JPS59123894A - Head phoneme initial extraction processing system - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

Detailed Description of the Invention (5) Technical Field of the Invention The present invention relates to a processing method for extracting the beginning of the leading end of a phoneme, and in particular, to extract the beginning of the leading end of the leading end of human speech in a tentative manner based on energy. The speech processing system is configured to extract the tentative starting point for each of the middle and high frequency components and the low frequency component of the input audio, and for the extraction, the true starting point is extracted from the other component. This invention relates to a processing method for extracting the beginning of a phoneme at the tip, which reduces the influence of leakage components.

(Bl Background and problems of the technology For example, in monosyllable recognition processing, it is necessary to precisely determine the starting point of the tip end phoneme of the input speech.For this purpose, it has traditionally been necessary to set the energy of the input speech to a threshold value. There is a known method in which the point exceeding the initial point is set as a temporary starting point, and the true starting point is extracted by, for example, extracting the features of the phoneme based on this temporary starting point. For example, the voiceless consonant r i (A
When trying to distinguish between J and the voiced consonant "GA", the middle and high frequency components and the low frequency components are separated, the beginnings of each are extracted, and the difference is made based on the presence or absence of each component and the location of the beginning. This is effective.

The first section shows the energy change of the unvoiced consonant rKAj, FIG. 1(B) shows the energy change of the low frequency component, and FIG. 1 (q shows the energy change of the middle and high frequency component).

In the case of the consonant rKAJ, the middle and high frequency components appear from the beginning, and the low frequency components appear later. Also, Figure 2 (5) shows the energy change of the voiced consonant "GA", and Figure 2 (C) shows the energy change of the voiced consonant "GA".
shows the energy change in the low frequency component, and Figure 2 (0) shows the energy change in the middle and high frequency component. In the case of the consonant rGAJ,
The low frequency components appear from the beginning, and the middle and high frequency components appear later.

Under such a background, when extracting the beginning of the leading end phoneme of the human voice, it is desirable to separately extract the low frequency component and the middle and high frequency component.

(q Purpose and structure of the invention The purpose of the present invention is to solve the above points,
Furthermore, the purpose is to stratify the energy leakage between the components so that the starting point can be extracted correctly. Therefore, the leading end phoneme start extraction processing method of the present invention, when extracting the leading end of the leading end phoneme of the input speech, temporarily determines the beginning end of the leading end phoneme by comparing the energy at the beginning end with a threshold value. , in the front end phoneme start extraction processing method, which extracts the start end from the tentatively determined tentative start end or its vicinity, a middle-high band filter and a low-pass filter are provided to extract the middle-high band components and low band components of the input voice. The mid-high range component start end detecting section detects the start end of the mid-high range component and a value proportional to the energy of the low range component and a predetermined value. and a low-frequency component start detection section for detecting the start of the low-frequency component, which is added in advance to a value proportional to the energy of the middle and high-frequency components. The t+if characteristic includes at least a determination processing section based on a dark value having a composite value of are doing.

This will be explained below with reference to the drawings.

Example FIG. 3 shows an example configuration of ``This issue''.

In Figure 3, 1 is a low-frequency fin, for example, 5.
2 is a medium-high pass filter that passes frequency components of, for example, I KH to 4.9 KH, and 3.4 is a power calculation unit, respectively. 5 and 6 are each 1!
1. The direct determination section 7 and 8 represent a starting end and a production section, respectively, which detect a tentative starting end according to the present invention.

As explained in connection with FIGS. 1 and 2, filters 1 and 2 are provided as shown in FIG. 3 in order to separately extract low frequency components and middle and high frequency components, respectively.

Energy PwL is calculated for the low frequency components that have passed through filter 1 in power calculation section 3, and energy PwH is calculated for middle and high frequency components that have passed through filter 2 in power calculation section 4.

The voiced start edge detection unit 7 basically uses the energy pw[
When + exceeds a predetermined threshold, the tentative beginning t...' of the voiced consonant is extracted. However, even if the original energy pwIi is sufficiently small, if the middle and high frequency component energies P and H are large, leakage occurs, and the apparent energy of the power calculation unit 3 is higher than the above threshold value. occurs. The threshold value determination unit 5 is configured to introduce the value of the energy PWH to determine the darkness value for the sun. The same applies to the threshold value determination unit 6, and the threshold value is determined by introducing the value of the energy pwL.

Temporary starting point t v', t in the starting point detection units 7 and 8
The extraction of u' is performed as follows.

