JP2001306086A - Device and method for deciding voice section - Google Patents

Abstract

PROBLEM TO BE SOLVED: To solve the problem in the convention system that the precision of voice section decision processing becomes deteriorated, if an input wave has abrupt change which is more than assumed in the voice section decision processing. SOLUTION: When the frequency of the voice section decision processing reaches previously set forgetfulness cycles, forgetfulness update processing for a criterion is performed by using an independent parameter which does not depend upon the input wave.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method of dividing an input wave in which an intermittently generated voice signal and environmental noise generated in the surrounding environment are divided into time sections and adapting the input wave to the input wave. The present invention relates to a voice section determination device and a voice section determination method for determining the presence or absence of a voice signal for each section using a threshold.

[0002]

2. Description of the Related Art For example, a battery of a mobile communication terminal is used only when an audio signal for communicating information is present in an input wave between mobile communication terminals that require efficient power saving. There is a need for a battery saving technology that enables communication by using a battery.

[0003] As such a battery saving technique, there is a voice section determination device that divides an input wave for each predetermined time section (also referred to as a frame) and detects and determines the presence or absence of a voice signal. The speech section determination device calculates various analysis parameters from the input wave divided into sections, and determines that a speech signal exists in the input wave when the analysis parameters satisfy a predetermined condition.

[0004] For example, in the past speech section determination apparatus, the input wave power for each section is calculated as an analysis parameter and is compared with a fixed determination threshold. A section having an input wave power equal to or higher than a predetermined determination threshold is determined as a sound section in which a voice is present (satisfies a predetermined condition), and a section having an input wave power lower than the predetermined determination threshold is a silent section in which no voice is present. It is determined to be a section (the predetermined condition is not satisfied). However, since the acoustic properties of audio signals and environmental noise are not always constant, and often change sharply in terms of level and time, a simple method of performing audio section determination processing using a constant determination threshold value Is often not enough.

[0005] For this reason, in recent voice section determination devices,
By various analysis parameters calculated from the input wave,
A method of adaptively changing (adapting) a determination threshold with respect to an input wave has become mainstream. The voice section determination device can be applied not only to voice communication technology including mobile communication, but also to voice recognition technology, noise suppression system, and the like.

[0006] Such a conventional voice section determination device includes:
For example, one disclosed in JP-A-6-266380 can be mentioned. In this voice section determination device, a metric method that uses the characteristics of human-generated voice and noise is used.

FIGS. 9 and 10 are flow charts showing the operation of the conventional voice section judging device. 9 and 1
0, a block BL110 is an audio frame metric calculation processing block for calculating an audio frame metric SFM (k) in the k-th frame, block B
L120 is a determination threshold calculation block that calculates a determination threshold Tm (k) based on the input wave power SE (k) input in the k-th frame.

A block BL140 is a noise frame metric calculation processing block for calculating a noise frame metric NFM (k) in the k-th frame, and a block BL150 is a noise frame metric calculation block in the k-th frame based on the calculated determination threshold value Tm (k). A voice section determination block BL160 for determining the presence or absence of voice is a parameter control block for controlling the voice frame metric SFM (k) and the determination threshold Tm (k).

Next, the operation will be described. When an audio signal of a pulse train mixed with environmental noise is input as an input wave to a conventional audio section determination device, a low-frequency noise component is removed by a filtering process, and after division into a certain frame length, the k-th The frame power SE (k) of the frame (k is a frame number) is calculated as one of the analysis parameters (step ST101).

In step ST111, the frame power SE (k) and the sound frame metric S
FM (k-1) is determined. SE (k) ≧ SF
If M (k−1), the process proceeds to step ST113,
If E (k) <SFM (k-1), step ST11
Move to 2. In step ST113, the audio frame metric SFM (k) of the k-th frame is set to SFM.
The process proceeds to step ST121 as (k) = SE (k).

On the other hand, in step ST112, the voice frame metric SFM (k) of the k-th frame is calculated by equation (1), and the process proceeds to step ST141. Here, BS is the decay time constant of the voice frame metric, and is given by an exponential function of equation (2), where ST is the sampling period and TB is the decay time.

SFM (k) = SFM (k−1) · BS + SE (k) · (1-BS) (1) BS = exp [− (ST / TB)] (2)

In step ST121, it is determined whether the voice section determination result VOUT (k-1) one frame before is a voice section determination or a silent section determination. VOUT (k-1) = 0,
That is, in the case of the silent section determination, in step ST131, the voice section determination result VOUT (k−
The result of VOUT (k-1) is stored in 2), and the process proceeds to step ST141.

On the other hand, in step ST121, VOU
In the case of T (k-1) = 1, that is, in the case of the sound section determination, in step ST122, tmp = SE (k) / fac
The parameter tmp obtained as 1 is calculated, and the process proceeds to step ST123 to determine whether the expression (3) or the expression (4) is satisfied.

Tmp> Tm (k−1) (3) VOUT (k−2) = 0 (4)

If either of the formulas (3) and (4) is satisfied, the k-th determination threshold value Tm (k) = STm in ST124.
tmp, steps ST131 and ST141.
Move to. If none of the equations (3) and (4) is satisfied, the process proceeds to steps ST131 and ST141.

In step ST131, VOUT (k
-2) = VOUT (k-1), step ST
At 141, a comparison is made between SE (k) and the noise frame metric NFM (k-1) one frame before.

If SE (k) ≧ NFM (k−1), then according to equation (5) of step ST142, on the other hand, SE (k) <
If NFM (k-1), the noise frame metric N of the k-th frame is calculated by the equation (6) in step ST143.
FM (k) is calculated. In the equation (5), the parameter BT is an attenuation time constant of the noise frame metric, and is obtained in the same manner as the equation (2).

NFM (k) = NFM (k−1) · BT + SE (k) · (1-BT) (5) NFM (k) = SE (k) (6)

As described above, from step ST101 to block BL110, block BL120, and step ST101.
131, the processing up to the block BL140, the determination threshold Tm (k), the voice frame metric SFM
When (k) and the noise frame metric NFM (k) are respectively calculated, a speech section determination process of the current frame is performed.

That is, in step ST151,
The upper limit determination threshold up is calculated according to the equation (7), and in step ST152, the upper limit determination threshold up and the SFM
A comparison with (k) is made. Here, Tu in equation (7) is an upper limit protection constant.

Up = Tm (k) + NFM (k) · Tu (7)

In step ST152, SFM (k)
If> up is determined, the k-th frame is determined to be a voiced section (VOUT (k) = 1, step ST15)
3) The process proceeds to Step ST161. On the other hand, if it is determined in step ST152 that SFM (k) ≦ up, in step ST154, the lower limit determination threshold value low is calculated according to equation (8), and in step ST155, the lower limit determination threshold value low is compared with SFM (k). Done. Here, Tl in the equation (8) is a lower limit protection constant.

Low = Tm (k) + NFM (k) · Tl (8)

In step ST155, SFM (k)
If it is determined to be <low, the k-th frame is determined to be a silent section (VOUT (k) = 0, step ST15).
6) The process proceeds to Step ST161. On the other hand, if it is determined in step ST155 that SFM (k) ≧ low, the voice section determination result VOUT of the k-th frame
(K) is the voice section determination result VOUT (k
-1) is determined not to change (VOUT (k) = V
OUT (k-1), step ST157), step S
The process moves to T161.

In step ST161, VOUT (k)
= 0 or VOUT (k-1) = 1 is determined. VOUT (k) = 0 or VOUT (k-1) =
1 is satisfied in step ST161, in step ST162, SFM (k) = SE
In step ST163, Tm (k) is calculated as in equation (9) to complete the voice section determination process for the k-th frame, and prepare for the voice section determination process for the next frame. In step ST161, VOUT (k) =
If none of the expressions of 0 or VOUT (k-1) = 1 is satisfied, the voice section determination processing of the k-th frame is completed as it is, and preparation is made for the voice section determination processing of the next frame.

Tbot <= Tm (k) = SE (k) / fac2 (9)

As described above, in the conventional voice section determination device, the determination threshold value Tm (k) adapted to the frame power SE (k) of the input wave is calculated, and the upper limit determination threshold value up is determined based on Tm (k). Since the voice section determination process is performed by calculating the lower limit determination threshold value low and the lower limit determination threshold value low, the determination can be performed in response to a change in the input wave.

[0029]

Since the conventional voice section determination apparatus is configured as described above, if a sudden change more than expected occurs in the input wave during the voice section determination process,
There has been a problem that the accuracy of the adaptation determination threshold is degraded by being affected by the steep change of the input wave, and the accuracy of the subsequent voice segment determination process is degraded.

