JPH01183232A - Presence-of-speech detection device - Google Patents

Abstract

PURPOSE:To precisely calculate a noise level and to attain presence-of-speech detection in which malfunction with respect to noise is less by setting a center clipping detection signal and a double talk detection signal to control signals which temporarily stop the update of a threshold. CONSTITUTION:An adaptive power threshold updating unit 264 in a presence-of- speech detection part 26 inputs the center clipping detection signal 512 and the double talk detection signal 514 from the coefficient calculator 126 in an echo cancel part 12. When either of the signals 512 or 514 is detected, an adaptive power threshold is not updated. A presence-of-speech decision unit 262 receives the short frame mean power output 700 of a power calculator 260 and the threshold output 702 of the threshold updating unit 264, compares them, and executes presence-of-speech decision. Thus, the noise level can precisely be calculated and presence-of-speech detection with less malfunction with respect to noise is attained.

Description

DETAILED DESCRIPTION OF THE INVENTION (Industrial Application Field) The present invention is a sound detection device, particularly a digital voice detection device that is advantageously applied to long-distance communication lines where line prices are high, such as international satellite lines or optical submarine cable lines. The present invention relates to a sound detection device in an insertion (O81) system.

(Conventional technology) With the development of communication networks, the demand for long-distance communication lines such as international satellite lines and optical submarine cable lines has increased dramatically, and methods for efficiently using these expensive lines are being considered. .

Such methods include, for example, <
There are O51 method, high efficiency-1, encoding method, etc.

A system that uses the former method is a DS+ system, a transcoder system that uses the latter method, and a digital line multiplexing system that combines the former method and the latter method. All of these systems are approaching the point where they can be put into practical use.

For DSI systems that apply silence compression technology, see, for example, ``Digital Call Voice System'' by Ohta and Ohno, Institute of Electronics and Communication Engineers Paper A.

Volume 56-A, No. 8, pp. 448-455 (1973
The O6I system is described below.

The call time in a telephone conversation includes the time during which the caller is listening to what the other party is saying, and the silent period, which is the pause time between words, phrases, and sentences when the caller is speaking. The DSI system aims to improve line efficiency by taking advantage of the fact that a normal telephone conversation has 50% or more of such silent periods. In other words, this system monitors the status IE of a large number of input channels, selects only the active channel where the caller is talking, and assigns it to the output channel for transmission, thereby reducing the number of transmission channels by half the number of communication channels. This allows you to make phone calls.

Therefore, the DSI system requires a utterance detector that has good sensitivity in detecting the voice part of the near-end speaker and has few malfunctions due to background noise. This is because a voice presence detector with low detection sensitivity causes deterioration of voice quality due to an increase in the number of disconnections at the beginning of speech, and a voice presence detector with many malfunctions due to noise reduces unnecessary voice average operation rate (ratio of voice presence on the line). This is because the increase causes a deterioration in efficiency.

Additionally, the sound detector in the DSI system has the problem of malfunctioning due to echoes in telephone communication using long-distance lines. The subscriber line is a 2-wire system, and the transmission path of a long-distance line is a 4-wire system.
Line 4-wire conversion is performed by the subscriber circuit of the exchange,
Impedance mismatch occurs due to variations in the characteristic impedance of the lines, and the received signal entering the 4-wire receiving side leaks to the 4-wire transmitting side and returns to the speaker side, causing an echo.
Since the DSI system regards such echoes as sound, if the average voice activity rate apparently exceeds 50%, the amount of voice cut-out increases significantly, resulting in a deterioration of efficiency. In order to prevent the harmful effects of such echoes, l
Echo control is essential in the I5I system.

Echo control methods include, for example, an echo suppressor method and an echo canceller method. The echo suppressor method is a method that causes a large loss to the transmission path on the transmitting side when the receiving level is higher than the transmitting level.

For this reason, there is a drawback that the voice is cut off when a call is made in both the sending and receiving directions simultaneously, and a satisfactory call quality cannot be obtained.

The echo canceller method estimates the impulse response of the echo caused by line impedance mismatch, creates a pseudo signal that is almost the same as the actual echo from this impulse response and the received signal, and then subtracts this signal from the actual echo to eliminate the echo. This is a method of canceling. Therefore, there is an advantage that echoes can be suppressed without essentially cutting off the audio. The echo canceller method requires complex arithmetic processing, but with recent advances in digital signal processing technology and semiconductor integrated circuit technology, high-performance and economical methods are being put into practical use. As described above, the echo canceller method has higher call quality than the echo suppressor method, and is therefore suitable for international communication lines where call quality is important.

