) is calculated, and this is defined as d (n-1). D (n) -4 *
The value of (d (n-1) +1) is calculated, and if this value is positive, p (n) is set to 0. These operations are performed for all the intervals in which the pitch period is continuously extracted for 5 frames or more.

EFFECTS OF THE INVENTION As described above, according to the present invention, when correcting an error in the pitch cycle extracted from the speech waveform, if one section in which the pitch cycle is extracted is one frame, four or more pitch cycles are continuously extracted. In this case, the absolute value of the difference between each of the pitch periods from the first frame to the third frame and the pitch period of the fourth frame is calculated, and the absolute value of each pitch period difference is determined in advance. If the absolute value of the pitch period difference is larger than the threshold value as compared with the threshold value, the pitch period is corrected, and the pitch period of the third frame is the pitch period of the fourth frame, The pitch cycle is replaced with the pitch cycle of the third frame, the pitch cycle of the first frame is replaced with the pitch cycle of the second frame, and the pitch cycle is 5 consecutive times.
When more than one frame is extracted, the absolute value of the difference between the pitch period of the last frame and the pitch period of one frame before the last frame, the pitch period of one frame before the last frame, and the two frames before the last frame. Calculate the absolute value of the difference with the pitch period, substitute the absolute value of each of the above pitch period difference into a predetermined formula, and if the calculation result is larger than the predetermined threshold value, determine the pitch period of the final frame. The average value of the extracted pitch periods is calculated, and the absolute value of the difference between each of the extracted pitch periods and the average value of the pitch periods is calculated to obtain the difference between the pitch periods. If there is a pitch period whose absolute value is larger than a predetermined threshold value, the pitch period is set to 0, so that the pitch can be quantitatively and easily. Since it is possible to perform error correction of the period, it can be processed easily in a short period of time on a computer.

[Brief description of drawings]

FIG. 1 is an explanatory view showing an example of a pitch extraction result for explaining a pitch error correction method in the first embodiment of the present invention, and FIG. 2 is a pitch error correction in the first embodiment of the present invention. FIG. 3 is a flow chart for explaining the method, FIG. 3 is an explanatory view showing an example of a pitch extraction result for explaining the pitch error correction method in the second embodiment of the present invention, and FIG. 4 is a second view of the present invention. 5 is a flowchart for explaining a pitch error correction method according to an embodiment of the present invention;
The figure is an explanatory view for explaining a conventional pitch error correction method.

Claims (3)

[Claims] 1. A method for correcting an error in a pitch period extracted from a voice waveform, wherein when one section for extracting the pitch period is one frame, it is used when four or more consecutive pitch periods are extracted. A value of each pitch period from the first frame to the third frame;
The absolute value of the difference from the pitch period of the frame is calculated, and the absolute value of each of the pitch period differences is compared with a predetermined threshold value. If the absolute value of the pitch period difference is greater than the threshold value, , The pitch period of the third frame is the pitch period of the fourth frame, the pitch period of the second frame is the pitch period of the third frame, and the pitch period of the first frame is the pitch period of the second frame. A pitch error correction method characterized in that it is replaced with a pitch period. 2. A method used when the pitch period is continuously extracted by 5 frames or more, wherein the absolute value of the difference between the pitch period of the final frame and the pitch period of one frame before the final frame, and the final value. The absolute value of the difference between the pitch period one frame before the frame and the pitch period two frames before the last frame is calculated, and the absolute value of each pitch period difference is substituted into a predetermined formula to obtain the calculation result. Is larger than a predetermined threshold value, the method is a method of correcting the pitch period, wherein the pitch period of the final frame is set to 0. The pitch error correcting method according to claim (1). 3. A method for adjusting the pitch period to a value within a certain difference from the average value, wherein an average value of the extracted pitch periods is calculated, and each of the extracted pitch periods and the average value of the pitch periods are calculated. The absolute value of the difference between the pitch period is calculated, and if there is a pitch period in which the absolute value of the difference between the pitch periods is larger than a predetermined threshold value, the pitch period is set to 0. ) The pitch error correction method described in section.