CI) Processing in the silent start detection unit 8.

An observation window having a time width of 1Qms is scanned for the above energy PwH so that sample points are taken at intervals of 2ms. The energy within the observation window at the time position becomes the threshold TH, = 3.0 + 0. I X PwL −(1
), the temporary beginning tu' of the voiceless consonant is extracted.

r■] Processing in the voiced start detection unit 7.

An observation window having a time width of 5 ms is scanned for the above energy PwL so that sample points are taken at intervals of 2 ms. Then, if the energy within the observation window at the time position (1) exceeds the threshold TI (L, = 0.5 - (2)), or (++) the temporary starting point tv obtained by the equation (2) If ' is later than the tentative starting point iu', the energy within the observation window at the above-mentioned acidic time position is the threshold T) (L-= 10.0 + 0.5 X PwH-□
The tentative beginning tv' of a voiced consonant is extracted either when (3) is exceeded.

The tentative starting points tu' and tv' are obtained as described above, but
The starting points iu and tv are extracted by performing the following processing in the vicinity of these temporary starting points.

(III) Silent start point tu0 When a point in time when the energy PwH suddenly changes is detected in the vicinity of the tentative start point tu', that is, within the time position from (tu'-5ms) to (tu'+20m5), Let that point be the starting point tu. If not detected, the temporary starting end tu' is set as the starting end tu.

Note that when tv'>tu' and (tu' + 20 ms) >tv', the above range is from (tu' - 5 ms) to tv'.

(IV) Voiced beginning jv.

From the vicinity of the tentative starting point tv'', that is, from the time position of tv' (t
Within the time position range of v'+5m5), the above energy p
If a point where J suddenly changes is detected, that point is set as the starting point t
・Suppose. If it is not detected, the temporary starting point tv' is set as the starting point jv.

As described above, each beginning 'A jLl , tv is extracted, but the above-mentioned point of sudden change in energy can be considered to be extracted as follows. That is, in the case of starting point t I+ extraction, a time width of 3 ms is also required for the above energy P=H.
) Two observation windows are connected, for example, and scanned to take sample points at intervals of 0.5 ms.

Then, the energy i%1 at the first observation window and the energy Pw at the second observation window! For , calculate , and the value is the threshold 3. Of: When exceeds, it is assumed that a sudden change has occurred at the time position corresponding to, for example, the connection position 1. 1. In the case of starting point tv extraction, two measurement windows with a time width of 3 ms are connected for the above energy PwL, and scanning is performed so as to take sample points at intervals of 0.5 m5 (43 intervals).

Then, the DPW side corresponding to the above equation (4) is the threshold value 2.
．． When it exceeds 0, at the time position corresponding to the relevant connection position, 8. Consider that a change is occurring.

(g) As described in detail, according to the present invention, it is arranged to individually extract mid-high frequency components and low frequency components with respect to the beginning of the leading end phoneme of input speech, and at that time, Even if there is a leakage of energy between the components, the tentative starting point can be extracted correctly, and accordingly, R'yFtA can be extracted correctly.

[Brief explanation of drawings]

FIGS. 1 and 2 are explanatory diagrams for explaining the prerequisite problems of the present invention, and FIG. 3 shows the configuration of one embodiment of the present invention. In the figure, 1 is a low-pass filter, 2 is a mid-high-pass filter, and 3.4
is the power calculation unit, 5 and 6 are the threshold value determination units, and 7.8 is the start yi
Represents the M detection section. Patent applicant: gshitsu Co., Ltd.

Claims (1)

[Claims] When extracting the starting edge of the leading end phoneme of the input speech, the starting end of the leading end phoneme is tentatively determined by comparing the energy at the starting end with a threshold value, and the starting end is extracted from the provisionally determined tentative starting end or its vicinity. The front end phoneme start point extraction processing method is configured to include a mid-high range filter and a low-pass filter to extract mid-high range components and low range components of the input voice, respectively, and detect the start end of each of them. At the same time, the mid-high range component starting edge detection unit for detecting the start end of the mid-high range component includes at least a determination processing unit using a threshold value having a composite value of a predetermined value and a value proportional to the energy of the low frequency component, and the low-frequency component starting edge detection section for detecting the starting edge of the low-frequency component includes at least a determination processing section using a threshold value having a composite value of a predetermined value and a value proportional to the energy of the mid-high frequency component; 1. A front end phoneme start extraction processing method, characterized in that the range component start detection section and the low frequency component start detection section each individually extract the provisional start.