When applying an echo canceller method to a DSI system, the echo cancellers are connected in series on the subscriber side of the DSI system. Specifically, for example, in a digital electronic exchange, an encoding/decoding circuit (Godec) is installed in a hybrid circuit that performs two-wire and four-wire conversion.
When using the echo canceller method, an echo canceller is inserted between the hybrid circuit and the encoding/decoding circuit.

(Problems to be Solved by the Invention) When a DSI system and an echo canceller are installed together, it is difficult to miniaturize the device, and there are problems in terms of maintenance and operation. Furthermore, in the conventional DSI system, echo cancellers are simply connected in cascade to the DSI system. For this reason, not only the echo component but also the noise component is attenuated, so a sound detector that uses an adaptive power threshold based on the noise power has the problem that the threshold is updated unnecessarily and there are many malfunctions due to noise. .

SUMMARY OF THE INVENTION An object of the present invention is to eliminate such drawbacks of the prior art and provide a DSI system incorporating an echo canceller function that is efficient in device design and manufacturing. Also 1
SUMMARY OF THE INVENTION An object of the present invention is to provide a voice presence detection device that accurately updates an adaptive power threshold based on noise power using a control function included in an echo canceller in a DSI system with a built-in echo canceller function. .

(Means for Solving the Problems) In order to solve the above-mentioned problems, the present invention provides a sound detection device that detects whether or not an input signal includes speech. , a calculation means for calculating the average power of the signal, and receiving the average power of the signal from the calculation means,
threshold updating means for calculating and updating a noise power average level, which is an average level of noise power, based on the average power; and receiving the average power of the signal and the noise power average level and comparing the two to determine whether the signal contains speech or not. and an input terminal for inputting a double talk detection signal and a center clipping detection signal, and when either the double talk detection signal or the center clipping detection signal is input to the input terminal, The threshold updating means suspends calculation and updating of the noise power average level during input, and maintains the power average level at the value before either the double talk detection signal or the center clipping detection signal is input.

Further, according to the present invention, the sound detection device that detects whether or not an input signal includes speech controls echo components included in the input signal, and also controls double talk and center clipping of the input signal. an echo control means for detecting; a calculation means for inputting a signal whose echo component has been controlled from the echo control means and calculating the average power of the signal;
Threshold updating means receives the average power of the signal from the calculation means and calculates and updates a noise power average level which is the average level of noise power using the average power; and receives the average power of the signal and the noise power average level and compares the two. and determining means for determining whether or not the signal includes audio, and when the echo control means detects either double talk or center clipping in the input signal, the echo control means updates the detection with a threshold value updating means. Upon receiving the notification, the threshold updating means does not update the calculation of the average noise power level while receiving the notification.

Further, according to the present invention, there is provided a digital audio insertion system that includes an echo control device that controls echoes generated on a line and inserts another audio signal into a silent section of a signal transmitted on the line. It has a speech presence detection device that detects whether a speech signal is included by comparing it with an adaptive power threshold that is updated based on the noise power of the transmitted signal.

The sound detection device is connected to receive an echo-controlled signal from the echo control device and to be able to input a center clipping detection signal and a double talk detection signal as echo control signals. When either the center fly pink detection signal or the double talk detection signal is input, the adaptive power threshold iQ is not updated.

(Function) According to the present invention, when the sound presence detection device receives either the center clipping detection signal or the double talk detection signal from the echo control device, the average power varies depending on the input signal. The calculation of the average noise power level is not updated even if the noise power level is calculated.

(Example) Prior to detailed description of the present invention, in order to help understanding of the present invention, reference will be made to FIG.
ital 5peech Interpolation
A conventional technique in which an echo canceller section is incorporated into a system (hereinafter referred to as DSI) will be described. What is the DSI system?
As mentioned above, based on the silence compression technology, active voice, which is the actual voice of another call, is inserted into the wall tone section of a call to improve the efficiency of the line. As shown in the figure, conventional DSI
The system is composed of a DSI transmitting device 4 and a DSI receiving device 3, and an echo canceller EC is connected to the device side of the hybrid circuit H for each telephone.

The audio signal sent from the telephone is converted into 2-wire and 4-wire by the hybrid circuit H, and is input to the echo canceller EC. The two-knee canceller EC is designed to prevent impedance mismatching due to variations in the characteristic impedance of the lines, and the received signal entering the 4-wire receiving side of the hybrid circuit leaks to the 4-wire transmitting side, causing an echo. This circuit prevents the DSI system from malfunctioning.

