JPS63155100A - Max. point detection for self-correlation function given by formant - 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] [Summary] A local maximum point of an autocorrelation function that has a local maximum value that is less than the absolute value of the minimum value of an autocorrelation function calculated from an input voice is determined to be a local maximum point given by the formant of the input voice. .

[Industrial application field]

The present invention relates to an improvement in a method for detecting a maximum point of an autocorrelation function given by a formant, which makes it possible to improve pinch extraction accuracy when pinches are extracted from speech using an autocorrelation function.

One method of extracting pinch, which is one of the important parameters from speech, is to use the maximum point of the autocorrelation function. Note that since the autocorrelation function exhibits maximum points not only in pitch but also in formant, it is strongly desired to improve the performance of means for preventing erroneous extraction of pitch due to this type of maximum point.

[Conventional technology]

The autocorrelation function of a voiced sound is a collection of cosine waves having the pinch frequency of the voiced sound and a frequency that is an integer multiple of that frequency, and the phase is aligned to an integer multiple of 2π at the pinch and its integer multiple point (delay ■). ing.

Therefore, the autocorrelation function shows a maximum value at a pinch and at an integral multiple of the delay amount.

FIG. 4 is a diagram showing an example of a conventional method for detecting a maximum point of an autocorrelation function given by a formant using such a principle.

In FIG. 4, an autocorrelation function calculation circuit 2 calculates a short-time autocorrelation function R(i) (hereinafter simply referred to as an autocorrelation function) normalized by energy for a voice X input from an input terminal l. 1) Calculate based on the formula and transmit it to the integrating circuit 3.

R(i)=ΣW(k) Books X (n+k) Books X (n+
+i) ÷ (ΣW (k) book X ”(n+k) book ΣW(k) book
・(11 However, W (k) two-window function X (n+k), X (+i to n+): Voice n:
Starting point i of the integral interval: Delay amount of the autocorrelation function: Integral interval (for example, 128) Note that iSn and k are both sampling period T (for example, 1
25 microseconds).

Further, the calculation interval of the autocorrelation function R(i) is set to a range (for example, t=1 to 192) that sufficiently covers the pinch frequency region (usually 50 hertz to 500 hertz).

The integrating circuit 3 integrates each delay amount i as shown in equation (2) in order to suppress the high frequency components and emphasize the low frequency components of the received autocorrelation function R(i). The integration result 5(i) is transmitted to the longest zero crossing interval extraction circuit 4.

S (i) −ΣR(j) ・・・
(2) The longest zero-crossing interval extraction circuit 4 calculates the average autocorrelation function Ra (i) by extracting only the polarity of the received integration result S (i) using equation (3).

Ra (i) = SGN (S(i) ) = SGN
(ΣR(j)) ・・・・・・(3) (However,
SGN (A) indicates the polarity of A. ) Furthermore, the longest zero-crossing interval extraction circuit 4 extracts the calculated average autocorrelation function Ra (
i), the zero intersection B M A X B (y) existing in the section (i-1 to 192) (where y= to Y
], and calculate the longest zero-crossing interval K among them.
BMAX is extracted and transmitted to the determination circuit 5.

On the other hand, the maximum point extraction circuit 6 extracts the maximum point (local maximum value RL and delay li=B) of the autocorrelation function R(i) in the interval (i=1 to 192) transmitted from the autocorrelation function calculation circuit
MAX) and transmits it to the judgment circuit '5.

The determination circuit 5 determines that among the maximum points transmitted from the maximum point extraction circuit 6, a maximum point having a delay amount i lower than the longest zero crossing interval KBMAX transmitted from the longest zero crossing interval extraction circuit 4 is higher than the pinch point. Since it is considered to have a frequency, it is determined that the maximum point is due to formant, and the maximum point due to formant is outputted from the output terminal 7 and transmitted to a pitch extraction circuit (not shown).

The pinch extraction circuit removes the maximum point given by the formant from the autocorrelation function R(i) received from the autocorrelation function calculation circuit 2, and extracts the pitch from the remaining maximum points in the following process.

The pitch extraction circuit extracts the maximum value R from among the target local maximum points.
1 (delay amount i = mR), and autocorrelation function values R (mH-1) and R (m * +1)
The average value of is set as the standard value AMAX8, and the standard value AMAXA
A maximum point having a maximum value exceeding
The maximum point where f1i shows the minimum value is determined to be the maximum point given by the pitch.

[Problem that the invention seeks to solve]

As is clear from the above explanation, in the conventional method for detecting the maximum point of the autocorrelation function given by a certain formant, by integrating the autocorrelation function R(i) of the target speech, the longest zero crossing interval KBMAX is determined by the pitch. It is assumed that the period approaches asymptotically, and the maximum point with a delay amount less than that is determined to be the maximum point given by the formant.

However, in the case of speech with a low formant frequency, the longest zero-crossing interval does not asymptote to the pinch period, and therefore the maximum point given by the formant l-, which has a small delay amount, cannot be removed, and the extraction circuit uses it as the maximum point given by the pinch. There was a problem with incorrect detection.

[Failure to solve the problem]

FIG. 1 is a diagram showing the principle of the present invention.

In FIG. 1, 100 is provided according to the present invention and is a process of calculating an autocorrelation function of input speech.

200 is a process provided according to the present invention, in which the absolute value of the minimum value of the autocorrelation function calculated in process 100 is set as a reference value.

300 is provided according to the present invention, and is a process of determining a maximum point of an autocorrelation function having a maximum value below a reference value as a maximum point given by the formant of the input speech.

[Effect]

As described above, the autocorrelation function R(i) of a voiced sound exhibits the maximum value R1 at a delay amount i of a pinch and an integral multiple thereof, and is less than the maximum value R1 at other delay amounts i.