This is because the decision threshold to be adapted is calculated as a constant value using the decay time constant, and can be applied even to a special situation such as the influence of a sudden change of the input wave that is more than expected. The determination threshold value is adapted, and the accuracy of the subsequent voice section determination process is degraded.

Further, in the conventional voice section determination apparatus, when calculating the determination threshold value adapted to the input wave, it is not considered to exclude the influence of the unnecessary power component of the frame transient section. There is a problem that the accuracy of the determination processing is deteriorated.

The present invention has been made to solve the above-described problem, and suppresses deterioration in accuracy of the subsequent voice section determination processing even when an unexpectedly sharp change occurs in an input wave. It is another object of the present invention to provide a voice section determination device and a voice section determination method capable of performing voice section determination processing with stable accuracy.

Further, the present invention provides a voice section determination apparatus and a voice section determination method capable of preventing the accuracy of voice section determination processing from deteriorating by taking into account the influence of unnecessary power components of a transient part frame. The purpose is to gain.

[0034]

According to the present invention, when the number of times of voice section determination processing reaches a preset forgetting cycle, the voice section determination apparatus of the present invention uses the independent parameter independent of the input wave to determine the determination threshold. Forgetting update processing is performed.

An audio section determining apparatus according to the present invention calculates an analysis parameter dependent on an input wave divided for each section, and performs an input wave analysis by comparing a determination threshold value with the analysis parameter; Based on the result of the input wave analysis, voice section determination means for performing voice section determination processing, adaptive update processing of analysis parameters, and output to the input wave analysis means a determination threshold calculated by the adaptively updated analysis parameters When the number of times of the voice section determination process reaches a preset forgetting cycle, the analysis parameter is subjected to a forgetting update process using an independent parameter that does not depend on the input wave. And a determination threshold updating means for outputting the determined threshold to the input wave analyzing means.

[0036] In the voice section determination apparatus according to the present invention, the input wave analysis means has a low-pass filter and a high-pass filter for processing the input wave, and a normal autocorrelation maximum value of the input wave divided for each section, a low-pass filter. The normal autocorrelation positive maximum value of the input wave processed by the above, the low-pass filter residual signal power, and the input wave power processed by the high-pass filter are used as analysis parameters.

In the voice segment determination apparatus according to the present invention, the determination threshold updating means compares a difference between a plurality of analysis parameters depending on an input wave and a difference threshold, and determines a difference between the plurality of analysis parameters based on the difference threshold. When the difference exceeds a predetermined difference range, an analysis parameter restricting means for performing a limit change process for keeping a difference between a plurality of analysis parameters at a difference limit value is provided.

[0038] In the voice segment determination apparatus according to the present invention, the determination threshold updating means compares the analysis parameter dependent on the input wave with the limit threshold, and determines that the analysis parameter exceeds the limit range determined by the limit threshold. In the limit range for performing the limit update process.

In the voice segment determination apparatus according to the present invention, the determination threshold updating means has a storage means for storing a plurality of independent parameters independent of the input wave, and the number of times of the voice section determination processing for each section is determined in advance. Each time the set forgetting cycle is reached, an independent parameter closest to the analysis parameter depending on the input wave is extracted from the storage means, and a forgetting update process of the determination threshold is performed.

The voice section determining apparatus according to the present invention includes a forgetting cycle adapting means for adapting a forgetting cycle to the analysis parameter calculated by the input wave analyzing means, and the determining threshold updating means comprises a forgetting cycle adapting means. The forgetting update process of the determination threshold is performed according to the forgetting cycle adapted by the above.

In the voice segment determination apparatus according to the present invention, the determination threshold updating means sets the forgetting cycle to be less than the maximum forgetting cycle in the initial frame in which the voice section determining process is started, and sets the forgetting cycle as the voice section determining process proceeds. This is to increase the forgetting cycle to the maximum value.

In the voice segment determination apparatus according to the present invention, the determination threshold updating means calculates the determination threshold using a weighting coefficient independent of the input wave, and assigns a weight in accordance with the analysis parameter calculated by the input wave analysis means. The coefficient is changed.

In the voice section determination method according to the present invention, when the number of times of the voice section determination processing reaches a preset forgetting cycle, the forgetting update processing of the determination threshold is performed using an independent parameter independent of the input wave. It is like that.

The voice section determination method according to the present invention includes a first step in which an analysis parameter depending on an input wave divided for each section is calculated, and an input wave analysis is performed by comparing a determination threshold value with the analysis parameter. A second step in which a speech section determination process is performed based on a result of the input wave analysis, and an adaptive update process of an analysis parameter is performed, and a determination threshold calculated based on the analysis parameter subjected to the adaptive update process is a first threshold value. Output to the step and when the number of times of the voice section determination process reaches a preset forgetting period, a forgetting update process of the analysis parameter is performed using an independent parameter independent of the input wave, and the forgetting update process is performed. And a third step in which the determination threshold value calculated based on the analysis parameter is output to the first step. .

The voice section determination method according to the present invention comprises:
In the step, the input wave is processed by the low-pass filter and the high-pass filter, and the normal auto-correlation maximum value of the input wave divided by each section, the normal auto-correlation positive maximum value of the input wave processed by the low-pass filter, the low-pass filter residual The difference signal power and the input wave power processed by the high-pass filter are used as analysis parameters.

The voice section determination method according to the present invention
In the step, the difference between the plurality of analysis parameters depending on the input wave is compared with the difference threshold, and when the difference between the plurality of analysis parameters exceeds the difference range defined by the difference threshold, the difference between the plurality of analysis parameters is set to the difference limit. A fourth step is provided in which a limit change process for keeping the value is performed.

The voice section determination method according to the present invention
In the step, the analysis parameter dependent on the input wave is compared with the limit threshold, and when the analysis parameter exceeds the limit range defined by the limit threshold, a limit update process for keeping the analysis parameter within the limit range is performed. It is provided with.

The voice section determination method according to the present invention
In the step, a plurality of independent parameters that do not depend on the input wave are stored, and each time the number of times of the voice section determination processing for each section reaches a preset forgetting cycle,
An independent parameter closest to an analysis parameter depending on an input wave is extracted from a plurality of stored independent parameters, and a forgetting update process of a determination threshold is performed.

The voice section determination method according to the present invention comprises:
And a fifth step in which the forgetting cycle is adapted to the analysis parameter calculated in the step (b). In the third step, the forgetting update process of the determination threshold is performed according to the forgetting cycle adapted in the fifth step. It is intended to be performed.

The voice section determination method according to the present invention
In the step, the forgetting cycle is set to be less than the forgetting cycle maximum value in the initial frame in which the voice section determining process is started, and the forgetting cycle is increased to the forgetting cycle maximum value as the voice section determining process proceeds. .

The voice section determination method according to the present invention has a third
In the step (a), the determination threshold value is calculated using a weighting coefficient independent of the input wave, and the weighting coefficient is changed according to the analysis parameter calculated in the first step.

[0052]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below. Embodiment 1 FIG. FIG. 1 is a diagram showing a configuration of a voice section determination device according to Embodiment 1 of the present invention. In FIG.
Reference numeral 10 denotes a pre-processing unit (input wave analysis means) that divides an input wave received by an input terminal into predetermined frames (sections) and performs various processes for analyzing the input wave. Reference numeral 20 denotes an input wave analysis unit (input wave analysis means) which calculates various analysis parameters from each processing of the preprocessing unit 10 and analyzes the input wave. Reference numeral 30 denotes a determination section (voice section determination means) which outputs a voice section determination result V from the analysis result of the input wave analysis section 20 and holds / changes the internal state of the voice section determination apparatus.

Numeral 40 denotes a judgment coefficient updating unit (judgment threshold updating means) which receives a voice section judgment result V of the judgment unit 30 and calculates a sound section judgment coefficient (analysis parameter) and a silent section judgment coefficient calculated from the analysis parameters. (Analysis parameters) is updated. Reference numeral 60 denotes a determination threshold calculating unit (determination threshold updating unit) that calculates a determination threshold from the sound section determination coefficient and the silent section determination coefficient updated by the determination coefficient update unit 40, and outputs the determination threshold to the determination unit 30. I do.

Next, the operation will be described. When an input wave in which an intermittently generated voice signal and environmental noise are mixed is received by an input terminal, the preprocessing unit 10 divides the input wave for each frame to improve the accuracy of the voice section determination process. Is subjected to high-pass filtering. Further, the SNR
(Signal to Noise Ratio, signal power to noise power ratio) is low, and low-pass filter processing is performed on the high-pass filter output in order to remove high-frequency components having low inter-frame correlation of the audio signal.