The audio signal output from the echo canceller EC is sent to the PC.
The signal is input to the M encoder, where it is converted to a PCM signal and sent to the multiplexer. The multiplexer multiplexes the input PCM signals and sends them to the DSI transmitter 4 as input PCM signals.

The PCM signal sent from the multiplexer is
The signal is input to the voice delay section and voice detection section of the transmitting device 4. The utterance detector is a utterance detector that detects utterance by updating an adaptive power threshold based on the noise power of background noise. The voice detector monitors each channel of the input PCM signal and, if there is voice, sends a transmission assignment request signal to the assignment processor. Further, the audio delay unit delays the input PCM signal for a certain period of time, and then outputs the signal to the audio memory.

When the allocation processor receives the transmission allocation request signal, the allocation processor controls the audio memory to select the audio signal of the allocated channel from this memory and send it to the output interface, and also sends allocation information to the output interface. Output. The output interface combines the audio signal and the allocation information and transmits the synthesized signal as an output DSI signal to the opposing DSI receiving device 3.

When the output DSI signal transmitted from the DSI transmitter 4 is input as an input DSI signal to the input interface of the DSI receiver 3, the input interface decomposes this signal into an audio signal and allocation information, and stores the audio signal in the audio memory. In addition, the allocation information is output to the allocation processor.

The allocation processor controls the audio memory based on the received allocation information and allocates audio signals to each channel. Since the input audio signal is a signal containing only the voiced portions, it is necessary to insert something equivalent to background noise into the silent portions in order to ensure the naturalness of the conversation. To this end, the allocation processor controls the noise generator so that background noise is inserted into the silent portions of each channel.

The audio signal into which the background noise has been inserted is sent from the DSI receiving device 3 to the demultiplexer as an output PCM signal. A demultiplexer is a circuit that separates multiplexed audio signals for each channel, and the separated audio signals are PCM
sent to the decoder. A PCM decoder converts digital audio signals into analog audio signals. The analog voice signal is sent via an echo canceller and hybrid circuit H to the called telephone.

In this way, in the DSI system according to the prior art.

Since an echo canceller is required for each terminal, it is difficult to make the system smaller and more economical, and there are problems in terms of maintenance and operation. Furthermore, since the utterance detector uses an adaptive power threshold based on noise power, simply connecting an echo canceller to the [lS1 system will attenuate noise components in addition to echo components. For this reason, there is a problem in that a voice detection section using an adaptive threshold based on noise power has a threshold updated unnecessarily and often malfunctions due to noise.

Next, embodiments of the sound detection control method according to the present invention will be described in detail with reference to the accompanying drawings.

Referring to FIG. 1, there is shown an embodiment in which a sound detection device according to the present invention is applied to a DSI system connected to, for example, an international satellite line or an optical submarine cable.

Note that this figure shows elements directly related to the present invention, that is, the transmitting device side of the DSI system, and directly related bulky elements on the receiving device side are omitted.

Analog voice signals sent from a plurality of telephones, for example, connected to the DSI system are converted into digital voice signals by a PCM encoder (not shown) and assigned to an input PCM channel CHI-C)In.
The digital audio signals of PGM channels CHI to CHn are sent to multiplexer 11 as output 500. The multiplexer 11 is a multiplexing device that time-division multiplexes the outputs 500. The multiplexer 11 has a transmission input (Sin
) It is connected to the echo canceling unit 12 via a signal line 502 and outputs the input PCM signal to the echo canceling unit 12.

The voice signal sent from the telephone is, for example, divided into two parts in a hybrid circuit (not shown) in the subscriber circuit of the exchange.
Line 41jit conversion is performed, but impedance mismatch occurs due to variations in the characteristic impedance of the line, as described above. Therefore, the received signal entering the 4-wire receiving side leaks to the 4-wire transmitting side and returns to the transmitting side, causing an echo. Since the ll5I system is based on silence compression technology, if there is such an echo, the average voice operation rate will apparently exceed, for example, 50%, which will dramatically increase the number of voices blocked out and cause a deterioration of efficiency. Become. For this reason, the echo canceling unit 12 uses an echo canceller (echo cancellation) method to cancel echoes occurring in the PCM channels CHI-CHn for each channel. Echo canceling section 12
is connected to the audio delay unit 20 via a transmission output (Sout) signal line 5047i, and is also connected to the n-channel sound detection unit 2
Connected to 6. The echo canceling unit 12 is also It is connected to the sound detection section 2B via a center clipping detection signal line 512 and a double talk detection signal line 514.

The sound detection unit 26 outputs a transmission output 504 which is a sound detection input.
The audio portions of channels CHI to CHn are detected.

Therefore, the utterance detection unit 26 is required to have good detection sensitivity for the near-end speaker's voice part and to have few malfunctions due to background noise. This is because the voice presence detection unit 28, which has low detection sensitivity, deteriorates the voice quality due to an increase in speech cutting, and the voice presence detection unit 2B, which often malfunctions due to noise, causes a decrease in efficiency due to an increase in unnecessary voice average operation rate. It is. Sound detection unit 26
is connected to the assignment processor 28 via the voice detection output 515.

The sound presence detection unit 2B outputs a sound presence detection output 515 to the allocation processor 28 as a channel allocation request signal if a voice signal is present in the transmission output 504.

The audio delay unit 20 delays the input transmission output signal 504 for a certain period of time in order to compensate for the processing delay of the voice detection unit 26. The audio delay unit 20 is connected to the audio memory 22 via an output 506.
The delayed audio signal is sent to memory 22 as output 508.

The allocation processor 28 is configured to, for example,
When the SI gain is 2, the control circuit that controls the audio memory 22 and the output interface 24 transmits the audio signal from the transmitter 2 to the receiver 3 using half the number of channels actually input. be. Allocation processor 28 connects audio memory 22 via memory control output 518.
It is connected to the output interface 24 via the allocation information output 51B. The processor 28 has an input PCM channel and an output OS that has undergone DSI processing.
Assignment information output 518 representing the correspondence with the I channel is sent to the interface 24.

The audio memory 22 stores the audio signal of the corresponding channel as sound in accordance with the control output 516. Audio memory 22 is connected to output interface 24 via output 508.
is connected to the interface 24, and only the sound signals stored in the memory are sent to the interface 24. The interface 24 connects the output 50B of the audio memory 22 and the allocation information output 51 of the processor 28.
8 and synthesizes these outputs to output DSI signal 5.
10 to the opposing DSI receiver row 3.

FIG. 2 shows a functional block diagram of the echo canceling section 12 and the sound detecting section 2B according to this embodiment. In addition,
In the figure, one channel of the echo canceller 12 and the sound detection section 2B is shown for ease of understanding.

The echo canceling section 12 includes an adder 120, a center critter 122, an N-order coefficient digital filter i24, and a coefficient calculator 126. The echo canceling unit 12 estimates the impulse response of the transmission path via the terminal side between the receiving output 522 and the transmitting input 502, that is, the echo path, and combines this impulse response with the receiving human power 522.
An echo pseudo signal Eni that is almost equal to the actual echo input from the input signal is generated, and this signal Eni is subtracted from the actual echo signal En to cancel the echo.

The adder 120 receives the audio signal En+Z as a transmitting input 502.
n and the echo pseudo signal -Eni from the digital filter 124 respectively to obtain one composite signal En+Zn-
This is an adder that creates Eni. Output 6 of adder 120
00 is connected to the center clipper 122 and the coefficient calculator 126, respectively, and outputs the composite signal En+Zn-Eni as an output 600.

Calculator 126 is also connected to receive input line 520,
The transmitter's power 502, the output 800, and the receiver's power 520 are input, and from these, the optimum value of the pseudo echo is estimated and the coefficients are calculated. The calculator 126 is connected to the digital filter 124 via an output 516, and outputs the counted value to the output 51.
6 to output to this. Calculator 126 also has the ability to detect center crimping and double talk by inputting the aforementioned signals. The calculator 126 is
It is connected to the center clipper 122 via the output 521, and to the adaptive power threshold updater 264 of the voice detection unit 26, which will be described later, via the double talk detection signal line 514 and the center clipping detection signal line 512, and detects double talk or center crimping. When detected, a detection report is made for each.

Double talk means that a transmission signal is superimposed on an echo, and FIG. 5 shows a double talk detection area according to this embodiment. That is, the coefficient calculator 126 sets the level threshold of the transmitting human power 502 to -31 dBmO, and the receiving input 52
The level threshold of 0 is -31 dBmO, the vertical axis is the level Ls of the sending human power 502, and the horizontal axis is the level L of the receiving human power 520.
When r is the transmission input level Ls greater than the straight line Ls=Lr−k (k is a constant), it is regarded as double talk detection.

The digital filter 124 has an output 516 whose coefficients have been calculated by estimating the optimum value of the pseudo echo, and a reception input 520.
are input, and a pseudo echo-Eni is created from these.

The filter 124 is connected to the adder 120 via an output 518, and the pseudo echo Eni is connected to the adder 120 as an output 518.
Output to 20.

The center clipper 122 is a circuit that receives the center clipping detection output 520 and removes minute residual echoes by manipulating the waveform of the output 600. FIG. 6 shows an input/output characteristic diagram of the center clipper 122, and as shown in the figure, the clipping voltage Vc changes depending on the received input level.

The sound detection unit 26 includes a power calculator 2 as shown in FIG.
60, utterance determiner 262 and adaptive power threshold updater 2
84. The power calculator 260 inputs the output 504 from the center clipper 122 and calculates the average power of the output 504 for each short frame of, for example, 10 m5. The calculator 126 is connected to the threshold updater 264 and the utterance determiner 282 via the output cuff 00, and outputs the average power calculation result 700 to these.

By inputting the average power calculation result 700, the threshold value updater 264 changes the condition during the silent section, for example, 100m.
The average power of continuous silent sections in five long frames, that is, the average level of noise power Pn is calculated, and the adaptive power threshold level Path is calculated. Path is calculated using the following equation.

Path=K m Pn Here, K is a weighting coefficient and is determined by the variance value of the noise. The adaptive power threshold Path is updated in a silent section with continuous long frames of, for example, 100 m5 or more. The threshold updater 264 is connected to the utterance determiner 262 via the output cuff 02, and the update of the adaptive power threshold Path is outputted to the utterance determiner 262 as the output cuff 02.

Threshold updater 264 receives center clipping detection signal 512 and double talk detection signal 514 from coefficient calculator 126 as described above. and threshold updater 264
When either the center clipping detection signal 512 or the double talk detection signal 514 is detected,
The above-mentioned adaptive power threshold is not updated. This is to prevent the noise level from following the center clipping detection state, since it is extremely reduced, and to prevent the noise level from following it, since the noise level fluctuates greatly in the gobble talk detection state.

The utterance determiner 262 receives the similar frame average power output ZOO of the power calculator 280 and the threshold output cuff 02 of the threshold value update 264, compares these, and determines the utterance. The determiner 262 is connected to the allocation processor 28 via the sound detection output 515, and outputs the sound detection unit cuff 06 to the allocation processor 28 if the average power output cuff 00 is greater than the threshold output cuff 02. .

When the average power output cuff 00 is smaller than the threshold output cuff 02, the utterance determiner 262 outputs the silence detection amount cuff 04 to the threshold updater 264.

Telephone voice PCM signal 500 for channels CHI to CHn
is input to the multiplexer 11, time-division multiplexed thereto, and transmitted to the echo canceling unit 12 as a transmission input 502.
sent to. The transmission input 502 is a signal obtained by adding the voice signal Zn from the near-end speaker and the echo signal En, as described above. Therefore, the echo pseudo signal -E old is calculated by the digital filter 124 and the coefficient calculator 12G,
An adder 120 adds the echo pseudo signals. The signal 600 added by the adder 120 is sent to the center clipper 122 to remove minute residual echoes.

The audio signal output from the center clipper 122 as a transmission output 504 is sent to the delay section 20 and the sound detection section 26.

When the transmission output 504 is input to the calculator 260 of the sound detection section 26, the average power 7 of the transmission output 504 is calculated.
00 is calculated. This average power 700 is determined by the judge 26
2 and output to the threshold updater 264.

Since the level of background noise differs depending on the input audio signal, it is necessary to update the noise value for each signal. The threshold updater 284 calculates the average value Pn of the noise power and outputs it to the determiner 262 as the reference value 702 of the noise power.

The determiner 262 determines whether there is a sound or no sound by comparing the average power 700 and the noise power 702, and if there is a sound, outputs a sound detection output 515 to the processor 28,
If there is no sound, the soundless detection output cuff 04 is updated by the threshold updater 28.
Output to 4.

Note that the threshold value updater 264 uses the double talk detection signal 51
4 and center clipping detection signal 512, the threshold value is not updated. For this reason, the threshold value updater 264 does not update the adaptive power threshold value following a noise level that is extremely reduced or a noise level that has large temporal fluctuations. Therefore, the sound detector 26 can detect the presence of sound with high accuracy.

