JPH0446400A - Pitch extracting circuit - Google Patents

Abstract

PURPOSE:To reduce the memory quantity for accumulation of the data necessary for pitch extraction and to shorten the delay time before data output by detecting the point of the time P pieces of voice data are accumulated in a memory means, successively reading out P pieces of the voice data before the inputting of P+1 pieces of the voice data and calculating correlation coefft. with P pieces of the voice data which are initially stored as reference values. CONSTITUTION:The input voice data is accumulated in the memory means 2 with the count value of a counter 1 which is reset at every 2P (P is the max. pitch period which can be detected) period and counts up at every inputting of the voice data as an address. The point of the time P pieces of the data are accumulated is detected by a P detecting circuit 2 and P pieces of the voice data before the inputting of P+1 pieces of the voice data are successively read out and the correlation are calculated in a correlation calculating circuit 4 with P pieces of the voice data which are initially accumulated as the reference values. The pitch period of the voice signals is detected in a correlation max. value detecting circuit 9 from the intervals between the input voice data at which the correlation value attains the max. value and the voice data of the reference values. The memory quantity is reduced to 2P and the time before the data output is shortened.

Description

[Detailed Description of the Invention] [Summary] Regarding a circuit for extracting a pitch cycle in an audio signal, a pitch extraction circuit is provided that can reduce the amount of memory for storing data required for pitch extraction and shorten the delay time until data output. A counter that is reset every 2P (P is the maximum detectable pitch period) period and counts up each time audio data is input, and a counter that stores input audio data using the count value of the counter as an address. P detection for detecting the point in time when P pieces of audio data are stored in the storage means and sequentially reading out P pieces of audio data before that including the input audio data in the storage means from the time when P+1 pieces of audio data are input. a correlation calculation circuit that calculates a correlation coefficient between the circuit and each audio data read out using the P audio data first accumulated in the storage means as a reference value; and an input for which the correlation value is a maximum value. The present invention includes a correlation maximum value detection circuit that detects the pitch period of the audio signal from the interval between the audio data and the reference value audio data.

[Industrial application field]

The present invention relates to a circuit for extracting a pitch period in an audio signal, and in particular to a pitch extraction circuit that starts extraction after inputting data of the maximum detectable pitch width in time-base companding coding in audio data compression transmission. be.

In the time-domain companding coding method (TDH3), an audio signal is sampled, and using the periodicity of each pitch period of the audio, 2 to 3 periods are compressed into one period, encoded, and transmitted. By doing so, the audio information is compressed and the transmission path is used efficiently.

Therefore, in such a case, a pitch extraction circuit is required to extract the pitch period from the audio signal input.

The pitch extraction circuit requires less buffer memory for pitch extraction and takes less time to complete pitch extraction, so when applied to time-based companding encoding, the delay time until compressed data is obtained is shorter. It is requested that it be a thing.

[Prior Art] Various pitch extraction methods are known, but
In the following, the correlation between the input audio data and the audio signal waveform captured for a time length corresponding to the pitch extraction range, that is, the maximum pitch width P to be extracted, is calculated, and this audio when the correlation reaches the maximum value is calculated. A method for determining the deviation width between a signal waveform and input audio data as a pitch period will be described.

The waveform correlation in this case is determined by an evaluation function. It is assumed that audio is sampled at a predetermined sampling period and converted into digital values. Below, input audio data is A (i) (i-1, 2, 3,...
...).

In the conventional pitch extraction method, if data starts to be input from A(1), the data from A(P+1) to A(2*P) is extracted when inputting up to A(2*P) is completed. The calculation of waveform correlation was started as a reference value.

FIG. 5 shows a conventional pitch extraction circuit,
11 is a counter that counts addresses; 12 is a dual port memory; 13 is a 2P detection circuit that detects the counter value 2P; 14 is a correlation calculation circuit that calculates correlation;
15 is a correlation maximum value detection circuit that detects the maximum value of correlation.

Audio input is sampled at regular intervals, converted to digital data, and input. The counter 11 increments the write address (W) for the dual port memory 12 by counting up each time data is input.
-ADR), thereby writing the audio data to the dual port memory 12.

On the other hand, the 2P detection circuit 13 detects that the counter value has become 2P, generates a read address (R-ADR), and reads the consecutive P data accumulated in the dual port memory 12 up to that point. Read out the data as the reference value.

From then on, the read address (R-ADR
), and P pieces of consecutive data, which are sequentially shifted by one from the first input data, are read out from the dual port memory 12 as comparison values and input to the correlation calculation circuit 14.

At the same time, an instruction is given from the 2P detection circuit 13 to the correlation calculation circuit 14, so that the correlation calculation circuit 14 starts calculating the correlation using the read reference value and comparison value.

When the correlation calculation circuit 14 finishes calculating the correlation, the maximum correlation value detection circuit 5 detects the position of the comparison value where the waveform correlation is maximum, and extracts the pitch width from the deviation between this and the reference value.

Such processing is repeated every 2P period, and when the processing ends, the counter 11 is reset and the processing starts from the beginning.

FIG. 6 explains pitch extraction in the conventional method.

In the figure, ↓ indicates data input, ■ indicates the start of waveform correlation calculation and calculation of correlation function F (B, C,), and ■ indicates correlation function F
Calculation of (B, C,), ■ Calculation of correlation function F (B, c,), ■ Calculation of correlation function F (B, Ci), ■ Calculation of correlation function F (B, CP-z) , ■ is the calculation of the correlation function F CB, CP-1), ■ is the calculation of the correlation function F (
B, C,) calculation and pitch extraction are shown.

As shown in FIG. 6, audio data group A (P+1
) z, '-'-', C3 A(1) A (2) ~A(2*P) is the reference value, and each audio data group indicated by C,, CC ~A (P) ~A(P+1) Using A(i) to A(i+P-1) A (P) to A (2*P+1) as comparison values, audio data group B and audio data group Cr
(i=1, 2. . . , p) to obtain waveform correlation. Then, when the waveform correlation function between audio data group B and audio data group C positive is F (B, C,), i, where the correlation number F (B, C,) has the maximum value, is extracted as the pitch period. do.

Here, the waveform correlation function F (B, C,) is F (B,
Ci).

[Problem to be solved by the invention]

In a conventional pitch extraction circuit, for example, if one sampling period is required to calculate one evaluation function, buffering of P data inputs is required until calculation of all evaluation values is completed.

Therefore, as shown in FIG.
A(1)) until pitch extraction requires (3*P) data input time, and a buffer is also required to store that much data.

The present invention aims to solve the problems of the conventional technology, and aims to reduce the amount of memory needed to store data necessary for pitch extraction in a pinch extraction circuit, and shorten the delay time until data output. The purpose of this invention is to provide a pitch extraction circuit that can perform the following steps.

[Means to solve the problem]

The present invention has a 2P (
P is the maximum pitch period that can be detected) A counter 1, which is reset every cycle, counts up each time voice data is input, and a storage means 2 stores human voice data using the column address of the counter I as an address.
The detection circuit 3 detects the point in time when 8 pieces of audio data are stored in the storage means 2, and sequentially records P pieces of audio data including the input audio data in the storage means 2 from the time when P+1 pieces of audio data are input. The reading and correlation calculation circuit 4 calculates the correlation coefficient with each read audio data using the P audio data initially accumulated in the storage means 2 as a reference value, and the correlation maximum value detection circuit 5
Accordingly, the pitch period of the audio signal is detected from the interval between the input audio data with the maximum correlation value and the audio data of the reference value.

Furthermore, the present invention is such that in this case, the storage means 2 comprises a dual port memory 6 in which audio data can be written and read simultaneously.

Further, the present invention provides that the storage means 2 includes a single port memory 7 and a selector that alternately writes and reads audio data to and from the single port memory 7 every time audio data is input from the time of P+1 audio inputs. It consists of 8.

[Effect]

A counter that is reset every 2P cycle is counted up every time audio data is input, the input audio data is stored in the storage means using this count value as an address, and the point in time when P pieces of audio data are stored in the storage means is detected. TeP
From the time when +1 audio data is input, the previous P audio data including the input audio data in the storage means are sequentially read out as comparison values, and the P audio data stored first in the storage means is used as a reference value. A correlation coefficient with each audio data of the comparison value is calculated, and the pitch period of the audio signal is detected from the interval between the input audio data of the comparison value and the audio data of the reference value at which the correlation value becomes the maximum value.

In this case, the storage means may be a dual port memory that can simultaneously write and read audio data.

Also, the storage means is single port memory and P+1
single voice input for each voice data input.
It may also include a selector that alternately writes and reads audio data to and from the port memory.

FIG. 2 provides a detailed explanation of the invention.

In the figure, ↓ indicates data input, ■ indicates the start of waveform correlation calculation and calculation of correlation function F(b, ci), and ■ indicates correlation function F(b, ci).
, c, ), ■ is the calculation of the correlation function F (b, ci-+), ■ is the calculation of the correlation function F (b, c2-1), ■ is the calculation of the correlation function F (b, cm2), ■ is the correlation function F (b,
Calculation of Ca-1), ■ is the correlation function F (b, c)
calculation and pitch extraction.

As shown in FIG. 2, audio data group A(1)--shown by b
With A(P) as the reference value, CI, C2, '-' + C4
+ '-''+ Each audio data group A(2) ~ shown on CD
Using A (p+1) A(3) to A (P+2) A(i-1) to A(P+1) A(P+1) to A(2*P) as comparison values, the audio data group becomes the audio data group Ci.
(i=1.2. . . , p) to obtain the waveform correlation. Then, when the waveform correlation function between the voice data group and the voice data group C8 is F (b, c,), i, where the correlation number F (b, c,) has the maximum value, is extracted as the pitch period. .

In the pitch extraction circuit of the present invention, first A(1) to A(P
), then input the data from A(P+1) to A(2*P
), the first input data is used as the reference value, and the waveform correlation is calculated one by one for each data input, and when the input of data A (2 * P) is completed, the pinch is extracted. I'm trying to make it happen.

Therefore, the time required for pinch extraction is shortened to (2*P) time, and the amount of buffer used is also reduced.

〔Example〕

FIG. 3 shows one embodiment of the present invention, and FIG.
The same parts as in the figure are indicated by the same numbers, and 16 is a P detection circuit that detects the counter value P.

Audio input is sampled at regular intervals, converted to digital data, and input. The counter 11 increments the write address (W) for the dual port memory 12 by counting up each time data is input.
-ADR), thereby writing the audio data to the dual port memory 12.

On the other hand, the P detection circuit 16 detects that the counter value has become P, generates a read address (R-ADR), and reads consecutive P data stored in the dual port memory 12 up to that point. Read out the data as the reference value.

From then on, the read address (R-ADR
), and P pieces of consecutive data, which are sequentially shifted by one from the first input data, are read out from the dual port memory 12 as comparison values and input to the correlation calculation circuit 14.

At the same time, an instruction is given from the P detection circuit 16 to the correlation calculation circuit 14, so that the correlation calculation circuit 14 starts calculating the correlation using the read reference value and comparison value.

When the correlation calculation circuit 14 completes the correlation calculation, the correlation maximum value detection circuit 15 detects the position of the comparison value where the waveform correlation is maximum, and extracts the pitch width from the deviation between this and the reference value.

Such processing is repeated every 2P period, and when the processing ends, the counter 11 is reset and the processing starts from the beginning.

FIG. 4 shows another embodiment of the present invention, in which the same parts as in FIG. is a single port memory.

In the embodiment shown in FIG. 4, the selector 17 receives the write address (WADR) generated by the counter 11 and the read address generated after the P detection circuit 16 detects that the value of the counter 11 becomes P. Address (R-ADR
) are alternately switched and output for each input data.

As a result, the digitized audio data is written to the single port memory 18 at the sampling period, and is written to the single port memory 18 before the P period.
The audio data written in 8 is sequentially read out and input to the correlation calculation circuit 14 as a comparison value. As a result, in the correlation circuit 14, when the P detection circuit 16 detects that the value of the counter 11 becomes P, the single port
Using the data read from the memory 18 as a reference value, correlation calculations are sequentially performed between the comparison value and the data read out from the memory 18.

When the correlation calculation circuit 14 completes the correlation calculation, the correlation maximum value detection circuit 15 detects the position of the comparison value where the waveform correlation is maximum and extracts the pitch width.

Such processing is repeated every 2P period, and when the processing ends, the counter 11 is reset and the processing starts from the beginning.

This kind of configuration is possible because if the memory write and read times are sufficiently short compared to the data sampling cycle, the data for correlation can be read by switching addresses within the sampling data input cycle interval. It's for a reason.

〔Effect of the invention〕

As explained above, according to the present invention, it is possible to reduce the number of buffer memories used for pitch extraction in the pitch extraction circuit, and the time until pitch extraction is shortened. When compressing audio data by applying the present invention to applications such as the above, it becomes possible to shorten the delay time until compressed data is obtained, thereby contributing to improving the performance of the audio multiplexing device.

[Brief explanation of drawings]

FIG. 1 is a diagram showing the basic configuration of the present invention, FIG. 2 is a diagram explaining the present invention in detail, FIG. 3 is a diagram showing one embodiment of the present invention, and FIG. 4 is a diagram showing another embodiment of the present invention. FIG. 5 is a diagram showing a conventional pitch extraction circuit, and FIG. 6 is a diagram illustrating pitch extraction in the conventional method. 1 is a counter, 2 is a storage means, 3 is a P detection circuit, 4 is a correlation calculation circuit, and 5 is a correlation maximum value detection circuit.

Claims (3)

[Claims] (1) A counter (1) that is reset every 2P (P is the maximum detectable pitch period) cycle and counts up each time audio data is input, and input audio data using the count value of the counter (1) as an address. storage means (2) for accumulating the storage means (2);
), the P detection circuit detects the point in time when P pieces of audio data are accumulated, and sequentially reads out the previous P pieces of audio data including the input audio data in the storage means (2) from the time when P+1 pieces of audio data are input. (3) and a correlation calculation circuit (4) that calculates a correlation coefficient with each of the read audio data using the P pieces of audio data initially stored in the storage means (2) as a reference value. and a correlation maximum value detection circuit (5) that detects the pitch period of the audio signal from the interval between the input audio data where the correlation value is the maximum value and the audio data of the reference value. extraction circuit. (2) The storage means (2) is a dual port memory (
6) The pitch extraction circuit according to claim 1, characterized in that the pitch extraction circuit comprises: (3) The storage means (2) includes a single port memory (7) and writes audio data into the single port memory (7) every time audio data is input from the time of the P+1 audio inputs. 2. The pitch extraction circuit according to claim 1, further comprising a selector (8) which alternately performs reading and reading.