The input wave analyzer 20 calculates analysis parameters for analyzing the input wave from each processing output of the preprocessor 10. In the first embodiment, the following parameters are used as analysis parameters. That is, the normal autocorrelation maximum value ac of the input wave divided by the frame, the normal autocorrelation positive maximum value rac of the lowpass filter residual signal obtained by performing linear prediction analysis on the lowpass filter output, the lowpass filter residual signal power rp, and This is the frame power fp of the output of the high-pass filter. These analysis parameters are output from the input wave analysis unit 20 to the determination unit 30.

FIGS. 2 and 3 show Embodiment 1 of the present invention.
5 is a flowchart showing the operation of the determination unit 30, the determination coefficient update unit 40, and the determination threshold value calculation unit 60 provided in the voice section determination device according to the first embodiment. The determination unit 30 determines the presence or absence of a voice signal in each frame (voice section determination processing), and the result of the determination is stored in the state memory state.
The state memory state changes in the range from −2 to 2. If the value is 0 or more (state ≧ 0), a sound section is determined. If a negative value (state <0), a silent section is determined.

First, the operation of the determination unit 30 will be described with reference to FIG. In block BL310, processing using the normal maximum positive value rac of the low-pass filter residual signal is performed in the order of the following steps. First, step ST3
At 11, the rac is compared with the determination threshold TH_rac for rac. If rac> TH_rac, the process proceeds to step ST312, and if rac ≦ TH_rac, the process proceeds to step ST321 of block BL320.

In step ST312, the value of state is determined. If state ≧ 0, the process proceeds to step ST313, and if state <0, the process proceeds to step ST315, where state = 0 and the process proceeds to step ST341.
In step ST313, it is determined whether or not state <2. If the state is less than the maximum value 2, the state value is incremented by one in step ST314, and the process proceeds to step ST341, where the maximum value 2 is stored in the state. If it has, the process moves to step ST341.

If NO is determined in the step ST311, the process proceeds to a block BL320. In this block BL320, the low-pass filter residual signal power r
Processing using p is performed in the order of the following steps. In step ST321, the judgment threshold value TH_rp regarding rp is compared with rp. If rp> TH_rp, the process proceeds to step ST322, and if rp ≦ TH_rp, the process proceeds to step ST331 of the block BL330.

In step ST322, state <2
Is determined, and if the state is less than the maximum value 2, the value of the state is set to 1 in step ST323.
Increment and proceed to step ST341,
If the maximum value 2 is stored in ate, step ST34
Move to 1.

If NO is determined in the step ST321, the process proceeds to a block BL330. In this block BL330, processing using the normal autocorrelation maximum value ac of the input wave divided by the frame is performed in the order of the following steps. In step ST331, the judgment threshold value TH_ac regarding ac is compared with ac. ac> TH_a
If c, the process proceeds to step ST332, where ac ≦ TH_
If it is ac, the process proceeds to step ST334.

In step ST332, state <2
Is determined, and if the state is less than the maximum value 2, the value of the state is set to 1 in step ST333.
Increment and proceed to step ST341,
If the maximum value 2 is stored in ate, step ST34
Move to 1.

In step ST334, state>-
It is determined whether it is 1 or not. If the state is larger than -1, the process proceeds to step ST335, the value of the state is incremented by 1, and the process proceeds to step ST341.
If the state is -1 or -2, step ST341
Move to.

Block BL340 and block BL3
At 50, the frame power fp of the high-pass filter output
Are performed in the order of the following steps. In step ST341, frame power differences dfp and df obtained by subtracting fp of the previous frame from fp of the k-th frame.
p is compared with a determination threshold value TH_dfp.
If dfp> TH_dfp, the process proceeds to step ST342, and if dfp ≦ TH_dfp, step ST35.
Move to 1.

In step ST342, state <2
Is determined, and if the state is less than the maximum value 2, the value of the state is set to 1 in step ST343.
Increment and proceed to step ST351,
If the maximum value 2 is stored in ate, step ST35
Move to 1.

In step ST351, the judgment threshold value TH_fp regarding fp is compared with fp. fp <TH_
If fp, the process proceeds to step ST352, where fp ≧ TH
If _fp, the process moves to step ST361. In step ST352, state = −2 and step S352.
The process moves to T361.

Based on the above-described processing using each determination threshold, in block BL360, the voice section determination result V is output from the value of state according to the following steps. In step ST361, the value of the state is determined. If a value from 0 to 2 is stored in the state, the state is determined to be a voiced section in step ST362 (V = 1).
If a value of -1 or -2 is stored in te, it is determined that there is a silent section in step ST363 (V = 0).

Next, the operation of the determination coefficient updating section 40 will be described with reference to FIG. For the sake of simplicity, only the low-pass filter residual signal power rp will be described from each analysis parameter, but other analysis parameters can be processed in the same manner.

In step ST411, the value of the voice section determination result V of the block BL360 is determined. V
If = 1, the process proceeds to block BL420 to perform the processing of steps ST421 and ST422. V = 0
If so, the process proceeds to block BL430 and proceeds to step ST.
431, the process of step ST432 is performed.

In BL420, the sound section determination coefficient Cs_r
Performs processing related to p. Cs_rp is the determination threshold TH_
This is one of the parameters used when calculating rp. First, in step ST421, rp and Cs_r
The absolute value | rp−Cs_rp | of the difference with p is compared with the sound section determination coefficient update threshold value tcs_rp. | Rp
If −Cs_rp | <tcs_rp, step ST
Then, the process proceeds to 422, where the adaptive update processing of Cs_rp is performed using equation (10) so as to adapt to rp, and block BL4
Move to 40. | Rp−Cs_rp | ≧ tcs_rp
If so, the process proceeds to block BL440 without performing the adaptive update process of Cs_rp in equation (10). (10)
ARs_rp in the expression is a smoothing coefficient for Cs_rp, and is a value of 0 or more and less than 1.

Cs_rp = ARs_rp · rp + (1−ARs_rp) · Cs_rp (10)

On the other hand, in BL430, the silent section determination coefficient C
Perform processing related to n_rp. Cn similar to Cs_rp
_Rp is also one of the parameters used when calculating the determination threshold TH_rp. First, in step ST431, the absolute value | rp-Cn of the difference between rp and Cn_rp
_Rp | is compared with a silence section determination coefficient update threshold value tcn_rp. If | rp−Cn_rp | <tcn_rp, the process shifts to step ST432 to perform an adaptive update process of Cn_rp so as to adapt to rp by using equation (11), and shift to block BL440. | Rp-Cn_
If rp | ≧ tcn_rp, then Cn_r in equation (11)
The flow shifts to the block BL440 without performing the adaptive update processing of p. Note that ARn_rp in equation (11) is Cn_r
This is a smoothing coefficient for p, and is a value of 0 or more and less than 1.

Cn_rp = ARn_rp · rp + (1−ARn_rp) · Cn_rp (11)

The block BL440 corresponds to the first embodiment.
Is a processing block that counts the number of times of the voice section determination processing (the section function subjected to the voice section determination processing) count that is repeatedly executed, and counts the number of processing count
Is a processing block for performing a forgetting update process of the voiced section determination coefficient Cs_rp and the silent section determination coefficient Cn_rp using the independent parameters RP_S and RP_N that do not depend on the input wave when a preset forgetting cycle update is reached.

As described above, the sound section judgment coefficient Cs_rp and the silent section judgment coefficient Cn_rp for calculating the judgment threshold value TH_rp regarding the low-pass filter residual signal power rp are determined by the block BL420 and the block B.
Analysis parameter rp depending on input wave in L430
Adaptation update processing is performed so as to adapt to.
Therefore, TH_rp is adapted through rp, Cs_rp, and Cn_rp even for a change in the input wave that is steeper than expected.

Since the change of the input wave that is steeper than expected is a factor of deteriorating the accuracy of the voice section determination processing in the subsequent frame, in the first embodiment, the number of processing times is increased every time the voice section determination processing is performed. The count is incremented, and each time the number of processes count reaches a preset forgetting cycle update, the block BL440
Independent parameters RP_S, RP_ independent of the input wave of
The influence of the input wave is reduced from the determination threshold by the forgetting update process using N, so that even if a steep change of the input wave occurs more than expected, the accuracy of the subsequent voice segment determination process is not deteriorated. I have to.

The forgetting update processing of block BL440 will be described below. In step ST441, it is determined whether or not the value stored in the number of processes count has reached the forgetting cycle update. If count has not reached update, the process shifts to step ST444 to increment count by 1 and returns to step ST60.
Move to 1. The value of “update” is empirically and experimentally determined in consideration of the use conditions of the voice section determination processing.

