JPH0834503B2 - Audio input device for electronic conference system - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION Object of the Invention (Field of Industrial Application) The present invention relates to a voice input device of an electronic conference system capable of removing room noise and reverberation.

(Prior Art) As a conventional electronic conference system, for example, one shown in FIG. 8 has been considered. As shown in the figure, this electronic conference system includes a microphone 1, a speaker 2, a television camera 3 for displaying an image of a conference room, a television camera 4 for displaying an original, a television monitor 5, a facsimile device 6, and a hard copy device as a printer. 7. A voice unit 8 for encoding and decoding voice signals of the microphone 1 and the speaker 2.
Consists of an image unit 9 for encoding and decoding image signals, a control unit 10 for controlling the image unit, an operation unit 11 for inputting instructions from the control unit 9, and a unit 12 for transmitting and receiving audio signals and image signals. To be done.

(Problems to be Solved by the Invention) However, in the audio unit 7 of the conventional electronic conference system as described above, the echo canceller is used to prevent the housing generated when the sound emitted from the speaker is picked up by the microphone 1. However, this echo canceller has a problem that if the room noise level is high, the echo canceling performance is deteriorated by the room noise.

The present invention has been made in view of the above problems, and an object of the present invention is to provide an input device for a voice capable of producing high echo canceling performance even when the level of room noise is high.

[Structure of the Invention] (Means for Solving Problems) The present invention relates to a voice input device for inputting and encoding a voice uttered by a conferees of an electronic conferencing system for conducting a conference between remote points. A plurality of microphones provided corresponding to attendees, a second microphone arranged with directivity in a direction different from that of the attendees, an adaptive filter having a learning function, and learning of the adaptive filter A first learning prohibition control circuit for prohibiting the learning when there is a transmitted voice or a speaker output therein, and input from the second microphone from the voice signal of the conference attendee input from the plurality of microphones. Room noise removal unit that removes room noise by removing the background noise signal that is generated, an adaptive filter with a learning function, and this adaptive filter. A second learning inhibition control circuit that inhibits the learning when there is both a speaker output and a transmitted voice during learning of the filter, and a dereverberation unit that removes reverberation from the room noise-removed voice signal. It is characterized by having.

(Operation) In the present invention, the room noise removing unit removes the background noise signal from the voice signal of the conference attendees, and the reverberation removing unit removes reverberation from the voice signal.

(Example) Hereinafter, an example of the present invention is described in detail based on a drawing.

FIG. 1 is a block diagram showing a conference unit system of an embodiment of the electronic conference system of the present invention.

The control device 900 is connected with input devices of drawing system, still image system, moving image system, and audio system, and is connected with output devices of soft copy system, hard copy system, and audio system. Signals from these input devices and signals to the output device are subjected to selection, multiplexing, and other processing by some control information signals, and transmitted.

As the input device, a plurality of drawing input devices 100 for inputting handwritten characters and graphics, a still image input device 200 for inputting document originals before or during the meeting, and a background image of the meeting attendees meeting place Video input device for inputting
300 and a voice input device 400 for inputting voice are connected.

Further, the output device makes a copy of the flat display device 150 and the graphic display device (1) 160, the graphic display device (1) 160, which displays the input information of its own terminal, the drawing system transmitted, and the still image system. Hard copy device 250, graphic device (2) 35 for displaying input information of the own terminal and moving image system transmitted
A voice output device 450 that outputs the voice input information that is transmitted is connected.

The storage device 600 records and reads the input / output information. The transmission device 700 is connected to the other two conference unit systems at the remote location.

The above input / output information flows according to signals from devices such as an operation input device 850, a state detection device 800 for detecting the presence or absence of a conference participant, and a speaker identification device 510 for determining a speaker from the voice of a talker. The destination is controlled.

 Next, the voice input device will be described.

The voice input device 400 converts the voices of the attendees into an electric signal and takes it into the electronic conference system, and FIG. 2 shows an embodiment thereof.

As shown in the figure, the voice input device 400 has four microphones 402-1 to 402-4 arranged so as to be able to collect the voices of all the attendees, and in a direction different from the conference attendees. Microphone 402- placed with directivity
5, a microphone amplifier (MAMP) 403 that amplifies the output signals of the microphones 402-1 to 402-5, a room noise removal unit (NCAN) 404 that removes room noise, and a reverberation removal unit (ECAN) that removes reverberation. 405, a variable gain unit (AGC) 406 that removes noise from the microphone of a speaker who is not busy, a mixer unit (VMIX) 407 that synthesizes the output signals of the microphones, and voice encoding that compresses and encodes voice (VCOD) 408.

These four microphones 402-1 to 402-4 are arranged in consideration of the directivity so that the voices of all the attendees can be collected, and pick up the voices emitted by the conference attendees. The microphone 402-5 picks up the background noise signal.

A microphone amplifier (MAMP) 403 amplifies the output signal of each microphone to an appropriate level.

 Next, the room noise elimination unit (NCAN) 404 will be described.

The room noise removing unit 404 is a circuit that removes background noise that always exists in the conference room, and uses the background noise signal picked up by the microphone 402-5 whose directionality is different from that of the conference attendees as a reference signal. Signal VB with background noise removed from microphone sounds VA-1 to VA-4 by noise canceller
-1 to VB-5 are generated.

 FIG. 3 is a circuit diagram of the room noise elimination unit 404.

As shown in the figure, it is composed of adaptive filters (NC1 to NC4) 404-1a to 404-1a of the room noise elimination unit 404, subtractors 404-2a to 404-2d, and a learning prohibition control circuit 409. To be done.

The adaptive filters 404-1a to 404-1d generate pseudo noises Yy1 to Yn4 using the output signal VA-5 of the microphone 402-5 as a reference signal. Subtractors 404-2a to 404-2d
Is the microphone 402-1 amplified by the microamplifier 403.
~ 402-4 output signals VA-1 ~ VA-4 subtracting the pseudo noise Yn1 ~ Yn4, the residual signal VB from which the noise is removed.
Outputs -1 to VB-4.

Further, the adaptive filters 404-1a to 404-1d are sequentially learned by a well-known learning identification method or the like so that the residual signal of the reference signal has no correlation. However,
Since learning is disturbed if there is a transmitted voice or speaker output during this learning, this is prevented by the learning prohibition control circuit 409 in the broken line in the figure.

The learning prohibition control circuit 409 includes microphones 402-1 to 402-1.
Level detection circuits (NLV1 to NLV5) for detecting the levels of the output signals VA-1 to VA-4 after amplification of 402-4 409-1a to 409-
1d and level detection circuits (NLVR to NLVL) 409-2 and 409-3 for detecting the levels of the left and right speaker output signals SR and SL
And a maximum value detection circuit (MAX1) 409-4 for detecting the maximum value among the level detection circuits 409.1a to 409-1d, and an adder for generating a sum signal of the level detection circuits 409-2 and 409-3. 409−
5, the output signal of the maximum value detection circuit 409-4 and the adder 409-
The maximum value detection circuit (MAX2) 409-6 that outputs the larger one of the output signals of 5 and the output signal of the level detection circuit 409-7 and the maximum value detection circuit 409-6 are compared, and the level detection circuit The output signal of 409-7 is the maximum value detection circuit 409-
Control signal NA that becomes "1" when it is larger than the output signal of 6
It is composed of a comparator 403-8 which outputs DP.

In the learning prohibition control circuit 409, the maximum value detection circuit 409-4 selects the maximum one of the amplified output signals of the microphones 402-1 to 402-4, and the maximum value detection circuit detects the maximum value. The larger signal is selected between the output signal of the circuit 409-4 and the sum signal of the output signals SR and SL of the left and right speakers.

In this way, the comparator 409-8 determines that the level of the reference signal VA-5 output from the level detection circuit 409-7 is the transmission voice VA-.
1 to VA-4 maximum level and left and right speaker output signals SR, SL
Output to the adaptive filters 404-1a to 404-1d only when it is larger than any of the sum levels of the control signals NADP "404-1a to 404-1d. "Only learn when.