Therefore, if the absolute value of the minimum value of the autocorrelation function R(i) is the reference value RN, then the maximum [Rm at the pinch and the delay Ii that is an integral multiple thereof exceeds the reference value R8, and the reference value R
A maximum point having a maximum value less than , can be determined not to be a maximum point given by a pinch and a frequency that is an integer multiple thereof, but a maximum point given by a formant.

Therefore, the formant frequency is low and the average autocorrelation function R
Even in speech in which the longest zero-crossing interval KBMAX of a (i) does not asymptotically approach the middle Rupisoti period, it is possible to remove the maximum point of the autocorrelation function given by the formant, and the accuracy of the pinch extraction process can be improved.

〔Example〕

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

FIG. 2 is a diagram showing an unvoiced sound detection method according to an embodiment of the present invention, and FIG. 3 is a waveform diagram illustrating the operation of FIG. 2. Note that the same reference numerals indicate the same objects throughout the figures.

In FIG. 2, the autocorrelation function calculation circuit 2 calculates the autocorrelation function R(i) based on (1) 2 for the audio X input from the input terminal 1, as described above, and the minimum value extraction circuit 10.

The minimum value extraction circuit 10 extracts a minimum value R3 from the autocorrelation function R(i) transmitted from the autocorrelation function calculation circuit 2, and transmits it to the reference value setting circuit 20.

The reference value setting circuit 20 sets the absolute value of the minimum value R5 transmitted from the minimum value extraction circuit 10 as a reference value R8, and transmits it to the determination circuit 30.

On the other hand, as described above, the local maximum point extraction circuit 6 extracts the local maximum point (local maximum value RL and delay FJ i = B M A X ) is extracted and transmitted to the determination circuit 30.

The determination circuit 30 determines the reference value R transmitted from the reference value setting circuit 20 among the maximum points transmitted from the maximum point extraction circuit 6.
A local maximum point having a local maximum value less than 8 is determined to be a local maximum point due to formant, and is outputted from the output terminal 7 and transmitted to a pinch extraction circuit (not shown).

As described above, the pitch extraction circuit removes the maximum points given by the formant from the autocorrelation function R(i) received from the autocorrelation function calculation circuit 2, and extracts the pitch from the remaining maximum points.

FIG. 3 is a diagram illustrating an autocorrelation function R(i) and an average autocorrelation function Ra (i) for human speech.

In FIG. 3, according to a conventional method for detecting the maximum point of the autocorrelation function given by a formant, the autocorrelation function R(i)
The maximum point B of is removed as a maximum point given by the formant because it has a delay li that is less than the longest zero-crossing interval KBMAX obtained from the average autocorrelation interval 9Ra (i).
The maximum point C is the delay 11 that exceeds the longest zero crossing interval KBMAX.
Therefore, it will not be removed as a maximum point given by the formant. Therefore, the pinch extraction circuit
When determining each maximum point using the reference value AMAXA, which is determined based on the maximum point A that indicates the maximum value R, , within 192),
Since the maximum value of maximum point C exceeds the reference value AMAXA and has the minimum delay amount i, maximum point C is mistakenly recognized as the maximum point given by pinch, even though A is originally the maximum point given by pinch. That will happen.

However, according to the method of detecting the maximum point of the autocorrelation function given by the formant according to the present invention, the interval (
Since the absolute value of the minimum value R8 in i=1 to 192) is set as the reference value RN, the maximum points B and C having the maximum value below the reference value R, are the maximum points of the autocorrelation function given by the formant. Even if the pinch extraction circuit determines each local maximum point based on the reference value AMA XA, the local maximum point A originally given by the pinch is correctly determined.

As is clear from the above explanation, according to this embodiment, the maximum value that is lower than the absolute value of the minimum value R8 of the autocorrelation function R(i) of the voice X is removed as the maximum point of the autocorrelation function given by the formant. Therefore, the maximum point of the autocorrelation function given by the formant can be removed without error even from the voice X with a low formant frequency.

Note that FIGS. 2 and 3 are merely examples of the present invention, and the waveforms to which the present invention is applied are not limited to those illustrated, and many other modifications may be considered. However,
In either case, the effects of the present invention remain the same.

〔Effect of the invention〕

As described above, according to the present invention, even in speech where the formant frequency is low and the longest zero crossing interval of the average autocorrelation function does not approach asymptotic to the pitch period, it is possible to remove the maximum point of the autocorrelation function given by the formant, and the pitch The accuracy of extraction processing can be improved.

[Brief Description of the Drawings] Fig. 1 is a diagram showing the principle of the present invention, Fig. 2 is a diagram showing a method for detecting the maximum point of an autocorrelation function given by a formant according to an embodiment of the present invention, and Fig. FIG. 2 is a waveform diagram illustrating the operation, and FIG. 4 is a diagram illustrating an example of a conventional method for detecting a local maximum point of an autocorrelation function given by a Polmant. In the figure, 1 is an input terminal, 2 is an autocorrelation function calculation circuit, 3 is an integration circuit, 4 is a longest zero crossing interval extraction circuit, 5 and 30 are judgment circuits, 6 is a local maximum point extraction circuit, 7 is an output terminal, 10 is the minimum value extraction circuit, 4 (the 4th month of the 4th month of the year).
, after the checkpoint tξ, on the other hand, point, Kenda direction Muro 2 ftnr'ns relaxation waveform diagram of Figure 7 Figure 3

Claims (1)

[Claims] Calculate an autocorrelation function of input speech (100), set the absolute value of the minimum value of the calculated autocorrelation function as a reference value (200), and set a local maximum value below the reference value. A method for detecting a maximum point of an autocorrelation function given by a formant, characterized in that a maximum point of the autocorrelation function given by a formant is determined to be a maximum point given by a formant of the input speech (300).