In this way, in the O51 system according to this embodiment, the echo canceller is disposed closer to the device than the multiplexer 11, so it is no longer necessary to dispose it for each terminal as in the past. As a result, the system can be made smaller and more economical, and maintenance and operation can be carried out smoothly. In addition, the presence detection unit 26
In the adaptive power threshold update of echo canceling unit 1
The center clipping detection signal 512 and the double talk detection signal 514 from 2 are output to the adaptive power threshold updater 264 as control signals for stopping the threshold update.

Therefore, it is possible to calculate the noise level with high accuracy, and it is possible to realize a sound detection device that is less likely to malfunction due to noise.

(Effects of the Invention) As described above, according to the present invention, the center clipping detection signal and the double talk detection signal from the echo control device are used in updating the adaptive power threshold of the voice presence detection device! This is a control signal that temporarily stops updating the V.23 value. Therefore, the noise level can be calculated with high precision, and the presence of sound can be detected with fewer malfunctions due to noise. That is, D
By applying the present invention to an SI system, highly accurate sound detection becomes possible, the average voice activity rate can be reduced, and as a result, an efficient system can be realized.

[Brief Description of the Drawings] Figure 1 shows an embodiment of the sound detection device according to the present invention.
A system configuration diagram applied to the system, FIG. 2 is a functional block diagram of a sound detection device applied to the DSI system of FIG. 1. Figure 3 is a system configuration diagram showing the configuration of a conventional DSI system, Figure 4 is a functional block diagram of the echo canceller of the conventional ll5I system shown in Figure 3, and Figure 5 is a system configuration diagram showing the configuration of a conventional DSI system. FIG. 6 is a detection area diagram showing the detection area, and FIG. 6 is an input/output characteristic diagram showing the input/output characteristics of the center clipper. Explanation of the symbols of important parts 2: DSI transmitter 11, multiplexer 12, n-channel 22-canceller 20, audio delay unit 22, audio memory 24, output interface 2E1 ．． , ,'n-channel sound detection unit 2B, ,assignment processor 120, ,adder 122, ,center clipper 124, ,N-order coefficient digital filter 12B
, ,Coefficient Calculator 280 , ,Power Calculator 282 , ,Speech Determiner 284 , ,Adaptive Power Threshold Updater Patent Applicant Oki Electric Industry Co., Ltd. Agent Takao Katori Takao Maruyama D/A Digital 7 Narok Conversion Conventional Echo canceller Fig. 4 Receiving input level (dBmo) Area of double talk detection Fig. 5 Input/output characteristics of center critter Fig. 6

Claims (1)

[Claims] 1. A sound detection device for detecting whether or not an input signal includes speech, the device comprising: calculation means for inputting the signal and calculating the average power of the signal; , threshold updating means that receives the average power of the signal from the calculation means and calculates and updates a noise power average level that is an average level of noise power using the average power, and receives the average power of the signal and the noise power average level. the input terminal for inputting a double talk detection signal and a center clipping detection signal; When either the signal or the center clipping detection signal is input, the threshold value updating means suspends calculation and updating of the noise power average level during the input, and updates the noise power average level to the double talk detection signal and the center clipping detection signal. A sound detection device characterized in that a clipping detection signal is maintained at a value before input. 2. In a sound detection device that detects whether or not an input signal contains speech, the device controls echo components contained in the input signal, and also controls double talk and center clipping of the input signal. echo control means for detecting; calculation means for inputting a signal whose echo component has been controlled from the echo control means and calculating the average power of the signal; and receiving the average power of the signal from the calculation means and calculating the average power based on the average power. threshold updating means for calculating and updating a noise power average level that is an average level of noise power; receiving the average power of the signal and the noise power average level and comparing the two to determine whether or not the signal includes speech; and determining means for determining whether the input signal is double talk or center clipping, and when the echo control means detects either double talk or center clipping in the input signal, it notifies the threshold updating means of the detection; A sound detection device characterized in that upon receiving a notification, the calculation update of the average noise power level is not performed while the notification is being received. 3. In a digital audio insertion system that has an echo control device that controls echoes generated on a line and inserts another audio signal into the silent section of the signal transmitted on the line, the system includes: a voice presence detection device that detects whether or not the echo control device includes a voice presence detection device by comparing it with an adaptive power threshold that is updated based on the noise power of the transmitted signal; is connected so that it can receive an echo-controlled signal and input a center clipping detection signal and a double talk detection signal as echo control signals, and the sound detection device receives the center clipping detection signal and double talk detection signal from the echo control device. A digital voice insertion system characterized in that the adaptive power threshold is not updated when any one of the detection signals is input.