Next, the dereverberation unit (ECAN) 405 will be described. The dereverberation unit 405 has a function of preventing a housing from being generated due to an echo generated when a sound emitted from a speaker is reflected by a wall or a window in a conference room, and a deterioration in call quality due to a reverberation. That is, the left and right speaker signals SR and SL are input and two-channel echo cancellers are used to generate reverberant signals VC-1 to VC-4.

 FIG. 4 is a circuit diagram of this dereverberation unit.

As shown in the figure, the dereverberation unit 405 includes echo cancellers 405-1a to 405-1d and subtractors 405-2a to 405-2d.
And a double-talk detector 410.

The echo cancellers 405-1a to 405-1d generate pseudo reverberation signals Ye1 to Ye4 from the output signals SR and SL of the left and right speakers.

Fig. 5 shows this echo canceller (eg 405-1a)
2 is a circuit diagram showing a configuration of the echo canceller 405-1a as shown in the figure, which comprises two adaptive filters (AF1, AF2) 405-3, 405-4 and an adder 405-5. The right speaker signal SR is input to the file 405-3, the left speaker signal SR is input to the adaptive filter 405-4, and the output signals of these two adaptive filters 405-3 and 405-4 are added to the adder 405-5.
Is added to generate a pseudo reverberation Ye1.

In FIG. 4, subtractors 404-2a to 405-2d subtract the pseudo-reverberation signals Ye1 to Ye4 from the signals VB-1 to VB-4 output from the room noise eliminator 404 to remove reverberation-removed residuals. The difference signals VC-1 to VC-4 are generated.

Further, the learning of the adaptive filters 405-3 and 405-4 is performed by, for example, a well-known learning identification method,
This is realized by performing sequential learning so as to eliminate the correlation between SR, RL, and residual signals VC-1 to VC-4.

Further, the double-talk detector 410 detects the double-talk state and prohibits the learning because learning is disturbed if learning is performed during double-talk in which both speaker output and transmitted voice exist.

The double-talk detector 410 includes level detectors (EVL1 to EVL4) 410-1a to 410-1d for detecting the levels of the output signals VB-1 to VB-4 of the room noise eliminator 404 and left and right speaker signals SL. , SR level detector (EVLL to EVL)
R) 410-2 to 410-3 and level detectors 410-1a to 410-1
It is composed of a maximum value detection circuit (MAX) 410-6 which removes the maximum one of the output signals of d and outputs it.

In this double-talk detector 410, first, the maximum value detection circuit 410-4 is used to detect level detectors 410-1a to 410-1.
The maximum output signal of d is selected, and the adder 410-
5, the sum of the output levels of the left and right speaker signals SL and SR is calculated, and the output signal of the maximum value detection circuit 410-4 and the adder 410 are added.
The level of the output signal of -5 is compared by the comparator 410-6.

This comparator 410-6 is "1" when the output signal of the adder 410-5 is smaller than the output signal of the maximum value detection circuit 410-4.
Signal EADY is generated, and this signal EADY is applied to the adaptive filter 405-1a in the echo cancellers 405-1a to 405-1d.
Input to 3, 405-4, ....

The learning of the adaptive filters 405-3, 405-4, ... Is prohibited when the signal EADY is “1”, that is, when the output level of the speaker is smaller than the maximum level of the voice output from each microphone.

Next, variable gain section (AGC) 406 and mixer section (UMIX) 407
Will be described.

Variable gain unit (AGC) 406 Four input signals V are generated by a speaker identification code TID provided from a speaker identification device 510 described later.
Of the C-1 to VC-4, it selectively amplifies the signal from the microphone with the directivity of the speaker who is currently speaking, and removes unnecessary noise from the microphone. )
Reference numeral 407 is for adding the output signals of the variable gain unit 406.

FIG. 6 is a circuit diagram of the variable gain section 406 and the mixer section 407. The variable gain section 406 is a variable gain amplifier 406-1 to 406-4.
And a control unit 406-5, and the mixer unit 407 is an adder 407-1.

When the speaker identification signal TID given from the speaker identification device 510 is input, the control unit 406-5 operates as a variable gain amplifier other than the variable gain amplifier (for example, 406-1) corresponding to the speaker who is currently speaking. Lower the gain.

The adder 407-1 of the mixer unit 407 is a variable gain amplifier 406-
The output signals of 1 to 406-4 are added.

 Next, the voice encoding unit (VCOD) 408 will be described.

The audio encoding unit (VCOD) 408 compresses the output signal V of the mixer unit 407 according to the transmission speed by the selection signal VRUI of the encoding / decoding algorithm sent from the control device 900,
Control device 900 after encoding and adding an identifier
Is output to

 FIG. 7 is a circuit diagram of the voice encoding unit 408.

As shown in the figure, the voice encoding unit 408 includes a compression algorithm storage unit 408-1 in which a compression algorithm program is stored, a selection switch 408-2 for selecting a compression algorithm, and compression for selecting voice data. It is composed of an encoding unit 408-3 that compresses and encodes with an algorithm and an identifier adding unit 408-4 that adds control information.

This compression algorithm storage unit 408-1 contains 64kADPCM.
Encoding program 408-1a, 32k ADPCM encoding program 4
08-1b, 16k APCAB encoding program 408-1c, and 8kK
The APCB encoding program 408-1d is stored.

First, the selection switch 408-2 is responsive to the selection signal VRUL sent from the control device 900, to select the compression algorithm storage unit 40.
Select one encoding program in 8-1.

Next, the encoding unit 408-3 uses the encoding program selected by the selection switch 408-2 to cause the mixer unit 40 to
The output signal V of 7 is encoded.

Then, the identifier adding unit 408-4 adds the sender identifier (TID) and the device address (SAD) to the encoded signal,
It is output to the control device 900 as a transmission signal VTX.

Therefore, in the present embodiment, the room noise eliminator 404 is provided with a plurality of microphones 402-1 to 402-2 provided corresponding to the attendees of the conference.
Room noise can be removed by removing the background noise signal input from the microphone 402-5 from the voice signal of the conference attendee input from -4.

Although the number of participants in the conference is four in the above embodiment, the present invention is not limited to this.

[Advantages of the Invention] As described in detail above, according to the present invention, echo cancellation can be sufficiently performed even when the level of room noise is high.

[Brief description of drawings]

1 is a block diagram of an electronic conference system, FIG. 2 is a block diagram of a voice input device, FIG. 3 is a block diagram of a room noise removing unit, FIG. 4 is a block diagram of a reverberation removing unit, and FIG. 5 is an echo. FIG. 6 is a block diagram of the canceller, FIG. 6 is a circuit diagram of a variable gain unit and a mixer unit, FIG. 7 is a block diagram of a voice encoding unit, and FIG. 8 is a block diagram of a conventional electronic conference system. 400 ... Voice input device 402-1 to 402-5 ... Microphone 404 ... Room noise removal unit 405 ... Reverberation removal unit 408 ... Voice coding unit

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Norimasa Kudo 3-1-1 Asahigaoka, Hino-shi, Tokyo Inside the Toshiba Hino Plant, Ltd. (72) Inventor Rouichi Nakamura 3-1-1 Asahigaoka, Hino-shi, Tokyo Toshiba Corporation Hino Factory (72) Inventor Hirokazu Sakano 3-1-1 Asahigaoka, Hino City, Tokyo Toshiba Corporation Hino Factory (56) References JP-A-60-116268 (JP, A) JP-A-60 -165165 (JP, A) JP-A-60-154764 (JP, A) JP-A-61-66458 (JP, A) Actually developed JP-A-61-52864 (JP, U)

Claims (1)

[Claims] 1. A voice input device for inputting and encoding a voice uttered by a conferees of an electronic conferencing system for conducting a conference between remote points, comprising a plurality of microphones provided corresponding to the conferees. A second microphone placed with a directivity in a direction different from the direction of the attendees, an adaptive filter having a learning function, and a transmission voice or speaker output during learning of this adaptive filter And a room noise by removing a background noise signal input from the second microphone from a voice signal of a conference attendee input from the plurality of microphones. Room noise eliminator, an adaptive filter with a learning function, and speaker output and transmission voice during learning of this adaptive filter. A second learning prohibition control circuit that prohibits the learning when both of the above are present, and a dereverberation unit that removes reverberation from the voice signal from which room noise is removed. Voice input device.