In step ST601, the judgment threshold value calculation unit 6 uses the sound section judgment coefficient Cs_rp and the silent section judgment coefficient Cn_rp obtained from the judgment coefficient update unit 40.
0 calculates the determination threshold TH_rp from equation (12). In the expression (12), wt_rp is a predetermined weighting coefficient. The calculated TH_rp is output to the determination unit 30 to be used for the voice section determination process of the next frame. In this case, the adaptive update processing of the determination threshold TH_rp is repeated so as to adapt to the input wave.

TH_rp = wt_rp · Cs_rp + (1−wt_rp) · Cn_rp (12)

The voice section determination process is repeated to increase the number of processes count. If it is determined in step ST441 that the count has reached the forgetting cycle update, the process proceeds from step ST441 to step ST4.
Move to 42. In step ST442, the sound section determination coefficient Cs_rp and the silent section determination coefficient Cn_rp
Are performed according to the equations (13) and (14), respectively. Where RP_S and RP
_N is an average value of the analysis parameter rp at the time of sound and at the time of no sound under ideal conditions under which sufficient accuracy of the voice section determination processing is obtained, and is an independent parameter independent of the input wave.

Cs_rp = ARs_rp · RP_S + (1-ARs_rp) · Cs_rp (13) Cn_rp = ARn_rp · RP_N + (1-ARn_rp) · Cn_rp (14)

As can be seen by comparing equations (10) and (13) and equations (11) and (14), (1)
Analysis parameters C depending on input waves in equations (0) and (11)
In the expressions (13) and (14), s_rp and Cn_rp are independent parameters RP_S and RP_N independent of the input wave.
Has been replaced by each. Thus, the processing count c
Every time when the count reaches the forgetting cycle update, the forgetting update process of Cs_rp and Cn_rp is performed using the independent parameters RP_S and RP_N which do not depend on the input wave. In this case, too, the influence of the sudden change of the input wave is reduced by the forgetting update process in the forgetting cycle,
It is possible to perform voice section determination processing with stable accuracy.

According to equations (13) and (14), when the forgetting update process of Cs_rp and Cn_rp is performed in step ST442, the count is reset to 0 in step ST443. Cs_rp and Cn_rp that have been subjected to the forgetting update processing are used to calculate TH_rp in step ST601 and output to the determination unit 30. Thereafter, the voice section determination process adapted to the input wave is repeated again using TH_rp, which has been subjected to the forgetting update process, as the determination threshold value of rp.

As described above, according to the first embodiment, the sound interval determination coefficient Cs_rp and the silent interval determination which are used to calculate the determination threshold value TH_rp and are adapted to the low-pass filter residual signal power rp of the input wave. Coefficient Cn_r
The determination coefficient updating unit 40 sets the independent parameters RP_S and RP_ independent of the input wave every time the number of times count of the voice section determination processing reaches the forgetting cycle update.
Since the forgetting update processing is performed using N, the influence of a change in the input wave that is steeper than expected with respect to the determination threshold TH_rp can be reduced by the forgetting update processing for each forgetting cycle update, and the resistance to the change in the input wave can be reduced. And voice segment determination processing with stable accuracy can be performed.

Further, according to the first embodiment, the pre-processing unit 10 for processing the input wave for each time interval, and the analysis parameters are calculated from the input wave processed by the pre-processing unit 10, An input wave analysis unit 20 that performs an input wave analysis by comparing the input wave analysis with the analysis parameter, a determination unit 30 that determines the presence or absence of a voice signal in a section based on the result of the input wave analysis, and outputs a voice section determination result V; In response to the voice section determination result V, the adaptive update processing of the analysis parameters Cs_rp and Cn_rp is performed, and the number c of processing of the voice section determination processing is performed.
out is a preset forgetting cycle update
, An independent parameter R independent of the input wave
Analysis parameter Cs_r using P_S and RP_N
judgment coefficient updating means 40 for performing a forgetting update process of p and Cn_rp, and a judgment threshold value TH based on analysis parameters Cs_rp and Cn_rp given from the judgment coefficient updating means 40
_Rp, and the judgment threshold value calculation unit 60 that outputs the calculated judgment threshold value to the judgment unit 30.
The influence of the change of the input wave that is steeper than expected on the determination threshold TH_rp can be reduced by the forgetting update processing for each forgetting cycle update, thereby improving the resistance to the change of the input wave and the voice section determination processing with stable accuracy. Is obtained.

Further, according to the first embodiment, the pre-processing section 10 includes the low-pass filter and the high-pass filter for processing the input wave, and the normal auto-correlation maximum value of the input wave divided for each section, the low-pass filter The normal auto-correlation positive maximum value of the input wave processed by the above, the low-pass filter residual signal power and the input wave power processed by the high-pass filter are used by the input wave analysis unit 20 as analysis parameters. The advantage is that the voice section determination process can be performed using the robust and robust analysis parameters.

In the first embodiment, (13),
The forgetting update processing of the sound section determination coefficient Cs_rp and the silent section determination coefficient Cn_rp is performed based on the equation (14). However, the method of the forgetting update processing is not limited to this, and the forgetting update method is adapted to the input wave. Any method can be used as long as it can reduce the influence of the input wave from the determination threshold value.

Embodiment 2 In the second embodiment, the determination coefficient updating unit shown in the first embodiment includes a determination coefficient limiter unit. FIG. 4 is a diagram showing a configuration of a voice section determination device according to Embodiment 2 of the present invention. In FIG. 4, reference numeral 40A denotes a sound section determination coefficient Cs_r.
p and a silent section determination coefficient Cn_rp are determination coefficient update units (determination threshold update means) that perform limit update processing on Cs_rp and Cn_rp while preventing excessive adaptation to an input wave, and are described in Embodiment 1. Judgment coefficient updating unit 40
Determination coefficient limiter 50 (analysis parameter limiting means)
It is provided with. Components that are the same as or correspond to those in FIG. 1 are denoted by the same reference numerals, and redundant description is omitted.

Next, the operation will be described. FIG. 5 is a flowchart showing the operation of the determination coefficient update unit 40A, the determination coefficient limiter unit 50, and the determination threshold value calculation unit 60 provided in the voice section determination device according to the second embodiment of the present invention. The same or equivalent components and operations as those in FIG.

As described in the first embodiment, block B
When the operation up to L420 or the block BL430 is performed, the operation of the determination coefficient limiter unit 50 is performed. For example, when the processing of the block BL420 is performed, the determination coefficient limiter 50 determines whether or not Cs_rp = Cn_rp is satisfied in step ST511. Cs_
If rp ≠ Cn_rp, the process proceeds to step ST531, and if Cs_rp = Cn_rp, step ST5.
Then, the process proceeds to step S12, and the limit update process is performed with Cs_rp = Cn_rp + C.

The operation of block BL520 is also performed in step S
At T521, it is determined whether Cn_rp = Cs_rp is satisfied. If Cn_rp ≠ Cs_rp, then step ST5 is performed.
Step S 41 when Cn_rp = Cs_rp
The process shifts to T522, and performs limit update processing as Cn_rp = Cs_rp-C. Note that, as shown in steps ST512 and ST522, Cs_rp, Cn_rp
Which is to be updated is determined by the voice section determination result V in step ST411, and processing blocks BL510 and BL520 are prepared for this purpose.

The processing here is to prevent Cs_rp and Cn_rp from adapting to unnecessary power components of the transient part frame and to prevent Cs_rp = Cn_rp, and to reduce the difference between Cs_rp and Cn_rp to a predetermined value. C
(Difference limit) only. This Cs_
The method for keeping the difference between rp and Cn_rp only at a predetermined value C (difference limit value) is described in steps ST512 and 522.
Not only in the case of the equal sign, but also with a margin and kept slightly larger than the predetermined value C (Cs_rp ≧ Cn_rp +
C).

The predetermined value C is empirically and experimentally determined in consideration of the use conditions of the voice section determination processing so that the voice section determination processing can be appropriately performed. Note that the conditions of step ST511 and step ST521 are not limited to the case of the equal sign, and the case where Cs_rp and Cn_rp approach to a certain extent, such as when Cs_rp and Cn_rp approach to a certain degree, is determined by the inequality using the difference threshold value that is the limit value of the difference between Cs_rp and Cn_rp. The difference range may be determined. In the case of the equal sign in step ST521, 0 corresponds to the difference threshold.

Block BL510 or block BL5
Subsequent to the processing of 20, the processing of either the block BL530 or the block BL540 is executed. In block BL530, first, in step ST531, Cs
It is determined whether or not _rp is greater than a predetermined value CS_RP_MAX (limit threshold). Cs_rp
If ≦ CS_RP_MAX, the process proceeds to block BL440. However, if Cs_rp> CS_RP_MAX, the process proceeds to step ST532 and Cs_rp = CS_R.
Restriction update processing is performed as P_MAX, and block BL440
Move to.

The operation of block BL540 is also performed in step S
At T541, it is determined whether or not Cn_rp is smaller than a predetermined value CN_RP_MIN (limit threshold value). If Cn_rp ≧ CN_RP_MIN, the process proceeds to block BL440, but Cn_rp <CN_RP_
In the case of MIN, the process proceeds to step ST542, where Cn
Restrict update processing is performed as _rp = CN_RP_MIN, and the process proceeds to block BL440.

In this processing, Cs_rp and Cn_rp are adapted to unnecessary power components of the transient part frame, and Cs_rp exceeds the limit range defined by the upper limit predetermined value CS_RP_MAX, and the lower limit predetermined value CN_RP
Cn_rp is prevented from exceeding the limit range defined by _MIN, and Cs_rp and Cn_rp fall within the limit range.

The method for keeping Cs_rp and Cn_rp within the limit range is as follows: Cs_rp = CS_RP_MAX, C
The method is not limited to the method of n_rp = CN_RP_MIN, but may have a margin and Cs_rp <CS_RP_MA
X, Cn_rp> CN_RP_MIN may be satisfied, and another method, for example, by setting a predetermined coefficient to Cs_rp,
A method of multiplying Cn_rp may be used.

The predetermined values CS_RP_MAX or CN_RP_MAX (CS_RP_MAX, CN_RP
The limit range defined by _MAX) is empirically and experimentally determined in consideration of various use conditions of the voice section determination device. When the processing operation of the determination coefficient limiter unit 50 is performed, as described in the first embodiment, the block BL4
The operations after 40 are performed.

As described above, according to the second embodiment, determination coefficient limiter 50 provided in determination coefficient updater 40A includes two analysis parameters Cs_rp and Cn_rp.
When the difference between Cs_rp and Cn_rp becomes a difference range determined by the difference threshold value by the adaptation, the limit update process is performed so that the difference between Cs_rp and Cn_rp maintains the difference limit value C. Because
It is possible to prevent the influence of unnecessary power components of the transient part frame on the analysis parameters Cs_rp and Cn_rp,
The effect is obtained that the voice segment determination processing can be performed with stable accuracy.

Further, according to the second embodiment, determination coefficient limiter 50 provided in determination coefficient updater 40A.
Is determined by changing the two analysis parameters Cs_rp and Cn_rp to an upper limit predetermined value CS_RP_MAX and a lower limit predetermined value CN.
_RP_MIN, Cs_rp and Cn_rp are set to predetermined values CS_R by adaptation to the input wave.
If P_MAX, CN_RP_ <MIN exceeds the limit range defined by MIN, Cs_rp or Cn_rp
Is performed within a limit range defined by CS_RP_MAX and CN_RP_ <MIN, it is possible to prevent the influence of unnecessary power components of the transient part frame on Cs_rp and Cn_rp, and achieve stable accuracy. This makes it possible to perform the voice section determination process.

Embodiment 3 FIG. In the first embodiment, the determination threshold value adapted to the input wave is subjected to the forgetting update processing for each forgetting cycle. In the third embodiment, the input wave is prevented from being excessively forgotten by the forgetting update processing. I have to. FIG. 6 is a flowchart showing the operation of the judgment coefficient updating unit 40B (judgment threshold updating means) provided in the voice segment judgment apparatus according to Embodiment 3 of the present invention. In FIG. 6, reference numeral 45 denotes a memory (storage means) provided in the determination coefficient updating unit 40B, and the analysis parameter r when sound is present
multiple independent parameters RP_ ideally modeling p
S (0), RP_S (1),..., RP_S (K) and a plurality of independent parameters RP_N (0), RP_
N (1),..., RP_N (J) are stored.

The section determination result V is received from the determination section 30, and the operation up to step ST451 is described in the first embodiment.
Is the same as In the third embodiment, step ST4
In 51, the number of processing count is equal to the forgetting cycle
e, the sound section determination coefficient Cs_rp and the silent section determination coefficient Cn_rp in step ST452.
RP_S (k) and RP_N that approximate
(J) is retrieved from the memory 45 and searched. However,
0 ≦ k ≦ K and 0 ≦ j ≦ J. That is, as shown in FIG. 6, the absolute value of the difference | Cs_rp-RP_S (k) |
| Cn_rp-RP_N (j) | is calculated, and the determination coefficients Cs_rp, Cn_r of step ST453 are determined using RP_S (k) and RP_N (j) that minimize this value.
The forgetting update process of p is performed according to the equations (15) and (16).

Cs_rp = ARs_rp · RP_S (k) + (1-ARs_rp) · Cs_rp (15) Cn_rp = ARn_rp · RP_N (j) + (1-ARn_rp) · Cn_rp (16)

In the forgetting update processing according to the third embodiment, independent parameters RP_S (k) and RP_N that approximate Cs_rp and Cn_rp most closely before the forgetting update processing so that the effect of adaptation is not excessively forgotten based on the determination threshold. (J) is retrieved from the memory 45 by the determination coefficient update unit 40B, and this RP_
Forgetting update processing is performed using S (k) and RP_N (j). RP_S (k) and RP_N (j) are
For example, the SNR of the input wave is changed, or various input waves are modeled and determined. (15),
As can be seen from the equation (16), the independent parameter R that most approximates the analysis parameter Cs_rp adapted to the input wave
Since the forgetting update process of Cs_rp is performed using P_S (k), the forgetting update process of Cs_rp can be performed while preventing the input wave from being forgotten more than necessary based on the determination threshold. The same applies to the silent section determination coefficient Cn_rp.

As described above, according to the third embodiment, the judgment coefficient updating unit 40B has a plurality of independent parameters RP_S (0), RP_S (1) that ideally model the analysis parameter rp when there is sound. ), ..., RP_S
(K) and a plurality of independent parameters RP_N ideally modeling the analysis parameter rp at the time of silence.
(0), RP_N (1),..., RP_N (J) are stored, and when the number of processes count reaches the forgetting cycle update, Cs_rp, Cn_rp
Since the RP_S (k) and RP_N (j) closest to the above are used to perform the forgetting update processing of the sound section determination coefficient Cs_rp and the silent section determination coefficient Cn_rp,
The effect that it is possible to perform a stable voice section determination process that appropriately copes with a sudden change in the input wave that is more than expected, while preventing the effect of the adaptation to the input wave from being unnecessarily forgotten based on the determination threshold. Is obtained.

In the third embodiment, (15),
Although the forgetting update processing of the sound section determination coefficient Cs_rp and the silent section determination coefficient Cn_rp is performed based on the equation (16), the method of the forgetting update processing is not limited to this. Any method can be used as long as it can reduce the influence of the input wave from the determination threshold value.

Embodiment 4 In the above embodiment, the voice section determination process is performed with the forgetting period update being a fixed value. In the fourth embodiment, however, the forgetting period update is used.
The data is adapted to the input wave.

FIG. 7 is a flow chart showing the operation of the power fluctuation calculating section, cycle changing section, and determination coefficient updating section provided in the voice section determining apparatus according to Embodiment 4 of the present invention. In FIG. 7, reference numeral 70 denotes a power fluctuation calculation unit (input wave analysis means), which calculates the power fluctuation dpow (analysis parameter) of the input wave from the frame power fp (analysis parameter) of the input wave. Numeral 80 denotes a cycle changing unit (forgetting cycle changing means) which compares the power fluctuation dpow calculated by the power fluctuation calculating unit 70 with a power fluctuation threshold th_dpow, and when the power fluctuation of the input wave is large, the forgetting period upd.
ate, and outputs the result to the determination coefficient updating unit 40 (determination threshold updating unit). Configurations and operations that are the same as or correspond to those in FIG.

Next, the operation will be described. When the frame power fp of the input signal of the frame to be subjected to the voice section determination processing is input to the power fluctuation calculation section 70, the power fluctuation calculation section 70 calculates H based on the past frame power subjected to the voice section determination processing. The average frame power fp_ave, which is the average value of the frame powers, is (1
7) Calculate according to equation.

Fp_ave = (1 / H) Σ {fp (h)] (h = 1,..., H) (17)

Here, the symbol Σ is the sum of fp (h), that is, fp (1) +.
+ Fp (H). H may be the frame number immediately before the frame having fp or may be the past frame number. Next, using the equation (18), the power fluctuation calculator 70 calculates a power fluctuation dpow, which is a difference between the input frame power fp and the average frame power fp_ave.

Dpow = fp−fp_ave (18)

When the power fluctuation dpow is calculated as described above, the cycle updating unit 80 calculates the power fluctuation dpow.
Then, in step ST801, a comparison is made between the power fluctuation threshold th_dpow and the power fluctuation dpow. When dpow <th_dpow, that is, when the power fluctuation of the input wave is smaller than the power fluctuation threshold, it is considered that the input wave is stable, and the process proceeds to step ST804 to set the forgetting cycle update to the maximum forgetting cycle value UPDA.
Embodiment 1 is adapted as TE (maintaining a forgetting cycle), and
Is performed.

On the other hand, in step ST801, dp
ow ≧ th_dpow, that is, when the power fluctuation of the input wave is equal to or larger than the power fluctuation threshold, the input wave is regarded as largely fluctuating, and the forgetting cycle update is performed so as to cope with a steep change of the input wave more than expected. Is shortened and adaptation is performed (reduction of the forgetting cycle). That is, the frequency of the forgetting update process is increased so that it is possible to cope with a large fluctuation of the input wave. In this case, the mobile terminal makes a transition to step ST802.

In step ST802, the forgetting cycle upd
If it is determined that “ate> UPDATE / 16”, the process proceeds to step ST803 to update = update ·
As (1/2), the forgetting cycle is adapted to a value smaller than the current forgetting cycle so that the frequency of the forgetting update processing increases,
The adaptive update value is output to the determination coefficient update unit 40. On the other hand, when it is determined in step ST802 that the forgetting cycle update is smaller than UPDATE / 16, the adaptation of the forgetting cycle update is stopped, and the update is output to the determination coefficient updating unit 40 as it is. UPDATE / 1 shown here
6 is not particularly limited to this, and another numerical value may be used.

Although not shown in FIG. 7, if the input wave that has fluctuated greatly becomes stable, the process proceeds to step ST80.
The value of the forgetting cycle update, which has been reduced by the process 3, may be increased, and may be adapted (increased forgetting cycle) so as to approach the maximum forgetting cycle value UPDATE. The determination coefficient updating unit 40 performs the same operation as in the first embodiment according to the forgetting cycle update provided from the cycle changing unit 80.

As described above, according to the fourth embodiment,
The power fluctuation calculating unit 70 calculates the frame power fluctuation dpow of the input wave, the cycle updating unit 80 determines the fluctuation of the input wave from the frame power fluctuation dpow, and the value of the forgetting period update according to the determined fluctuation of the input wave. Is applied to the input wave (increase, decrease, and maintain the forgetting period), and the adapted forgetting period update is output to the determination coefficient updating unit 40. Therefore, the forgetting is performed according to the magnitude of the fluctuation of the input wave. The frequency of the update process can be changed, and the effect that the voice segment determination process can be performed with stable accuracy can be obtained.

Although the description has been made using the fluctuation dpow of the frame power of the input wave as an example of the analysis parameter indicating the fluctuation of the input wave, the fourth embodiment is not limited to this. It is also possible to adapt the forgetting cycle update using the analysis parameters.

Embodiment 5 FIG. In the fifth embodiment, the judgment coefficient updating unit 40 (judgment threshold updating unit) sets the forgetting cycle update from the initial frame to the predetermined frame of the voice section judgment processing without depending on the comparison result with the coefficient comparison thresholds tcs_rp and tcn_rp. Assuming that the value is 1, the low-pass residual signal power r obtained in each frame during this period
The speech section determination coefficient Cs_rp and the silent section determination coefficient Cn_rp are updated using only p, and when a predetermined frame has elapsed, the forgetting cycle update is changed to the forgetting cycle maximum value UPDATE (# 1), and the speech section is updated. The determination process may be performed. By doing so, the influence of unnecessary power components of the transient part frame can be prevented, and the adaptation of the decision threshold value TH_rp can be accelerated.

The process of changing the forgetting cycle update from the initial frame to the predetermined frame is performed by changing the forgetting cycle update = 1 to update = UPDA as described above.
Not limited to the method of changing to TE, for example, the forgetting period update is set to 1,
, UPDATE may be changed stepwise so as to prevent inappropriate adaptation to analysis parameters in each frame at the time of system startup. Further, the forgetting cycle update may be started not only from 1 but also from a value smaller than the forgetting cycle maximum value UPDATE.

As described above, according to the fifth embodiment, in the initial frame in which the voice section determination processing is started,
The forgetting cycle upda so that the frequency of the forgetting update processing becomes high.
The value of te is set to a value smaller than the maximum forgetting cycle value UPDATE, and the forgetting cycle u
Since pdate is increased so that it becomes the value of the forgetting cycle maximum value UPDATE in a predetermined frame set in advance, the influence of unnecessary power components of the transient part frame is prevented, and the adaptation of the decision threshold value TH_rp to the input wave is performed quickly. This makes it possible to perform the speech segment determination process with stable accuracy from the initial frame in which the speech segment decision process is started to a predetermined frame.

The predetermined frame in which the forgetting cycle update has the maximum forgetting cycle value UPDATE is determined empirically and experimentally in consideration of the use conditions of the voice section determination processing.

Embodiment 6 FIG. FIG. 8 is a flowchart showing the operation of the judgment threshold value calculation unit provided in the voice segment judgment device according to Embodiment 6 of the present invention. In FIG.
Reference numeral 60A denotes a judgment threshold value calculation unit (judgment threshold value updating means) for adapting the weighting coefficient wt_rp used for calculating the judgment threshold value TH_rp to the input wave. The other components are denoted by the same reference numerals, and redundant description will be omitted.

Next, the operation will be described. When the sound section determination coefficient Cs_rp and the silent section determination coefficient Cn_rp are output from the determination coefficient update section 40 to the determination threshold calculation section 60A, in step ST611, the determination threshold calculation section 6 is determined.
0A is the difference between Cs_rp and Cn_rp and the threshold th_w
Comparison with t_rp is performed as in equation (19).

Cs_rp−Cn_rp <th_wt_rp (19)

When the expression (19) is satisfied, the judgment threshold value calculation unit 6
When 0A is determined (when the conditions of the sound determination and the silence determination are close), the determination threshold calculation unit 60A determines that the SNR of the input wave is low, and proceeds to step ST612, where the weighting coefficient wt_rp Is a predetermined value WT_RP_
Set to HIGH. On the other hand, if Expression (19) is not satisfied, that is, if it is determined that Expression (20) holds, it is determined that the SNR of the input wave is high, and the process shifts to Step ST613 to set the weighting coefficient wt_rp. The value is set to a predetermined value WT_RP_LOW.

Cs_rp−Cn_rp ≧ th_wt_rp (20)

Here, the predetermined value WT_RP_HIGH is 0
It is a value that is larger and less than 1, and a value in which the voice section determination result V is likely to be a voiced section determination (V = 1) is experimentally and empirically obtained. Also, a predetermined value WT_RP_LOW
Is also a value greater than 0 and less than 1, and is experimentally and empirically determined so that the voice section determination process is appropriately performed in a state where the SNR is high. As described above, the influence of environmental noise is determined from the SNR of the input wave, and the weighting coefficient wt_rp is changed to an appropriate value for the voice section determination processing in accordance with the determination, thereby analyzing parameters Cs_rp, Cn_rp for calculating the determination threshold TH_rp. Is adapted to an inappropriate value according to the input wave, the voice section determination processing with stable accuracy can be performed.

As described above, according to the sixth embodiment, the judgment threshold value calculation unit 60A sets the sound section judgment coefficient Cs_
rp and a silent section determination coefficient Cn_rp, and a determination threshold value TH_rp is calculated from a fixed value weighting coefficient wt_rp. Since the estimation threshold is changed and the judgment threshold wt_th is changed according to the estimated input wave, the analysis parameters Cs_rp,
Even when Cn_rp is adapted to an inappropriate value according to the input wave, the voice segment determination processing with stable accuracy can be performed.

In the sixth embodiment, as an example of a method of estimating an input wave, an example using a difference between a sound section determination coefficient and a silent section determination coefficient is used. Determine the weighting coefficient wt
_Rp may be changed.

The above first to sixth embodiments can also be applied to a voice section determination apparatus and a voice section determination method using a determination threshold adapted to another input wave, such as a voice frame metric method. .

[0132]

As described above, according to the present invention, when the number of times of the voice section determination processing reaches a preset forgetting cycle, the forgetting update processing of the determination threshold is performed using an independent parameter independent of the input wave. Since it is performed, the influence of the change of the input wave that is steeper than expected can be reduced by the forgetting update processing for each forgetting cycle, the tolerance to the change of the input wave is increased, and the voice segment determination with stable accuracy is achieved. The effect that processing can be performed is acquired.

According to the present invention, an input wave analyzing means for calculating an analysis parameter depending on an input wave divided for each section and performing an input wave analysis by comparing a judgment threshold value with the analysis parameter; Based on the result, voice section determination means for performing voice section determination processing, and adaptive update processing of analysis parameters, and output a determination threshold calculated based on the analysis parameters subjected to adaptive update processing to input wave analysis means, and When the number of times of the section determination processing reaches a preset forgetting cycle, a forgetting update processing of an analysis parameter is performed using an independent parameter independent of an input wave, and a determination threshold calculated based on the analysis parameter subjected to the forgetting update processing is determined. Since the judgment threshold updating means for outputting to the input wave analyzing means is provided, the influence of the change of the input wave which is sharper than expected The now be reduced by forgetting updating process for each forgetting cycle, enhance the resistance to change of the input wave, there is an advantage that it is possible to perform a stable speech segment determination processing with accuracy.

According to the present invention, the input wave analyzing means has the low-pass filter and the high-pass filter for processing the input wave, and the normal auto-correlation maximum value of the input wave divided for each section is processed by the low-pass filter. Since the normal autocorrelation positive maximum value of the input wave, the low-pass filter residual signal power, and the input wave power processed by the high-pass filter are used as analysis parameters, use analysis parameters having properties that are robust against environmental noise. This makes it possible to perform the voice section determination process.

According to the present invention, the determination threshold updating means includes:
A difference between a plurality of analysis parameters depending on an input wave is compared with a difference threshold, and when a difference between the plurality of analysis parameters exceeds a difference range defined by the difference threshold, the difference between the plurality of analysis parameters is kept at a difference limit value. Since the analysis parameter limiting means for performing the limit change processing is provided, the effect of unnecessary power components of the transient part frame can be prevented, and the effect of performing the voice segment determination processing with stable accuracy can be obtained. can get.

According to the present invention, the determination threshold updating means includes:
An analysis parameter limiting means for comparing an analysis parameter depending on an input wave with a limit threshold and performing a limit update process for limiting the analysis parameter to a limit range when the analysis parameter exceeds a limit range defined by the limit threshold is provided. Therefore, the effect of unnecessary power components of the transient part frame can be prevented, and the effect of performing the voice segment determination processing with stable accuracy can be obtained.

According to the present invention, the determination threshold updating means includes:
It has storage means for storing a plurality of independent parameters that do not depend on the input wave, and each time the number of times of the sound section determination processing for each section reaches a preset forgetting cycle, an analysis parameter that depends on the input wave Since the closest independent parameter is extracted from the storage unit and the forgetting update process of the determination threshold is performed, it is possible to prevent the effect of the adaptation from being unnecessarily forgotten from the determination threshold, and increase the resistance to the change of the input wave, An effect is obtained that stable voice section determination processing can be performed.

According to the present invention, there is provided the forgetting cycle adapting means for adapting the forgetting cycle to the analysis parameter calculated by the input wave analyzing means, and the determination threshold updating means is adapted by the forgetting cycle adapting means. Since the forgetting update process of the judgment threshold is performed according to the forgetting cycle
The frequency of the forgetting update process can be changed according to the input wave, and the effect that the voice segment determination process can be performed with stable accuracy can be obtained.

According to the present invention, the determination threshold updating means includes:
In the initial frame in which the voice section determination process is started, the forgetting cycle is set to be less than the forgetting cycle maximum value, and the forgetting cycle is increased to the forgetting cycle maximum value as the voice section determining process proceeds. The effect of the power component can be prevented, and the adaptation of the determination threshold can be accelerated, so that the effect that the voice section determination processing with stable accuracy can be performed even at the beginning of the voice section determination processing is obtained.

According to the present invention, the determination threshold updating means includes:
The determination threshold is calculated using a weighting coefficient independent of the input wave, and the weighting coefficient is changed according to the analysis parameter calculated by the input wave analysis means. Even when the value is adapted to an inappropriate value according to the wave, an effect is obtained that the voice section determination processing can be performed with stable accuracy.

According to the present invention, when the number of times of the voice section determination processing reaches a preset forgetting cycle,
Since the forgetting update process of the determination threshold is performed using an independent parameter that does not depend on the input wave, the influence of the change of the input wave that is steeper than expected can be reduced by the forgetting update process for each forgetting cycle. This has the effect of increasing the resistance to changes in the input wave and performing the voice segment determination process with stable accuracy.

According to the present invention, the first step in which the analysis parameter depending on the input wave divided for each section is calculated, and the input wave analysis is performed by comparing the determination threshold with the analysis parameter, Based on the result of the above, a second step in which a speech section determination process is performed and an adaptive update process of an analysis parameter are performed, and a determination threshold calculated based on the analysis parameter subjected to the adaptive update process is output to the first step. At the same time, when the number of times of the voice section determination processing reaches a preset forgetting cycle, the forgetting update processing of the analysis parameter is performed using the independent parameter independent of the input wave, and the calculation is performed based on the analysis parameter subjected to the forgetting update processing. And a third step in which the determined determination threshold is output to the first step, so that the input threshold becomes steeper than expected. The influence of the wave changes in will be able to reduce the forgetting process of updating each forgetting cycle, enhance the resistance to change of the input wave, there is an advantage that it is possible to perform a stable speech segment determination processing with accuracy.

According to the present invention, in the first step,
The input wave is processed by the low-pass filter and the high-pass filter, the normal auto-correlation maximum value of the input wave divided for each section, the normal auto-correlation positive maximum value of the input wave processed by the low-pass filter, the low-pass filter residual signal power and Since the input wave power processed by the high-pass filter is used as the analysis parameter, an effect is obtained that the speech section determination process can be performed using the analysis parameter having a property robust to environmental noise.

According to the present invention, in the third step,
A difference between a plurality of analysis parameters depending on an input wave is compared with a difference threshold, and when the difference between the plurality of analysis parameters exceeds a difference range defined by the difference threshold, a limit for keeping the difference between the plurality of analysis parameters at a difference limit value. Since the fourth step in which the change processing is performed is provided, the influence of unnecessary power components of the transient part frame on the analysis parameters can be prevented, and the voice segment determination processing with stable accuracy can be performed. The effect is obtained.

According to the present invention, in the third step,
An analysis parameter dependent on an input wave is compared with a limit threshold, and when the analysis parameter exceeds a limit range defined by the limit threshold, a limit update process for keeping the analysis parameter within the limit range is performed. Therefore, the effect of unnecessary power components of the transient part frame can be prevented, and the effect of performing the voice segment determination processing with stable accuracy can be obtained.

According to the present invention, in the third step,
A plurality of independent parameters independent of the input wave are stored,
Each time the number of times of the voice segment determination processing for each section reaches the preset forgetting cycle, the independent parameter closest to the analysis parameter depending on the input wave is extracted from the stored plurality of independent parameters to determine the determination threshold. Since the forgetting update process is performed, it is possible to prevent the effect of the adaptation from being forgotten more than necessary from the determination threshold value, increase the resistance to changes in the input wave, and perform the voice segment determination process with stable accuracy. The effect that it can be obtained is obtained.

According to the present invention, there is provided a fifth step in which the forgetting period is adapted to the analysis parameter calculated in the first step, and the third step is adapted in the fifth step. Since the forgetting update process of the determination threshold is performed according to the forgetting cycle, the frequency of the forgetting update process can be changed according to the input wave, and the voice segment determination process with stable accuracy can be performed. The effect is obtained.

According to the present invention, in the third step,
In the initial frame in which the voice section determination process is started, the forgetting cycle is set to be less than the maximum value of the forgetting cycle, and the forgetting cycle is increased to the maximum value of the forgetting cycle as the voice section determining process proceeds. Effect of the power section can be prevented, and the adaptation of the decision threshold can be accelerated, so that the effect that the speech section determination processing with stable accuracy can be performed even at the beginning of the speech section determination processing is obtained. .

According to the present invention, in the third step,
The determination threshold value is calculated using the weighting coefficient independent of the input wave, and the weighting coefficient is changed according to the analysis parameter calculated in the first step. Is adapted to an inappropriate value in accordance with the input wave, an effect is obtained that the voice section determination processing with stable accuracy can be performed.

[Brief description of the drawings]

FIG. 1 is a diagram showing a configuration of a voice section determination device according to a first embodiment of the present invention.

FIG. 2 is a flowchart illustrating operations of a determination unit, a determination coefficient update unit, and a determination threshold calculation unit included in the voice section determination device according to the first embodiment of the present invention.

FIG. 3 is a flowchart illustrating operations of a determination unit, a determination coefficient update unit, and a determination threshold calculation unit included in the voice section determination device according to the first embodiment of the present invention.

FIG. 4 is a diagram showing a configuration of a voice section determination device according to a second embodiment of the present invention.

FIG. 5 is a flowchart illustrating operations of a determination coefficient update unit, a determination coefficient limiter unit, and a determination threshold value calculation unit provided in the voice section determination device according to the second embodiment of the present invention.

FIG. 6 is a flowchart illustrating an operation of a determination coefficient updating unit provided in the voice section determination device according to the third embodiment of the present invention.

FIG. 7 is a flowchart illustrating operations of a power fluctuation calculating unit, a cycle changing unit, and a determination coefficient updating unit included in a voice section determining device according to a fourth embodiment of the present invention.

FIG. 8 is a flowchart illustrating an operation of a determination threshold value calculation unit provided in a voice section determination device according to a sixth embodiment of the present invention.

FIG. 9 is a flowchart showing the operation of the conventional voice section determination device.

FIG. 10 is a flowchart showing the operation of a conventional voice section determination device.

[Explanation of symbols]

Reference Signs List 10 preprocessing unit (input wave analysis unit), 20 input wave analysis unit (input wave analysis unit), 30 judgment unit (voice section judgment unit), 40, 40A, 40B judgment coefficient update unit (judgment threshold update unit), 45 Memory (storage means), 50 determination coefficient limiter section (analysis parameter limiting means), 60, 6
0A judgment threshold calculation unit (judgment threshold update unit), 70 power fluctuation calculation unit (input wave analysis unit), 80 cycle change unit (forgetting cycle change means).

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) G10L 101: 14 G10L 9/08 301A

Claims (18)

[Claims] 1. A decision threshold for dividing an input wave in which an intermittently emitted audio signal and an environmental noise generated in the surrounding environment of the audio signal are respectively divided into time sections and adapting the input wave to the input wave. In the voice section determination device that performs the voice section determination processing for each section using the above, when the number of times of the voice section determination processing reaches a preset forgetting cycle, the above-mentioned using an independent parameter independent of the input wave A voice section determination device for performing a forgetting update process of a determination threshold. 2. A decision threshold value for dividing an input wave in which an intermittently emitted audio signal and an environmental noise generated in the surrounding environment of the audio signal are divided into respective time sections, and adapted to the input wave. A voice section determination device that performs a voice section determination process for each section by using the above. An analysis parameter dependent on the input wave divided for each section is calculated, and input is performed by comparing the determination threshold with the analysis parameter. Input wave analysis means for performing a wave analysis; voice section determination means for performing the voice section determination processing based on a result of the input wave analysis; and adaptive analysis processing for the analysis parameter, The judgment threshold value calculated by the parameter is output to the input wave analysis means, and the number of times of the voice segment judgment processing reaches a preset forgetting cycle. Upon reaching, a determination threshold for performing a forgetting update process of the analysis parameter using the independent parameter independent of the input wave, and outputting the determination threshold calculated by the analysis parameter subjected to the forgetting update process to the input wave analysis means. An audio section determination device, comprising: updating means. 3. The input wave analysis means has a low-pass filter and a high-pass filter for processing the input wave, and has a normal autocorrelation maximum value of the input wave divided for each section, and the input signal processed by the low-pass filter. 3. The voice section determination device according to claim 2, wherein a normal autocorrelation positive maximum value of the wave, a low-pass filter residual signal power, and the input wave power processed by the high-pass filter are used as analysis parameters. 4. The determination threshold updating unit compares a difference between a plurality of analysis parameters depending on an input wave and a difference threshold, and determines that the difference between the plurality of analysis parameters exceeds a difference range defined by the difference threshold. 4. The apparatus according to claim 2, further comprising an analysis parameter restriction unit for performing a restriction change process for maintaining a difference between the plurality of analysis parameters at a difference limit value. 5. The determination threshold updating means compares an analysis parameter dependent on an input wave with a limit threshold, and when the analysis parameter exceeds a limit range defined by the limit threshold, sets the analysis parameter to the limit range. 4. The apparatus according to claim 2, further comprising an analysis parameter restriction unit for performing a restriction update process.
The voice section determination device according to the above. 6. The determination threshold updating means has a storage means for storing a plurality of independent parameters independent of an input wave, and the number of times of the voice section determination processing for each section reaches a preset forgetting cycle. 4. The voice section determination device according to claim 2, wherein the independent parameter closest to the analysis parameter dependent on the input wave is retrieved from the storage unit and the forgetting update process of the determination threshold is performed every time. . 7. A forgetting cycle adapting means for adapting a forgetting cycle to an analysis parameter calculated by the input wave analyzing means, wherein the judgment threshold updating means is adapted to the forgetting cycle adapted by the forgetting cycle adapting means. 4. The voice segment determination device according to claim 2, wherein the forgetting update process of the determination threshold is performed according to a cycle. 8. The determination threshold updating means sets the forgetting cycle to be less than the maximum value of the forgetting cycle in the initial frame in which the voice section determining processing is started, and sets the forgetting cycle to the maximum value of the forgetting cycle with the progress of the voice section determining processing. The voice section determination device according to claim 2 or 3, wherein the number is increased. 9. The determination threshold updating means calculates a determination threshold using a weighting coefficient independent of an input wave, and changes the weighting coefficient according to the analysis parameter calculated by the input wave analysis means. The voice section determination device according to claim 2 or 3, wherein: 10. A decision threshold value for dividing an input wave in which an intermittently emitted audio signal and an environmental noise generated in the surrounding environment of the audio signal are respectively divided into time sections and adapting the input wave to the input wave. In the voice section determination method in which the voice section determination processing for each section is performed using the above, when the number of times of the voice section determination processing reaches a preset forgetting cycle, an independent parameter independent of the input wave is used. A voice section determination method, wherein a forgetting update process of the determination threshold is performed. 11. A decision threshold value for dividing an input wave in which an intermittently emitted audio signal and an environmental noise generated in the surrounding environment of the audio signal are respectively divided into time sections and adapting the input wave to the input wave. In the voice section determination method in which the voice section determination processing for each section is performed using, the analysis parameter dependent on the input wave divided for each section is calculated, and the determination threshold and the analysis parameter are compared. A first step in which an input wave analysis is performed; a second step in which the speech section determination processing is performed based on the result of the input wave analysis; and an adaptive update processing of the analysis parameter, the adaptive update processing The determination threshold value calculated based on the analyzed analysis parameter is output to the first step,
When the number of times of the voice section determination process reaches a preset forgetting cycle, a forgetting update process of the analysis parameter is performed using an independent parameter that does not depend on the input wave, and the forgetting update process performs the forgetting update process. A third step in which the calculated determination threshold is output to the first step. 12. In a first step, an input wave is processed by a low-pass filter and a high-pass filter, and a normal autocorrelation maximum value of the input wave divided for each section is calculated. The voice section determination method according to claim 11, wherein the normal autocorrelation positive maximum value, the low-pass filter residual signal power, and the input wave power processed by the high-pass filter are used as analysis parameters. In a third step, a difference between a plurality of analysis parameters dependent on an input wave and a difference threshold are compared, and when the difference between the plurality of analysis parameters exceeds a difference range defined by the difference threshold, 13. The voice segment determination method according to claim 11, further comprising a fourth step of performing a limit change process for maintaining a difference between the plurality of analysis parameters at a difference limit value. 14. In a third step, an analysis parameter dependent on an input wave is compared with a limit threshold, and when the analysis parameter exceeds a limit range defined by the limit threshold, the analysis parameter is set to the limit range. 13. The method according to claim 11, further comprising a fourth step of performing a limit update process.
Voice section determination method described. 15. In a third step, a plurality of independent parameters independent of an input wave are stored.
Each time the number of times the voice section determination processing for each section reaches the preset forgetting cycle, the independent parameter closest to the analysis parameter depending on the input wave is extracted from the stored plurality of independent parameters. 13. The voice segment determination method according to claim 11, wherein a forgetting update process of a determination threshold is performed. 16. A fifth step in which a forgetting cycle is adapted to the analysis parameter calculated in the first step, wherein the third step includes the step of adapting the forgetting cycle adapted in the fifth step. 13. The voice segment determination method according to claim 11, wherein a forgetting update process of the determination threshold is performed according to the following. 17. In a third step, the forgetting cycle is set to be smaller than the maximum forgetting cycle in the initial frame in which the voice section determining process is started, and the forgetting cycle is set to the maximum value of the forgetting cycle as the voice section determining process proceeds. 13. The method according to claim 11, wherein
Voice section determination method described. 18. In a third step, a determination threshold is calculated using a weighting coefficient that does not depend on an input wave, and the weighting coefficient is changed according to the analysis parameter calculated in the first step. The voice section determination method according to claim 11 or 12, wherein: