JPS6238718B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field of the Invention] The present invention relates to an apparatus for extracting pitch information from a signal having approximately periodicity such as an audio signal.

[Technical background of the invention and its problems]

Pitch information is one of the basic parameters to be extracted from audio signals when performing audio analysis. In other words, while the consonant part of the speech signal waveform has a noisy continuous spectrum signal,
The vowel part is a pseudoline spectrum signal having strong frequency components at approximately constant frequency intervals, and the intervals of this pseudoline spectrum are called pitch frequencies. In other words, the vowel part is a signal that has approximately periodicity in terms of time, and when this approximate period is T (sec), T is the pitch period, and its reciprocal 1/T (Hz) is called the pitch frequency. These are collectively called pitch information.

FIG. 1 shows the configuration of a conventional pitch information extraction device, which utilizes an autocorrelation method. In the figure, an audio signal sampled at constant intervals τ is applied to an input terminal 11, and is input to a delay circuit 12 with a tap. The delay time mτ, (m+1)τ, (m+2)τ,
The output signals of the taps 13 ₁ , 13 ₂ , 13 ₃ . . . are supplied to the arithmetic circuit 14 . In the arithmetic circuit 14, correlators _{15 1} , 15 _{2 , 15 3 ,} . ) and outputs the result to the determination circuit 16. In the determination circuit 16, correlators 15 ₁ , 15 ₂ , 1
5. Find the time average value of the correlation coefficients calculated in ₃ , ... according to the control signal input to the terminal 17, and then compare these time average values to find the tap where this value is the maximum. The value of the delay time is output to the terminal 18 as pitch information (period). For example, if it is determined that the time average value of the correlation coefficient calculated by the correlator ₁₅₂ is the maximum, tap 1
The delay time (m+1)τ of ₃₂ is the pitch period. That is, as mentioned above, since the vowel part of the audio signal has approximately periodicity, the correlation coefficient calculated by the correlator connected to the tap whose delay time is approximately equal to the period becomes large, and is output to the terminal 18. The information on the delay time provided by the signal provides pitch information.

However, in the apparatus having the configuration shown in FIG. 1, erroneous extraction of pitch information of about 5 to 10% is unavoidable, which poses a problem in terms of reliability. FIG. 2 shows a typical waveform example of the vowel part of the audio signal, which has approximately periodicity at time T. Such an audio signal is expressed as τ
Input terminal 1 in Figure 1 is sampled at a time interval of
1, depending on the relationship between the approximate period T of the audio signal and the delay time of each tap in the tap delay circuit 12, T may be extracted correctly or may be incorrectly detected as an integral multiple thereof.

Figure 3 shows that the audio signal as shown in Figure 2 is input to the input terminal 11.
Correlators 15 ₁ , 15 ₂ , 15 when given to
The correlation coefficients calculated in ₃ , . . . are shown in correspondence with the delay times of the respective taps 13 ₁ , 13 ₂ , 13 ₃ , . Here, FIG. 3a shows a case where T is approximately equal to an integral multiple of τ (m+1)τ, and in this case, the output of the correlator _15o connected to the tap _13o of the delay time of (m+n)τ is Therefore, T is correctly extracted.
On the other hand, in Fig. 3b, T is (m+n)τ<T<
(m+n+1)τ, and in this case, not only the outputs of the correlators 15 _o and 15 _o+1 but also the taps 13 _2(n+o) with a delay time close to twice T,
13 _2(n+o)+1 , correlator 1 connected to 13 _2(n+o+2)
The output of 5 _2o , 15 _2o+1 , 15 _2o+2 also increases,
In particular, there is a possibility that the output of correlator 15 _2o+1 becomes maximum. In this case, the delay time 2(m+n)τ+τ of the tap 13 _2(n+o)+1 corresponding to the correlator 15 _2o+1 ,
In other words, a time approximately twice as long as T will be erroneously extracted as the pitch period.

This erroneous extraction of pitch information is due to the fact that the true maximum value of the autocorrelation coefficient may be missing because discrete values of the autocorrelation coefficient are compared. . Therefore, this false detection can be reduced to some extent by increasing the sampling frequency of the input signal in order to calculate the autocorrelation coefficient more precisely, but for example, doubling the sampling frequency will also double the circuit size of the device. , which increases the cost burden and is not a practical idea.

[Purpose of the invention]

An object of the present invention is to provide a pitch information extraction device that can effectively improve pitch information extraction accuracy without increasing the circuit scale.

[Summary of the invention]

In this invention, the sampling interval of the input signal is τ, an arbitrary value of 2 or more is m, and N=2 ⁱ (i=
0, 1, 2, ...), the delay time of the delay circuit with taps is divided into multiple ranges represented by N{mτ~(2m-1)τ}, and the time interval of Nτ is set for each of these ranges. The present invention is characterized in that the output signal of the tap is supplied to an arithmetic circuit that performs arithmetic operations to extract pitch information.

In other words, the search for pitch information is not performed at fixed time intervals τ as in the past, but at mτ~(2m
−1) In the range of τ, τ, 2{mτ ~ (2m−1)
2τ in the range of τ}, 4{mτ~(2m−1)
In the range of 4τ, . . . , in the range of 4τ, .

〔Effect of the invention〕

According to this invention, even when the pitch period is between the delay times of adjacent taps in the delay circuit with taps, times that are twice, four times, etc. of the true pitch period are incorrectly determined as the pitch period. This prevents erroneous extraction of pitch information caused by searching for pitch information discretely. This makes it possible to improve the accuracy of extracting pitch information.

In addition, in this invention, it is possible to prevent incorrect extraction of pitch information without increasing the sampling frequency of the input signal more than necessary, and the time interval between the taps for supplying the output signal from the delay circuit with taps to the arithmetic circuit is the minimum. τ, which becomes larger in the region where the delay time is longer, so that the circuit scale including the delay circuit with tap and the entire arithmetic circuit can be simplified.

[Embodiments of the invention]

FIG. 4 shows the configuration of a pitch information extraction device using an autocorrelation method according to an embodiment of the present invention. In the figure, an input terminal 41 receives a roughly periodic signal, such as an audio signal, sampled at constant intervals τ, and is applied to a delay circuit 42 with a tap. The delay circuit 42 with tap has a delay time m
In the range τ~(2m-1)τ, there are taps 43 ₁ , 43 ₂ ,... with time intervals of τ, and the delay time is 2.
In the range of {mτ to (2m-1)τ}, there are taps 43 _P , 43 _P+1 , ... with time intervals of 2τ, and in the range of delay time 4 {mτ to (2m-1)τ}, there are taps of 4τ. It has taps 43 _Q , 43 _Q+1 , . . . with time intervals of .
And these taps 43 ₁ , 43 ₂ ,...43
The output signals of _P , 43 _P+1 , . . . 43 _Q , 43 _{Q + 1} , .
45 _P , 45 _P+1 , . . . 45 _Q , 45 _Q+1 , . That is, the delay time of the tap delay circuit 42 is N{mτ
〜(2m−1)τ}
=2 ⁱ , i=0, 1, 2 . . . , m: any value), and output signals of taps at time intervals of Nτ are supplied to the arithmetic circuit 44 within each of these ranges.

In the arithmetic circuit 44, as in the case of FIG. 1, correlators 45 ₁ , 45 ₂ ,...45 _P , 45 _P+1 ,
The time average value of the correlation coefficients calculated at 45 _Q , 45 _Q+1 , . Then, set the delay time value of the tap that maximizes this value as pitch information (period).
It is output to terminal 48 as .

According to the above configuration, the delay times of the tapped delay circuit 42 corresponding to each of the correlation coefficients calculated within the arithmetic circuit 44 and compared with each other are mτ, (m+1)τ, (m+2)τ, . . . (2m−
1) τ, 2mτ, 2(m+1)τ, 2(m+2)
τ,...2(2m-1)τ, 4mτ, 4(m+1)
τ, 4(m+2)τ, . . . 4(2m−1)τ, . . . Therefore, no matter what value the pitch period T has, it can be extracted without a large error. That is, T is (m
Even if it is between +n)τ and (m+n+1)τ, 2(m+n)τ+τ calculated with the conventional configuration
The correlation coefficient corresponding to the delay time of is excluded. As a result, the time average value of the correlation coefficient corresponding to the delay time of (m+n)τ or (m+n+1)τ becomes maximum, and the value of this delay time is extracted as the pitch period T. The extraction error in this case is within τ,
Sufficient extraction accuracy can be obtained.

In this way, the calculation error of the true maximum value of the correlation coefficient due to comparing the time average values of the correlation coefficient corresponding to discrete delay times is twice, or even four times, the pitch period T. The time interval for calculating the correlation coefficient (delay time difference between adjacent taps) so that it occurs under the same conditions near the delay time of 8 times...
is set corresponding to each range of delay time of the delay circuit with tap 42, thereby preventing the pitch period from being erroneously detected as twice, four times, . . . .

In addition, in this invention, as mentioned above, the correlation coefficient is calculated at coarser intervals in the region where the delay time is longer, so for example, the pitch period T is T=(2m+
1) In the case of τ+τ, the pitch period is 2(m+1)
It will be judged as either τ or 2(m+1)τ+2τ and an error will occur, but this type of error is small compared to the error that the device inherently has.
No problem at all. This is because pitch information is extracted from the sampled signal, and the longer the pitch period (lower the pitch frequency), the higher the frequency resolution. For example, if the sampling frequency is 8KHz (τ = 125μsec) and m = 16, the extracted pitch information (frequency) is T = m
Near τ, 1/mτ=500Hz 1/(m+1)τ=470Hz, and the frequency resolution is 15Hz. On the other hand, T=
Near 2mτ, 1/2mτ=250Hz 1/(2m+1)τ=242, 42Hz, and the frequency resolution becomes as high as 3.8Hz. Even if the correlation coefficient corresponding to the delay time of 2(m+1)τ is excluded according to the present invention and 1/2mτ=250Hz 1/2(m+1)τ=235.3Hz, the frequency resolution is 7.41Hz, and T =m
much higher than near τ. Therefore, the present invention does not increase new extraction errors.

As described above, according to the present invention, erroneous extraction of pitch information can be prevented with a simple circuit configuration, and extraction accuracy is significantly improved.

Note that this invention is applicable to methods other than the autocorrelation method, such as
It can also be applied to a pitch information extraction device using the AMDF method. In that case, instead of using the correlator 45 of FIG. 4, the terminals 51, 5 as shown in FIG.
The subtracter 53 calculates the difference between the input signal given to each of the input signals 2 and the output signal of the tap delay circuit 42,
This is input to the cumulative adder 55 through the absolute value circuit 54, the minimum value of the cumulative addition result obtained at the terminal 56 is detected by the judgment circuit, and the tap delay time of the tap delay circuit 42 that gives this minimum value is determined by the pitch. It can be extracted as a cycle.

[Brief explanation of the drawing]

Figure 1 is a configuration diagram of a conventional pitch information extraction device.
Figure 2 is a diagram showing a typical waveform example of the vowel part of a voice signal, Figures 3a and b are diagrams showing the autocorrelation coefficients in response to the delay time of the delay circuit with taps, and Figure 4 is a diagram showing typical waveforms of the vowel part of the audio signal. FIG. 5 is a block diagram of a pitch information extraction device according to an embodiment of the invention. FIG. 5 is a block diagram of main parts of another embodiment of the invention. 41...Input terminal, 42...Delay circuit with tap, 4
3 ₁ , 43 ₂ ,...Tap, 44... Arithmetic circuit, 45
... Correlator, 46... Judgment circuit, 47... Control signal input terminal, 48... Output terminal.

Claims (1)

[Claims] 1. A delay circuit with taps into which a sampled signal with approximately periodicity is input, and an operation for extracting pitch information of the input signal from the output signals of a plurality of taps of this delay circuit and the input signal. In the pitch information extraction device equipped with a circuit, the sampling interval of the input signal is τ, an arbitrary value of 2 or more is m, and N=2.
ⁱ (i=0, 1, 2,...), the delay time of the tap delay circuit is N{mτ~(2m-
1) A pitch information extraction apparatus characterized in that the pitch information is divided into a plurality of ranges represented by τ}, and output signals of taps at time intervals of Nτ are supplied to the arithmetic circuit for each of these ranges. 2 The arithmetic circuit calculates the time average value of the correlation coefficient between each output signal of the delay circuit with taps and the input signal at fixed time intervals, and extracts the value of the delay time of the tap at which this value becomes the maximum as pitch information. A pitch information extraction device according to claim 1, wherein the pitch information extraction device is a pitch information extraction device. 3 The arithmetic circuit calculates the cumulative addition value of the absolute value of the difference between each output signal of the delay circuit with taps and the input signal at fixed time intervals, and extracts the value of the delay time at which this value is the minimum as pitch information. A pitch information extraction device according to claim 1, characterized in that: