JP2005148301A - Speech processing system and speech processing method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a speech processing system and speech processing method capable of specifying the speech signal of a main utterer, subjecting the same to translation processing and outputting the translated clear speech signal. <P>SOLUTION: The speech processing system is so constituted that a DSP 25 specifies the main utterer from the speech signals collected from a plurality of microphones MC 1 to MC 6 and the specified one sound collection signal is echo canceled by a DSP 26 and thereafter the sound collection signal is subjected to speech recognition processing, the translation processing and speech synthesis processing in a DSP 40 for translation and is then outputted. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to an audio processing apparatus and an audio processing method for processing audio signals emitted from a plurality of conference attendees, for example.

In recent years, with internationalization, opportunities to communicate with foreigners are increasing. For example, it is common for Japanese to have a conference call with a foreigner in a remote location.
In such a case, the Japanese who conducts the conference call cannot speak a language that can be understood by the foreigner at the conference call, or conversely, the foreigner at the conference call cannot speak Japanese. In some cases, it is difficult to communicate with each other, which hinders the progress of the conference.
Therefore, in the past, a conference was conducted through a person who can interpret, or a specific close-talking microphone was used, a dedicated headset was attached, and the operator translated each speech or performed automatic translation. I went there.

  However, when there are many conference participants in the conference call or when the conference call takes a long time, it is practically difficult to use the dedicated headset described above. That is, in the state where the headset is worn, there is a problem in communication between Japanese who are conference participants, and there is a problem that translation is difficult when a plurality of conference participants speak at the same time.

  The present invention has been made in view of such circumstances, and an object of the present invention is to specify a speech signal of a main speaker, perform translation processing, and output a translated clear speech signal in all directions. A processing device and a voice processing method are provided.

  In order to achieve the above object, a first aspect of the present invention is selected by a plurality of microphones, a microphone signal selection means for selecting one of the collected sound signals of the plurality of microphones, and the microphone signal selection means. Speech recognition means for generating first text data based on the collected microphone sound signal, and text translation means for translating the first text data generated by the voice recognition means to generate second text data And speech synthesizing means for generating a speech signal based on the second text data generated by the text translation means.

  Preferably, the voice recognition means includes voice print recognition means for performing voice print authentication on whether or not the selected microphone sound collection signal matches a voice print registered in advance. Only when it is determined that the voice print obtained from the microphone sound collection signal selected by the signal selection means matches the voice print registered in advance, the first text data is generated based on the microphone sound collection signal. .

  In order to achieve the above object, a second aspect of the present invention includes a step of selecting one of a plurality of microphone sound collection signals, and the first text data based on the selected microphone sound collection signal. Generating, translating the generated first text data to generate second text data, and generating an audio signal based on the generated second text data. This is a voice processing method.

  According to the speech processing apparatus according to the first aspect of the present invention, when the microphone signal selection means selects one of the microphone sound collection signals collected from the plurality of microphones facing the speaker, the speech recognition means. Generates first text data based on the selected microphone sound collection signal, and the text translation means translates the first text data generated by the voice recognition means to generate second text data. The speech synthesis unit generates and generates a speech signal based on the second text data generated by the text translation unit, so that the main speaker is identified from the collected sound signals from a plurality of microphones and translated. Can be performed.

  According to the present invention, the voice signal of the main speaker can be identified and translated, and a translated clear voice signal can be output. Therefore, there is an advantage that communication performance is improved.

Hereinafter, for convenience of description of the first to third embodiments of the present invention to be described later, a communication device (two-way communication device) as a microphone selection means of the speech processing device of the present invention will be described first.
FIGS. 1A to 1C are configuration diagrams showing an example to which the communication device of the present invention is applied.
As illustrated in FIG. 1A, communication devices 1A and 1B are installed in two remote conference rooms 901 and 902, respectively, and these communication devices 1A and 1B are connected by a telephone line 920. Yes.
As illustrated in FIG. 1B, in the two conference rooms 901 and 902, the two-way communication devices 1A and 1B are placed on the tables 911 and 912, respectively. However, in FIG. 1B, only the two-way communication device 1A in the conference room 901 is illustrated for simplification of illustration. The same applies to the two-way communication device 1B in the conference room 902. FIG. 2 shows an external perspective view of the two-way communication devices 1A and 1B.
As illustrated in FIG. 1C, a plurality (six in this embodiment) of conference participants A1 to A6 are located around the two-way communication devices 1A and 1B, respectively. However, in FIG. 1C, only conference participants around the two-way communication device 1A in the conference room 901 are illustrated for simplification. The arrangement of conference participants located around the two-way communication device 1B in the other conference room 902 is the same.

The two-way communication device of the present invention can respond by voice via the telephone line 920 between two conference rooms 901 and 902, for example.
Normally, a conversation via the telephone line 920 is performed by one speaker and one speaker, that is, one-on-one, but the two-way communication device of the present invention uses one telephone line 920. A plurality of conference participants A1 to A6 can talk with each other. Although details will be described later, the number of speakers at the same time (same time zone) is limited to one each other in order to avoid voice congestion.
Since the two-way communication device of the present invention is intended for voice (call), only voice is transmitted via the telephone line 920. In other words, a large amount of image data as in the video conference system is not transmitted. Furthermore, since the two-way communication device of the present invention compresses and transmits conference participants' calls, the transmission burden on the telephone line 920 is light.

Configuration of Interactive Communication Device The configuration of an interactive communication device as an embodiment of the present invention will be described with reference to FIGS.
FIG. 2 is a perspective view of a two-way communication device as an embodiment of the present invention.
FIG. 3 is a sectional view of the two-way communication apparatus illustrated in FIG.
4 is a plan view of the microphone / electronic circuit housing portion of the two-way communication apparatus illustrated in FIG. 1, and is a plan view taken along line X-XY in FIG.

As illustrated in FIG. 2, the two-way communication device 1 includes an upper cover 11, a sound reflection plate 12, a connecting member 13, a speaker housing portion 14, and an operation portion 15.
As illustrated in FIG. 3, the speaker housing 14 includes a sound reflecting surface 14 a, a bottom surface 14 b, and an upper sound output opening 14 c. The reception / reproduction speaker 16 is accommodated in a lumen 14d which is a space surrounded by the sound reflection surface 14a and the bottom surface 14b. The sound reflecting plate 12 is positioned above the speaker housing portion 14, and the speaker housing portion 14 and the sound reflecting plate 12 are connected by a connecting member 13.

  A constraining member 17 passes through the connecting member 13, and the constraining member 17 is between the constraining member / lower fixing portion 14 e on the bottom surface 14 b of the speaker housing portion 14 and the constraining member fixing portion 12 b of the sound reflecting plate 12. Is restrained. However, the restraining member 17 only penetrates the restraining member / penetrating portion 14 f of the speaker housing portion 14. The reason why the restraining member 17 penetrates the restraining member / penetrating portion 14f and is not restrained here is that the speaker housing portion 14 vibrates due to the operation of the speaker 16, but the vibration is restrained around the upper sound output opening 14c. This is to prevent it from happening.

The voice spoken by the speaker in the speaker partner conference room is removed from the upper sound output opening 14c through the reception / reproduction speaker 16, and is transmitted between the sound reflecting surface 12a of the sound reflecting plate 12 and the sound reflecting surface 14a of the speaker accommodating portion 14. It spreads in all directions of 360 degrees around the axis CC along the defined space.
As illustrated, the cross section of the sound reflecting surface 12a of the sound reflecting plate 12 depicts a gentle trumpet arc. The cross section of the sound reflecting surface 12a has a cross-sectional shape illustrated over 360 degrees (omnidirectional) about the axis CC.
Similarly, as illustrated in the cross section of the sound reflection surface 14a of the speaker housing portion 14, a gentle convex surface is drawn. The cross section of the sound reflecting surface 14a also has the illustrated cross sectional shape over 360 degrees (omnidirectional) about the axis CC.

The sound S emitted from the reception / reproduction speaker 16 passes through the upper sound output opening 14c, passes through a sound output space having a trumpet-shaped cross section defined by the sound reflection surface 12a and the sound reflection surface 14a, and the voice response device 1 Along the surface of the placed table 911, the sound spreads in all directions 360 degrees around the axis C-C, and is heard at a volume equal to all conference participants A1 to A6. In the present embodiment, the surface of the table 911 is also used as part of the sound propagation means.
The diffusion state of the sound S output from the receiving / reproducing speaker 16 is shown by arrows.

The sound reflecting plate 12 supports the printed circuit board 21.
On the printed circuit board 21, as illustrated in a plan view in FIG. 4, the microphones MC <b> 1 to MC <b> 6, the light emitting diodes LED <b> 1 to 6, the microprocessor 23, the codec (CODEC) 24, and the first digital signal Various electronic circuits such as a processor (DSP 1) DSP 25, a second digital signal processor (DSP 2) DSP 26, an A / D converter block 27, a D / A converter block 28, and an amplifier block 29 are mounted on the sound reflector. Reference numeral 12 also functions as a member that supports the microphone / electronic circuit housing portion 2.

  The printed circuit board 21 has a damper 18 that absorbs vibration from the reception / reproduction speaker 16 so that vibration from the reception / reproduction speaker 16 is transmitted to the sound reflector 12 and does not enter the microphones MC1 to MC6. It is attached. The damper 18 includes a screw and a cushioning material such as an anti-vibration rubber inserted between the screw and the printed board 21, and the cushioning material is screwed to the printed board 21 with a screw. That is, the vibration transmitted from the reception / reproduction speaker 16 to the printed circuit board 21 is absorbed by the buffer material. Thereby, the microphones MC1 to MC6 are not affected by the sound from the speaker 16.

As shown in FIG. 4, six microphones MC <b> 1 to MC <b> 6 are located radially from the central axis C of the printed circuit board 21 at equal intervals (60 degrees in this embodiment). Each microphone is a unidirectional microphone. Its characteristics will be described later.
Each of the microphones MC1 to MC6 is swingably supported by a first microphone support member 22a and a second microphone support member 22b, both of which are flexible or elastic (in order to simplify the illustration, the microphones Only the first microphone support member 22a and the second microphone support member 22b in the MC1 portion are illustrated), and is not affected by the vibration from the reception / reproduction speaker 16 by the damper 18 using the above-described cushioning material. In addition to the countermeasures, the first microphone support member 22a and the second microphone support member 22b having flexibility or elasticity absorb the vibration of the printed circuit board 21 that is vibrated by the vibration from the reception / reproduction speaker 16, and reproduce the reception. The noise of the receiving / reproducing speaker 16 is avoided so as not to be affected by the vibration of the speaker 16.

As illustrated in FIG. 3, the reception / reproduction speaker 16 is oriented perpendicularly to the central axis CC of the plane on which the microphones MC1 to MC6 are located (in the present embodiment, it is directed upward) With the arrangement of the reception / reproduction speaker 16 and the six microphones MC1 to MC6, the distance between the reception / reproduction speaker 16 and each of the microphones MC1 to MC6 is equal. The sound reaches the microphones MC1 to MC6 with almost the same volume and phase. However, due to the configuration of the sound reflecting surface 12a of the sound reflecting plate 12 and the sound reflecting surface 14a of the speaker housing portion 14, the sound of the receiving and reproducing speaker 16 is not directly input to the microphones MC1 to MC6. In addition, as described above, by using the damper 18 using the buffer material, the first microphone support member 22a and the second microphone support member 22b having flexibility or elasticity, the reception / reproduction speaker 16 is provided. The influence of vibration is reduced.
As shown in FIG. 1C, the conference participants A1 to A6 are usually almost equal to the vicinity of the microphones MC1 to MC6 arranged at intervals of 60 degrees in the direction of 360 degrees around the voice response device 1. Located at intervals.

Light- emitting diodes Light-emitting diodes LED1 to 6 are arranged in the vicinity of the microphones MC1 to MC6 as means for reporting that a speaker to be described later has been determined.
The light emitting diodes LED1 to 6 are provided so as to be visible from all the conference participants A1 to A6 even when the upper cover 11 is attached. Therefore, the upper cover 11 is provided with a transparent window so that the light emitting states of the light emitting diodes LED1 to LED6 can be visually recognized. Of course, the upper cover 11 may be provided with openings in the portions of the light emitting diodes LEDs 1 to 6, but a light transmitting window is preferable from the viewpoint of dust prevention to the microphone / electronic circuit housing portion 2.

The printed circuit board 21 includes a first digital signal processor (DSP1) 25, a second digital signal processor (DSP2) 26, and various electronic circuits 27 to 29 for performing various signal processing described later. It is arranged in a space other than the part where MC6 is located.
In the present embodiment, the DSP 25 is used as signal processing means for performing processing such as filter processing and microphone selection processing together with various electronic circuits 27 to 29, and the DSP 26 is used as an echo canceller.

FIG. 5 is a schematic configuration diagram of the microprocessor 23, the codec 24, the DSP 25, the DSP 26, the A / D converter block 27, the D / A converter block 28, the amplifier block 29, and other various electronic circuits.
The microprocessor 23 performs overall control processing of the microphone / electronic circuit housing unit 2. The codec 24 compresses and encodes the voice to be transmitted to the other party conference room.
The DSP 25 performs various signal processing described below, such as filter processing and microphone selection processing.
The DSP 26 functions as an echo canceller.
In FIG. 5, four A / D converters 271 to 274 are illustrated as an example of the A / D converter block 27, and two D / A converters are illustrated as an example of the D / A converter block 28. The converters 281 to 282 are illustrated, and two amplifiers 291 to 292 are illustrated as an example of the amplifier block 29.
In addition, as the microphone / electronic circuit housing portion 2, various circuits such as a power supply circuit are mounted on the printed circuit board 21.

In FIG. 4, a pair of microphones MC1-MC4: MC2-MC5: MC3-M6 arranged in a straight line at symmetrical (or opposite) positions with respect to the central axis C of the printed circuit board 21 each have two channels. The analog signals are input to A / D converters 271 to 273 that convert digital signals. In this embodiment, one A / D converter converts a 2-channel analog input signal into a digital signal. Therefore, the detection signals of two (one pair) microphones, for example, microphones MC1 and MC4, which are positioned on a straight line across the central axis C, are input to one A / D converter and converted into digital signals. Yes. Further, in this embodiment, in order to identify the speaker of the voice to be sent to the other party's conference room, the difference between the two microphones positioned on a straight line, the volume of the voice, etc. are referred to. When the signals of two microphones located on the line are input to the same A / D converter, the conversion timing is also substantially the same, and there is little timing error when taking the difference between the audio outputs of the two microphones. There are advantages such as being easy.
The A / D converters 271 to 274 can also be configured as A / D converters 271 to 274 with a variable gain amplification function.
The collected sound signals of the microphones MC1 to MC6 converted by the A / D converters 271 to 274 are input to the DSP 25, and various signal processing described later is performed.
As one of the processing results of the DSP 25, the result of selecting one of the microphones MC1 to MC6 is output to the light emitting diodes LED1 to LED6.

The processing result of the DSP 25 is output to the DSP 26 and an echo cancellation process is performed. The DSP 26 includes, for example, an echo cancellation transmission processing unit and an echo cancellation reception unit.
The processing result of the DSP 26 is converted into an analog signal by the D / A converters 281 to 282. The output from the D / A converter 281 is encoded by the codec 24 as necessary, and output to the line-out of the telephone line 920 (FIG. 1 (A)) via the amplifier 291 to the partner conference room. The sound is output as a sound through the reception / reproduction speaker 16 of the installed voice response device 1.
Voice from the two-way communication device 1 installed in the other party's conference room is input via the line-in of the telephone line 920 (FIG. 1A), converted into a digital signal by the A / D converter 274, The signal is input to the DSP 26 and used for echo cancellation processing. In addition, the voice from the two-way communication device 1 installed in the other party's conference room is applied to the speaker 16 through a route (not shown) and output as sound.
The output from the D / A converter 282 is output as a sound from the reception / reproduction speaker 16 of the bidirectional communication apparatus 1 via the amplifier 292. In other words, in addition to the voice of the speaker selected in the other party's conference room from the reception / reproduction speaker 16 described above, the conference participants A1 to A6 also use the reception / reproduction speaker 16 for the voice uttered by the speaker in the conference room. Can be heard through.

Microphones MC1 to MC6
FIG. 6 is a graph showing the characteristics of the microphones MC1 to MC6.
Each unidirectional characteristic microphone changes its frequency characteristic and level characteristic as illustrated in FIG. 6 depending on the arrival angle of sound from the speaker to the microphone. The plurality of curves indicate directivity when the frequency of the sound collection signal is 100 Hz, 150 Hz, 200 Hz, 300 Hz, 400 Hz, 500 Hz, 700 Hz, 1000 Hz, 1500 Hz, 2000 Hz, 3000 Hz, 4000 Hz, 5000 Hz, and 7000 Hz. However, in order to simplify the illustration, FIG. 6 typically illustrates the directivity for 150 Hz, 500 Hz, 1500 Hz, 3000 Hz, and 7000 Hz.

7A to 7D are graphs showing the analysis results of the position of the sound source and the sound collection level of the microphone. A speaker is placed at a predetermined distance, for example, a distance of 1.5 meters, from the two-way communication device 1. The result of performing fast Fourier transform (FFT) on the sound collected by each microphone at regular time intervals is shown. The X axis represents frequency, the Y axis represents signal level, and the Z axis represents time.
When the microphone having directivity shown in FIG. 6 is used, strong directivity is shown in front of the microphone. In the present embodiment, using such characteristics, the DSP 25 performs a microphone selection process.

When a non-directional microphone is used instead of a directional microphone as in the present invention, since all sounds around the microphone are collected, the S / N between the voice of the speaker and the ambient noise is confused. Good sound cannot be collected. In order to avoid this, in the present invention, S / N with surrounding noise is improved by collecting sound with one directional microphone.
Furthermore, a microphone array using a plurality of omnidirectional microphones can be used as a method for obtaining the directivity of the microphone. However, in such a method, the time axis (phase) of a plurality of signals is complicated, and thus complicated. Since processing is required, it takes time and response is low, and the apparatus configuration is complicated. That is, the DSP signal processing system also requires complicated signal processing. The present invention solves such a problem by using the directional microphone illustrated in FIG.
Further, in order to synthesize a microphone array signal and use it as a directional sound pickup microphone, there is a disadvantage that the outer shape is restricted by the pass frequency characteristic and the outer shape becomes large. The present invention also solves this problem.

Effects of the device configuration of the communication device The communication device configured as described above exhibits the following advantages.
(1) The positional relationship between the even number of microphones MC1 to MC6 arranged radially at equal angles and at equal intervals and the reception / reproduction speaker 16 is constant, and the distance from the reception / reproduction speaker 16 is very short. The level at which the output sound returns directly to the microphones MC1 to MC6 via the conference room (room) environment is overwhelmingly dominant. Therefore, the characteristics (signal level (intensity), frequency characteristics (f characteristic), phase) that the sound reaches from the speaker 16 to the microphones MC1 to MC6 are always the same. That is, there is an advantage that the two-way communication device 1 in the embodiment of the present invention always has the same transfer function.
(2) Therefore, there is no change in the transfer function when the output of the microphone sent to the other party's conference room is switched when the speakers are different, and there is no need to adjust the gain of the microphone system every time the microphone is switched. Have advantages. In other words, there is an advantage that once the adjustment is made at the time of manufacturing the interactive communication apparatus, it is not necessary to redo the adjustment.
(3) Even if the microphones are switched when the speakers are different for the same reason as described above, only one echo canceller (DSP 26) is required. The DSP is expensive, and it is not necessary to arrange a plurality of DSPs on the printed circuit board 21 on which various members are mounted and the space is small, and the space for arranging the DSPs on the printed circuit board 21 may be small. As a result, the printed circuit board 21, and thus the communication device of the present invention can be made smaller.
(4) As described above, since the transfer function between the reception and reproduction speaker 16 and the microphones MC1 to MC6 is constant, for example, the sensitivity difference adjustment of the microphone itself having ± 3 dB can be performed by the microphone unit of the two-way communication device alone. There is an advantage. Details of the sensitivity difference adjustment will be described later.
(5) The table on which the two-way communication device 1 is mounted normally uses a round table or a polygonal table. However, a single reception / reproduction speaker 16 in the two-way communication device 11 can provide sound of equal quality. A loudspeaker system that is uniformly distributed (diffused) in all directions of 360 degrees around the axis C has become possible.
(6) The sound emitted from the receiving / reproducing speaker 16 is transmitted to the table surface of the round table (boundary effect), effectively and evenly delivering high-quality sound to the conference participants, and facing the ceiling direction of the conference room There is an advantage that the sound and phase on the side are canceled and become a small sound, and there are few reflected sounds from the ceiling direction to the conference participants, and as a result, a clear sound is distributed to the participants.
(7) Since the sound emitted from the reception / reproduction speaker 16 reaches all the microphones MC1 to MC6 arranged radially and at equal intervals at the same angle at the same volume at the same time, it is determined whether the sound is the voice of the speaker or the received voice. It becomes easy. As a result, erroneous determination of microphone selection processing is reduced. Details thereof will be described later.
(8) Even number, for example, six microphones are arranged at equal angles radially and at equal intervals, and a pair of opposing microphones are arranged in a straight line, so that level comparison for direction detection can be easily performed.
(9) By the damper 18, the microphone support member 22, and the like, it is possible to reduce the influence of the vibration due to the sound of the reception and reproduction speaker 16 on the sound collection of the microphones MC1 to MC6.
(10) As illustrated in FIG. 3, structurally, the sound of the reception / reproduction speaker 16 does not directly propagate to the microphones MC1 to MC6. Therefore, in the two-way communication apparatus 1, the influence of noise from the reception / reproduction speaker 16 is small.

The communication device 1 described with reference to FIGS. 2 to 3 has the reception reproduction speaker 16 disposed in the lower portion and the microphones MC1 to MC6 (and related electronic circuits) disposed in the upper portion. The positions of 16 and microphones MC1-MC6 (and associated electronic circuitry) can also be turned upside down as illustrated in FIG. Even in such a case, the above-described effects are exhibited.

  The number of microphones is not limited to six, and any number of microphones such as four, eight, etc. may be arranged in a straight line (in the same direction) with a plurality of pairs of axes C radially and equally spaced at the same angle. They are arranged in a straight line like MC1 and MC4. The reason why the two microphones MC1 and MC4 are arranged to face each other is to select a microphone and specify a speaker.

Signal Processing Contents Hereinafter, processing contents mainly performed by the first digital signal processor (DSP) 25 will be described.
FIG. 9 is a diagram illustrating an outline of processing performed by the DSP 25. The outline is described below.

(1) Measurement of ambient noise As an initial operation, preferably, ambient noise where the two-way communication device 1 is installed is measured.
The two-way communication device 1 can be used in various environments (conference rooms). In order to improve the performance of the two-way communication device 1 in order to ensure the accuracy of selection of the microphone, in the present invention, noise in the surrounding environment where the two-way communication device 1 is installed is measured in the initial stage. It is possible to eliminate the influence from the signal collected by the microphone.
Of course, when the two-way communication apparatus 1 is repeatedly used in the same conference room, noise measurement is performed in advance, and this process can be omitted when the noise state does not change.
Note that noise measurement can also be performed in a normal state.
Details of the noise measurement will be described later.

(2) Selection of Chairperson For example, when the two-way communication device 1 is used for a two-way conference, it is beneficial to have a chairman who manages the proceedings in each conference room. Therefore, as one aspect of the present invention, the chairperson is set from the operation unit 15 of the interactive communication device 1 in the initial stage of using the interactive communication device 1. As a chairperson setting method, for example, the first microphone MC1 located in the vicinity of the operation unit 15 is used as a chairperson microphone. Of course, the chairman's microphone can be arbitrary.
Note that this processing can be omitted when the chairperson who repeatedly uses the interactive communication device 1 is the same. Or you may decide the microphone of the position where a chairperson sits beforehand. In that case, there is no need to select a chairman each time.
Of course, the selection of the chair is not limited to the initial state, and can be performed at any timing.
Details of the chairperson selection will be described later.

(3) Microphone sensitivity difference adjustment As an initial operation, preferably, the gain or attenuation unit of the amplification unit that amplifies the signals of the microphones MC1 to MC6 so that the acoustic coupling between the reception reproduction speaker 16 and the microphones MC1 to MC6 is equal. Automatically adjust the attenuation value.
The sensitivity difference adjustment will be described later.

Various processes exemplified below are performed as normal processes.
(4) Microphone selection / switching process When a plurality of conference participants make a call at the same time in one conference room, voices are mixed and difficult for the conference participants A1 to A6 in the other conference room. Therefore, in the present invention, in principle, one person is allowed to talk at a time. For this reason, the DSP 25 performs microphone selection / switching processing.
As a result, only the call from the selected microphone is transmitted to the voice response device 1 in the other party conference room via the telephone line 920 and output from the speaker. Of course, as described with reference to FIG. 5, the LED in the vicinity of the selected speaker's microphone is turned on, and the selected speaker's voice is also output from the speaker of the interactive communication device 1 in the room. Can hear and recognize who is an authorized speaker.
The purpose of this processing is to select a signal from a unidirectional microphone facing the speaker and send a signal having a good S / N to the other party as a transmission signal.
(5) Display of the selected microphone Lights so that all the conference participants A1 to A6 can easily recognize which conference participant's microphone is selected and allowed to speak. The corresponding ones of the diodes LEDs 1 to 6 are turned on.
(6) As a background art of the microphone selection process described above, or in order to accurately perform the microphone selection process, various signal processes exemplified below are performed.
(A) Band separation and level conversion processing of microphone collected signal (b) Start / end determination processing of speech
To be used as a trigger to start selecting the microphone signal that faces the speaker direction.
(C) Speaker direction microphone detection processing
To analyze the collected sound signal of each microphone and determine the microphone used by the speaker.
(D) Speaker direction microphone switching timing determination processing, and microphone signal selection switching processing facing the detected speaker
An instruction to switch to the microphone selected from the above processing result is given. (E) Measurement of floor noise during normal operation

Measurement of floor (environment) noise This process is divided into an initial process and a normal process immediately after the two-way communication device is turned on.
This process is performed under the following exemplary preconditions.

Immediately after turning on the power of the interactive communication apparatus 1, the DSP 25 performs the following noise measurement described with reference to FIGS.
The initial processing of the DSP 25 immediately after turning on the power of the two-way communication device 1 is to measure the floor noise and the reference signal level, and based on the difference between them, a guideline of the effective distance between the speaker and the present system and the speech start / end determination threshold level. To set up.
The level value peak-held by the sound pressure level detection unit in the DSP 25 is read at a constant time interval, for example, 10 mSec, and an average value of unit time values is calculated and used as floor noise. Then, the DSP 25 determines a threshold value for a speech start detection level and a speech end detection level based on the measured floor noise level.

FIG. 10, Process 1: Test Level Measurement The DSP 25 outputs a test tone to the line-in terminal of the reception signal system illustrated in FIG. 5 according to the process illustrated in FIG. The sound is collected at ~ MC6, and the average value is obtained using the signal as a speech start reference level.

FIG. 11, Process 2: Noise measurement 1
In accordance with the process illustrated in FIG. 11, the DSP 25 collects the level of the collected sound signal from each of the microphones MC1 to MC6 as a floor noise level for a predetermined time and obtains an average value.

FIG. 12, Process 3: Effective distance trial DSP 25 compares the speech start reference level with the floor noise level according to the process illustrated in FIG. And the effective distance between the speaker who works well in the two-way communication device 1 and the two-way communication device 1 is calculated.

As a result of the microphone selection prohibition determination process 3, if the floor noise is larger (higher) than the speech start reference level, the DSP 25 determines that there is a strong noise source in the direction of the microphone, and automatically selects a microphone in that direction. The prohibition is set, and this is displayed, for example, on the light emitting diodes LED1 to 6 or the operation unit 15.

As illustrated in FIG. 13, the threshold value determination DSP 25 compares the speech start reference level and the floor noise level, and determines the threshold values of the speech start and end levels from the difference.

  As far as noise measurement is concerned, the next process is a normal process, so the DSP 25 sets each timer (counter) and prepares for the next process.

The noise normal processing DSP 25 performs noise processing according to the processing shown in FIG. 14 in the normal operation state after the noise measurement during the initial operation of the two-way communication device 1, and each of the six microphones MC1 to MC6. The volume level average value of the selected speaker and the noise level after detection of the end of the speech are measured, and the speech start / end determination threshold level is reset in a fixed time unit.

FIG. 14, Process 1 : The DSP 25 determines branching to Process 2 or Process 3 based on the determination of whether the speech is in progress or the end of speech.

FIG. 14, Process 2 : Speaker Level Measurement The DSP 25 averages and records the level data for a unit time, for example, 10 seconds, for a plurality of times, for example, 10 times, as a speaker level.
If the utterance ends within the unit time, the time measurement and the utterance level measurement are stopped until a new utterance starts, and the measurement process is resumed after the new utterance is detected.

FIG. 14, Process 3 : Floor noise measurement 2
The DSP 25 averages the noise level data for a unit time, for example, 10 seconds from the detection of the end of the speech to the start of the speech, and records the average as a floor noise level a plurality of times, for example, 10 times.
If there is a new message within the unit time, the DSP 25 stops the time measurement and noise measurement on the way, and restarts the measurement process after detecting the end of the new message.

FIG. 14, Process 4 : Threshold Determination 2
The DSP 25 compares the speech level and the floor noise level, and determines the threshold values for the speech start and end levels from the difference.
In addition to this, since the average value of the speaking level of the speaker is obtained, the speaking start and end detection threshold levels specific to the speaking party facing the microphone can be set.

Generation of Various Frequency Component Signals by Filter Processing FIG. 15 is a configuration diagram showing filtering processing performed by the DSP 25 using sound signals collected by a microphone as preprocessing. FIG. 15 shows processing for one microphone (channel (one sound collection signal)).
The collected sound signal of each microphone is processed by an analog low cut filter 101 having a cutoff frequency of 100 Hz, for example, and a filtered audio signal from which a frequency of 100 Hz or less has been removed is output to the A / D converter 102. , Digital high-cut filters 103a to 103e (collectively referred to as “collection signals”) having cut-off frequencies of 7.5 KHz, 4 KHz, 1.5 KHz, 600 Hz, and 250 Hz, respectively. 103), high frequency components are removed (high cut processing). The results of the digital high cut filters 103a to 103e are further subtracted for each filter signal of the adjacent digital high cut filters 103a to 103e in subtractors 104a to 104d (collectively 104).
In the embodiment of the present invention, the digital high cut filters 103a to 103e and the subtractors 104a to 104d are actually processed in the DSP 25. The A / D converter 102 can be realized as one of the A / D converter blocks 27.

  FIG. 16 is a frequency characteristic diagram showing the filter processing result described with reference to FIG. Thus, a plurality of signals having various frequency components are generated from a signal collected by a microphone having one directivity.

The start and end of speech is determined as one of the triggers for starting the band-pass filter processing and microphone signal level conversion processing microphone selection processing. A signal used for this purpose is obtained by the bandpass filter processing and level conversion processing illustrated in FIG. FIG. 17 shows only 1CH during input signal processing of 6 channels (CH) collected by the microphones MC1 to MC6.
The band-pass filter processing and level conversion processing unit in the DSP 25 respectively collects the collected sound signals of the microphones of each channel at 100 to 600 Hz, 200 to 250 Hz, 250 to 600 Hz, 600 to 1500 Hz, 1500 to 4000 Hz, 4000 to 7500 Hz. Band-pass filters 201a to 201a having band-pass characteristics (collectively, band-pass filter block 201), original microphone sound collection signals, and level converters 202a to 202g (for level conversion of the band-pass sound collection signals) Collectively, it has a level conversion block 202).

  Each level conversion unit 202 a to 202 g includes a signal absolute value processing unit 203 and a peak hold processing unit 204. Therefore, as illustrated in the waveform diagram, the signal absolute value processing unit 203 inverts the sign and converts it to a positive signal when a negative signal indicated by a broken line is input. The peak hold processing unit 204 holds the maximum value of the output signal of the signal absolute value processing unit 203. However, in the present embodiment, the held maximum value is somewhat lowered with the passage of time. Of course, the peak hold processing unit 204 can be improved so that the maximum value can be held for a long time by reducing the decrease.

A bandpass filter will be described. The band-pass filter used for the two-way communication device 1 is composed of, for example, a secondary IIR high-cut filter and a microphone signal input stage low-cut filter only.
In the present embodiment, it is utilized that if the signal that has passed through the high-cut filter is subtracted from the signal having a flat frequency characteristic, the rest is substantially equivalent to the signal that has passed through the low-cut filter.
In order to match the frequency-level characteristics, an extra band-pass bandpass filter is required for one band, but the band required by the number of filter stages equal to the number of bands of the required bandpass filter + 1 and the filter coefficient A pass is obtained. The band frequency of the hand pass filter required this time is the following 6 band pass filter per channel (CH) of the microphone signal.

In this method, the calculation program of the above IIR filter in the DSP 25 is only 6CH (channel) × 5 (IIR filter) = 30.
Contrast with the conventional band-pass filter configuration. Assuming that the band-pass filter uses a second-order IIR filter and a 6-band band-pass filter is prepared for each of six microphone signals as in the present invention, in the conventional method, 6 × 6 × 2 = 72. Circuit IIR / filtering is required. This processing requires considerable program processing even with the latest excellent DSP, and affects other processing.
In the embodiment of the present invention, the 100 Hz low cut filter is processed by an analog filter in the input stage. There are five types of cutoff frequencies of the prepared second-order IIR high cut filters: 250 Hz, 600 Hz, 1.5 KHz, 4 KHz, and 7.5 KHz. Of these, the high-cut filter with a cutoff frequency of 7.5 kHz is not necessary because the sampling frequency is actually 16 KHz. Deliberately rotate the phase of the attenuator to reduce the phenomenon.

  FIG. 18 is a flowchart when processing by the DSP 25 is performed according to the configuration illustrated in FIG.

  In the DSP 25 illustrated in FIG. 18, a high-pass filter process is performed as the first stage process, and a subtraction process from the result of the first-stage high-pass filter process is performed as the second stage process. FIG. 16 is an image frequency characteristic diagram of the signal processing result. [X] below shows each processing case in FIG.

First stage [1] The input signal is passed through a 7.5 kHz high cut filter for the whole band pass filter. This filter output signal becomes a bandpass filter output of [100Hz-7.5KHz] by matching the analog low cut of the input.

  [2] Pass the input signal through a 4KHz high cut filter. This filter output signal becomes a bandpass filter output of [100Hz-4KHz] by combining with the input analog low cut filter.

  [3] Pass the input signal through a 1.5 kHz high cut filter. This filter output signal is combined with the input analog low cut filter [100Hz-1.5KHz] is combined with the input analog low cut filter [100Hz-1.5KHz] When combined with the input analog low cut filter [100Hz -1.5KHz] bandpass filter output.

  [4] Pass the input signal through a 600 kHz high cut filter. This filter output signal becomes a bandpass filter output of [100Hz-600Hz] by combining with the input analog low cut filter.

  [5] The input signal is passed through a 250 kHz high cut filter. This filter output signal becomes a bandpass filter output of [100Hz-250Hz] by combining with the input analog low cut filter.

The second stage [1] band pass filter (BPF5 = [4KHz ~ 7.5KHz]) executes the process of filter output [1]-[2] ([100Hz ~ 7.5KHz]-[100Hz ~ 4KHz]) The signal output is [4KHz to 7.5KHz].
[2] The bandpass filter (BPF4 = [1.5KHz to 4KHz]) will perform the above processing when the filter output [2]-[3] ([100Hz to 4KHz]-[100Hz to 1.5KHz]) is executed. Output [1.5KHz ~ 4KHz].
[3] The bandpass filter (BPF3 = [600Hz to 1.5KHz]) performs the above processing when the filter output [3]-[4] ([100Hz to 1.5KHz]-[100Hz to 600Hz]) is executed. Output [600Hz ~ 1.5KHz].
[4] The bandpass filter (BPF2 = [250Hz to 600Hz]) performs the process of filter output [4]-[5] ([100Hz to 600Hz]-[100Hz to 250Hz]). ~ 600Hz]. [5] The bandpass filter (BPF1 = [100 Hz to 250 Hz]) uses the signal [5] as it is as the output signal [5].
[6] The bandpass filter (BPF6 = [100 Hz to 600 Hz]) uses the signal [4] as it is and outputs it as the output signal (4).
The bandpass filter output required by the above processing in the DSP 25 is obtained.

  The input microphone sound collection signals MIC1 to MIC6 are constantly updated in the DSP 25 as the sound pressure level of the entire band and the sound pressure level of the six bands that have passed through the bandpass filter as shown in Table 5.

In Table 5, for example, L1-1 indicates a peak level when the collected sound signal of the microphone MC1 passes through the first bandpass filter 201a.
The start and end of speech is determined using a microphone sound collection signal that has passed through the 100 Hz to 600 Hz bandpass filter 201a shown in FIG. 17 and whose sound pressure level has been converted by the level converter 202b.

  The conventional band-pass filter is configured by combining a high-pass filter and a low-pass filter for each stage of the band-pass filter. Therefore, a 36-band band-pass filter of the specification used in this embodiment is used. When constructed, 72 circuits of filter processing are required. In contrast, the filter configuration of the embodiment of the present invention is simplified as described above.

Sentence start / end determination processing The first digital signal processor (DSP1) 25, based on the value output from the sound pressure level detector, as shown in FIG. If it rises and exceeds the threshold of the speech start level, it is determined that the speech starts.If the level continues to be higher than the threshold of the start level, the floor noise is determined if the level is lower than the threshold of speech end during speech. The speech end determination time is determined, for example, when it is continued for 0.5 seconds, the speech end is determined.
The start and end of speech is determined based on sound pressure level data (microphone signal level (1)) that has passed through a 100 Hz to 600 Hz bandpass filter whose sound pressure level has been converted by the microphone signal conversion processing unit 202b illustrated in FIG. It is determined that the utterance has started when the threshold level illustrated in FIG. 19 is reached.
In order to avoid malfunction due to frequent microphone switching, the DSP 25 does not detect the next speech start after the speech start determination time, for example, 0.5 seconds.

The microphone selection DSP 25 performs speaker direction detection and automatic selection of a microphone signal facing the speaker in the mutual communication system based on a so-called “star chart method”.
FIG. 20 is a graph illustrating the operation mode of the interactive communication device 1.
FIG. 21 is a flowchart showing normal processing of the interactive communication device 1.

As illustrated in FIG. 20, the two-way communication device 1 performs voice signal monitoring processing according to the collected sound signals from the microphones MC1 to MC6, performs speech start / end determination, performs speech direction determination, and selects a microphone. The result is displayed on the light emitting diodes LED1 to LED6.
The operation will be described below with the DSP 25 in the two-way communication device 1 as a main component with reference to the flowchart of FIG. The overall control of the microphone / electronic circuit housing unit 2 is performed by the microprocessor 23, and the processing of the DSP 25 will be mainly described.

Step 1: Level Conversion Signal Monitoring Signals collected by the microphones MC1 to MC6 are respectively obtained in the band-pass filter block 201 and the level conversion block 202 described with reference to FIGS. Therefore, the DSP 25 constantly monitors seven types of signals for each microphone sound collection signal.
Based on the monitoring result, the DSP 25 proceeds to any one of the speaker direction detection processing 1, the speaker direction detection processing 2, and the speech start / end determination processing.

Step 2: Speech Start / End Determination Processing The DSP 25 determines the start and end of speech according to the method described in detail below with reference to FIG. When the DSP 25 detects the start of speech, the DSP 25 informs the speaker direction determination processing in step 4 of the start of speech.
When the speech start / end determination process in step 2 is performed, when the speech level becomes lower than the speech end level, a speech end determination time (for example, 0.5 second) timer is activated and the speech end determination time and the speech level are When the level is lower than the end level, it is determined that the speech has ended.
If it becomes larger than the speech end level within the speech end determination time, it waits until it becomes smaller than the speech end level again.

Step 3: Speaker Direction Detection Processing The speaker direction detection processing in the DSP 25 is continuously performed by continuously searching for the speaker direction. Thereafter, the data is supplied to the speaker direction determination processing in step 4.

Step 4: Speaker direction microphone switching processing The timing determination processing in the speaker direction microphone switching processing in the DSP 25 has been selected from the results of the processing in step 2 and step 3 and the speaker detection direction at that time. If the speaker direction is different, the microphone selection in step 4 is instructed to select a microphone in a new speaker direction.
However, if the chairman's microphone is set from the operation unit 15 and the chairman's microphone and another conference participant speak at the same time, the chairman's comment is given priority.
At this time, the selected microphone information is displayed on the light emitting diodes LED1 to LED6.

Step 5: Transmission of microphone sound collecting signal In the microphone signal switching process, only the microphone signal selected by the process of Step 4 from the six microphone signals is used as the transmission signal, and the two-way communication device 1 through the telephone line 920. For transmission to the other party's two-way communication device, the data is output to the line-out of the telephone line 920 illustrated in FIG.

Processing for setting a speech start level threshold and a speech end threshold 1: Immediately after turning on the power, the floor noise for a predetermined time, for example, 1 second, of each microphone is measured.
The DSP 25 reads out the peak-held level value of the sound pressure level detection unit at a constant time interval, for example, 10 mSec interval in this embodiment, and calculates an average value of values for a predetermined time, for example, 1 minute to obtain floor noise. And
The DSP 25 determines a speech start detection level (floor noise +9 dB) and a speech end detection level threshold (floor noise +6 dB) based on the measured floor noise level. The DSP 25 thereafter reads the peak-held level value of the sound pressure level detector at regular time intervals.
When it is determined that the speech has ended, the DSP 25 functions as a floor noise measurement, detects the start of speech, and updates the threshold for the detection level of speech end.

  According to this method, since the floor noise level at the position where the microphone is placed is different in this threshold setting, a threshold can be set for each microphone, and erroneous determination in selection of the microphone by the noise source can be prevented.

Process 2: Dealing with ambient noise (large floor noise) rooms Process 2 performs the following as a countermeasure when it is difficult to detect the start and end of speech when the floor level is high in Process 1 and the threshold level is automatically updated. .
The DSP 25 determines a threshold for the detection level of the speech start and the detection level of the speech end based on the predicted floor noise level.
The DSP 25 sets the speech start threshold level to be greater than the speech end threshold level (for example, a difference of 3 dB or more).
The DSP 25 reads the level value peak-held by the sound pressure level detector at regular time intervals.

  According to this method, since the threshold value is the same value for all microphones, the person who is behind the noise source and the person who is not so have the same voice volume and can recognize the start of speech.

Talk start judgment
Process 1 : The output level of the sound pressure level detector corresponding to the six microphones is compared with the threshold value of the speech start level.
When the output level of the sound pressure level detector corresponding to all the microphones exceeds the threshold of the speech start level, the DSP 25 determines that the signal is from the reception / reproduction speaker 16 and does not determine that the speech is started. This is because the distance between the reception / reproduction speaker 16 and all the microphones MC1 to MC6 is the same, so that the sound from the reception / reproduction speaker 16 reaches almost all the microphones MC1 to MC6.

Process 2 : Two unidirectional microphones (with microphones MC1 and MC1) with the directional axes shifted by 180 degrees in the opposite direction at an equal angle of 60 degrees with respect to the six microphones illustrated in FIG. Three sets of MC4, microphones MC2 and MC5, and microphones MC3 and MC6) are used, and the level difference of the microphone signal is used. That is, the following calculation is performed.

The DSP 25 compares the absolute values [1], [2], and [3] with the threshold value of the speech start level, and determines that the speech is started when the threshold value of the speech start level is exceeded.
In the case of this process, since all the absolute values do not become larger than the threshold value of the speech start level as in process 1 (because the sound from the reception / reproduction speaker 16 reaches all the microphones equally), the reception / reproduction speaker 16 It is not necessary to determine whether the sound is from the speaker or from the speaker.

Speaker Direction Detection Processing For detecting the speaker direction, the characteristics of the unidirectional microphone illustrated in FIG. 6 are used. As illustrated in FIG. 6, the frequency characteristics and level characteristics of the unidirectional microphone change depending on the sound arrival angle from the speaker to the microphone. The results are illustrated in FIGS. 7 (A) to (D). FIGS. 7A to 7D show a fast Fourier transform (FFT) at a predetermined time interval for the sound collected by each microphone with a speaker placed at a predetermined distance from the two-way communication device 1, for example, a distance of 1.5 meters. ) Result. The X axis represents frequency, the Y axis represents signal level, and the Z axis represents time. The horizontal line represents the cut-off frequency of the band-pass filter, and the level of the frequency band sandwiched between the lines is the 5-band band pass from the microphone signal level conversion processing described with reference to FIGS.・ Data converted to sound pressure level through the filter.

A determination method applied as an actual process for detecting the direction of the speaker in the two-way communication device 1 as one embodiment of the present invention will be described.
Appropriate weighting processing is performed on the output level of each band-pass filter (0 for 1 dB full span (1 dBFs) step, 0 for 0 dBFs, 3 for -3 dBFs, or vice versa). This weighting step determines the processing resolution.
The above weighting process is executed for each sample clock, and the weighted score of each microphone is added and averaged with a fixed number of samples, and the microphone signal having a small (large) total score is determined as a microphone facing the speaker. To do. Table 7 below is an image of this result.

In this example illustrated in Table 7, the smallest total point is the first microphone MC1, so the DSP 25 determines that there is a sound source in the direction of the first microphone MC1 (there is a speaker). The DSP 25 holds the result in the form of a sound source direction microphone number.
As described above, the DSP 25 performs weighting on the output level of the bandpass filter in the frequency band for each microphone, and ranks the microphone signals in the order of smaller (or larger) scores of the output of each bandbandpass filter. The microphone signal having the first rank in three or more bands is determined as the microphone facing the speaker. Then, the DSP 25 creates a score table as shown in Table 8 below assuming that there is a sound source in the direction of the first microphone MC1 (there is a speaker).

  Actually, the performance of the first microphone MC1 is not necessarily the best in the output of all bandpass filters due to the reflection of sound and the influence of standing waves depending on the characteristics of the room, but the majority in the 5 bands If it is 1st place, it can be determined that there is a sound source in the direction of the first microphone MC1 (there is a speaker). The DSP 25 holds the result in the form of a sound source direction microphone number.

  The DSP 25 sums the output level data of each band band pass filter of each microphone in the form shown in Table 9 below, determines that the microphone signal having a high level is the microphone facing the speaker, and determines the result as the sound source direction microphone number. Hold in the form of.

Talker direction microphone switching timing determination processing When activated by the speech start determination result of step 2 in FIG. 21 and when a new speaker microphone is detected from the speaker direction detection processing result of step 3 and past selection information, The DSP 25 issues a microphone signal switching command to the microphone signal selection switching process in step 5 and notifies the light emitting diodes LED1 to 6 that the speaker microphone has been switched. Informs that the communication device 1 has responded.

In order to eliminate the influence of reflected sound and standing waves in a room with high reverberation, the DSP 25 prohibits the activation of a new microphone selection command if the speech end determination time (for example, 0.5 seconds) has not elapsed after switching the microphone.
In this embodiment, two microphone selection switching timings are prepared from the result of the microphone signal level conversion process in step 1 in FIG. 21 and the result of the speaker direction detection process in step 3.

First method : When it is possible to clearly determine the start of speech When speech from the direction of the selected microphone has ended and there is a new speech from another direction.
In this case, the DSP 25 starts speaking after all the microphone signal level (1) and the microphone signal level (2) are equal to or lower than the speech end threshold level and more than the speech end determination time (for example, 0.5 seconds). When any microphone signal level (1) is equal to or higher than the speech start threshold level, it is determined that speech has started, and a microphone facing the speaker direction is properly collected based on the information of the microphone number in the sound source direction. The microphone is determined, and the microphone signal selection switching process in step 5 is started.

Second method : When there is a new louder voice from another direction while the voice is continuing In this case, the DSP 25 starts the voice from the start of the voice (when the microphone signal level (1) exceeds the threshold level) and the voice ends. The determination process starts after the determination time (for example, 0.5 seconds) has elapsed. If it is determined that the sound source direction microphone number from the process 3 is changed and is stable before the end of the speech is detected, the DSP 25 is more than the speaker currently selected for the microphone corresponding to the sound source direction microphone number. It is determined that there is a speaker who is speaking loudly, the sound source direction microphone is determined as a valid sound collecting microphone, and the microphone signal selection switching process in step 5 is started.

The microphone signal selection switching processing DSP 25 facing the detected speaker is activated by the command selected and determined by the command from the speaker direction microphone switching timing determination processing in step 4 of FIG.
The microphone signal selection switching process of the DSP 25 is composed of a 6-circuit multiplier and a 6-input adder as illustrated in FIG. In order to select the microphone signal, the DSP 25 sets the channel gain (channel gain: CH Gain) of the multiplier to which the microphone signal to be selected is connected to [1], and the CH gains of the other multipliers to [0]. By doing so, the selected signal of (microphone signal × [1]) and the processing result of (microphone signal × [0]) are added to the adder, and a desired microphone selection signal is obtained at the output.

  When the channel gain is switched between [1] and [0] as described above, there is a possibility that a click sound is generated due to the level difference of the microphone signal to be switched. Therefore, in the two-way communication device 1, as illustrated in FIG. 23, in order to change the change in CH Gain from [1] to [0] and from [0] to [1], for example, By continuously changing and crossing in a time of 10 milliseconds, the generation of a click sound due to the difference in the level of the microphone signal is avoided.

  Further, by setting the maximum channel gain to other than [1], for example, [0.5], the echo cancellation processing operation in the DSP 25 at the subsequent stage can be adjusted.

  As described above, the call device according to the embodiment of the present invention is not affected by noise, and can be effectively applied to a call device such as a conference.

The communication device according to the embodiment of the present invention has the following advantages in terms of structure.
(1) The positional relationship between a plurality of microphones having a single directivity and the reception / reproduction speaker is constant, and furthermore, since the distance is very close, the sound emitted from the reception / reproduction speaker passes through the conference room (room) environment. The level that returns directly to the multiple microphones is overwhelmingly dominant. Therefore, the characteristics (signal level (intensity), frequency characteristics (f characteristic), phase) for sound to reach a plurality of microphones from the receiving / reproducing speaker are always the same. That is, there is an advantage that the transfer function is always the same in the communication device.

  (2) Therefore, there is no change in the transfer function when the microphone is switched, and there is an advantage that it is not necessary to adjust the gain of the microphone system every time the microphone is switched. In other words, there is an advantage that it is not necessary to start over once the adjustment is made at the time of manufacturing the communication device.

  (3) Even if the microphone is switched for the same reason as described above, only one echo canceller configured by a digital signal processor (DSP) may be used. The DSP is expensive, and the space for placing the DSP on a printed circuit board on which various members are mounted and there is little space may be small.

  (4) Since the transfer function between the receiving / reproducing speaker and the plurality of microphones is constant, there is an advantage that the sensitivity difference of the microphone itself having ± 3 dB can be adjusted by the unit alone. (5) The table on which the communication device is mounted normally uses a round table, but a speaker system that evenly distributes sound of equal quality in all directions with one reception / reproduction speaker in the communication device can be realized. became.

  (6) The sound emitted from the receiving / reproducing speaker is transmitted to the table surface (boundary effect), and the sound is effectively and evenly delivered to the conference participants, and the sound on the opposite side to the ceiling of the conference room. The phase is canceled to produce a small sound, and there is an advantage that the conference participant has less reflected sound from the ceiling direction, and as a result, a clear sound is distributed to the participant.

  (7) Since the sound emitted from the reception / reproduction speaker reaches all of the plurality of microphones at the same volume at the same time, it is easy to determine whether the sound is the speaker's voice or the reception voice. As a result, erroneous determination of microphone selection processing is reduced.

  (8) By arranging even number of microphones at equal intervals, level comparison for direction detection can be easily performed.

  (9) Due to a damper using a cushioning material, a microphone support member having flexibility or elasticity, vibration due to the sound of the reception and reproduction speaker that can be transmitted through the printed circuit board on which the microphone is mounted is collected by the microphone. Can reduce the influence.

  (10) The sound of the receiving / reproducing speaker does not directly enter the microphone. Therefore, in this two-way communication device, the influence of noise from the reception / reproduction speaker is small.

The above communication device has the following advantages from the signal processing aspect.
(A) A plurality of unidirectional microphones are arranged radially at equal intervals so that the direction of the sound source can be detected, and the microphone signal is switched to collect (collect) sound with good S / N and clear sound. Can be sent to the other party.
(B) Sound from surrounding speakers can be collected with good S / N and a microphone facing the speaker can be automatically selected.
(C) In the present invention, signal analysis is simplified by dividing a passing voice frequency band as a method of microphone selection processing and comparing levels of the divided frequency bands.
(D) The microphone signal switching processing of the present invention is realized as DSP signal processing, and a plurality of signals are all cross-fade processed so as not to generate a clicking sound at the time of switching.
(E) The microphone selection result can be notified to display means such as a light emitting diode or to the outside. Therefore, for example, it can be used as speaker position information for a television camera.

The configuration and processing operation of the interactive communication device 1 have been described in detail above.
A technique using the above-described effects and features of the interactive communication device 1 is a communication device with a translation function as a speech processing device of the present invention described in the following embodiments.

First Embodiment FIG. 24 is an example of a circuit block diagram of the translation function-equipped calling device 1a in the present embodiment.
The circuit block diagram of the communication device with translation function 1a shown in FIG. 24 is compared with the circuit block diagram of the two-way communication device 1 shown in FIG. 5 with a translation DSP 40 that is a third digital signal processor (DSP). The speech recognition memory 41, the translation dictionary memory 42, and the phoneme piece memory 43 are different.
These added components may be configured outside the two-way communication device 1 described above, but in the present embodiment, they are mounted in an empty area on the circuit board shown in FIG.

As shown in FIG. 25, the translation DSP 40 includes a speech recognition unit 401 as a speech recognition unit, a translation unit 402 as a text translation unit, and a speech synthesis unit 403 as a speech synthesis unit. 401 is connected to the speech recognition memory 41, the translation unit 402 is connected to the translation dictionary memory 42, and the speech synthesis unit 403 is connected to the phoneme segment memory 43.
Hereinafter, each of these components will be described.

The voice recognition unit 401 converts the signal S26, which is a voice signal output from the DSP 26, into character string data (text data).
Specifically, when the signal S26 is input, the input voice signal is analyzed, and an acoustic feature quantity (acoustic feature) is extracted from the acoustic model stored in the voice recognition memory 41, as will be described later. That is, for the input speech signal S26, acoustic features are extracted for each basic sound unit used for speech recognition, that is, for each unit of small human pronunciation (phonemes) such as consonants and vowels.
Furthermore, the speech recognition unit 401 refers to the acoustic features for each phoneme of the extracted speech signal S26 with the recognition dictionary stored in the speech recognition memory 41, and the speech signal S26 input in the text data to be recognized. The candidate closest to each phoneme is output as a speech recognition result (signal S401, which is text data). That is, since the recognition dictionary describes text data corresponding to the acoustic features in units of phonemes, the acoustic features extracted from the signal S26 are compared with the acoustic features described in the recognition dictionary. Selects and outputs text data corresponding to the close acoustic feature.

  At that time, in order to improve the recognition rate in the above-described voice recognition, the registrant's voice can be recognized particularly well by previously registering the word to be recognized with the voice of a specific person. It is also possible (specific speaker voice recognition). Therefore, when the speaker (conference participant) who carries out the conference is specified in advance using the telephone device with translation function 1a, the acoustic features of these conference participants are registered in the speech recognition memory 41. Therefore, the recognition rate of voice recognition can be improved.

The voice recognition memory 41 stores the above-described acoustic model and recognition dictionary.
In the acoustic model, a small unit (phoneme) of human pronunciation is described by an acoustic feature, and it is possible to refer to an acoustic feature corresponding to the phoneme unit of an audio signal. This acoustic feature is statistical acoustic feature information of phonemes obtained from the voices of many speakers.
Text data for speech recognition is described in the recognition dictionary, and text data corresponding to acoustic features in phonemes can be referred to.

The translation unit 402 receives the text data S401 from the speech recognition unit 401, performs translation processing, and then outputs the text data S402.
As the translation processing, various translation processing software is known as a publicly known technology, and it is possible to apply them. For example, when the text data of the language X is translated into the text data of the language Y, the following processing is generally required.
(1) Syntax analysis of text data in language X (signal S401) Text data in language X (signal S401) is input and syntax analysis is performed. That is, the input text data of the language X is analyzed, the subject, predicate, etc. of each word are specified, and data (syntax tree) in which the relationship between words is described in a tree structure is created.
(2) Generation of syntax tree of language Y The obtained syntax tree of language X is converted according to a predetermined rule. That is, referring to the translation dictionary (language X → language Y) stored in the translation dictionary memory 42, the syntax tree obtained in (1) is converted into a language Y syntax tree in units of words. Then, the obtained syntax tree of the language Y is converted according to a predetermined rule such as conversion of the word order adapted to the language Y, addition of a particle, or the like.
(3) Generation of Language Y Text Data Language Y text data (signal S402) is generated from the converted language Y syntax tree.

As described above, in the translation dictionary memory 42, text data in units of words in a specific language and text data in units of words in different languages are described in association with each other as a translation dictionary.
Needless to say, the translation dictionary memory 42 can be additionally registered as appropriate by accessing the translation dictionary memory 42 from outside the call device with translation function 1a.

The speech synthesizer 403 receives the text data S402 translated by the translator 402 and converts it into a speech signal (speech synthesis).
Specifically, the following processing is performed.
(1) Text analysis processing The input text data S402 is analyzed with reference to a word dictionary stored in the phoneme memory 43.
Text analysis uses the above word dictionary to add information such as accent setting and pose to the input text data based on the grammatical connection relationship of the connected words. This is because the position where the accent is set may differ depending on the context before and after the same word.
(2) Speech synthesis processing In the speech synthesis processing, text data to which accents and the like generated by the text analysis processing are added is converted into speech signals. That is, the text data subjected to the text analysis process is converted into a corresponding voice signal with reference to the phoneme piece memory 43 in which the voice signal for each phoneme is stored (voice signal S40).

The phoneme memory 43 stores the word dictionary and phoneme data as described above.
The word dictionary is information such as reading for each word (text data), accent setting / pause, and the like, and is referred to in text analysis processing executed by the speech synthesizer 403.
The phoneme piece data is data of a voice signal corresponding to each phoneme, and is referred to in a voice synthesis process executed by the voice synthesis unit 403.

  Needless to say, when the translation unit 402 described above converts Japanese text into English text, the English word dictionary and phoneme piece data are stored in the phoneme piece memory 43.

Heretofore, each component of the translation device with translation function 1a added to the bidirectional communication device 1 has been described.
Next, the operation of the translation device with translation function 1a in the present embodiment will be described with reference to FIG. In the following description of the operation, a case where the translation DSP 40 has a translation function from Japanese to English will be described as an example.

Similarly to the signal processing operation of the bidirectional communication device 1 described above, in the communication device with translation function 1a shown in FIG. 24, when analog audio signals in Japanese collected from the six microphones MC1 to MC6 are input, Each of them is converted into a digital signal by A / D converters 271 to 273 of two channels.
The converted digital signal is subjected to the above-described signal processing by the DSP 25, that is, microphone selection / switching processing. As a result, the main speaker is identified and the voice signal is output to the DSP 26.

  In the DSP 26, an echo cancellation process is performed on the voice signal of the identified speaker, and is output to the translation DSP 40.

In the translation DSP 40, the Japanese speech signal subjected to echo cancellation processing in the speech recognition unit 401 is converted into a character string (text data), and the text data is converted in the translation unit 402 into English text data.
Further, the speech synthesis unit 403 generates an English speech signal based on the English text data converted by the translation unit 402 (signal S40).

  Therefore, the speaker's Japanese speech signal (digital) selected by the DSP 25 is converted into an English speech signal by the translation DSP 40, converted to an analog signal by the D / A converter 28, and amplified by the amplifier 291. And output from the line-out.

Next, an application example of the call device with translation function 1a will be described.
In FIG. 26, for example, a plurality of Japanese A10, A11 and a plurality of Americans A20, A21, each at a remote location, hold a conference via the telephone line 920 using the telephone devices with translation functions 1a, 1a ′, respectively. It is a figure which shows the example of application in the case of performing.
The call device with translation function 1a has a translation function from Japanese to English, and the call device with translation function 1a 'has a translation function from English to Japanese. Therefore, the translation DSP 40, the speech recognition memory 41, the translation dictionary memory 42, and the phoneme memory 43 are different between the translation device with translation function 1a and the translation device with translation function 1a ′.

As described above, a plurality of Japanese A10, A11 speaks (speech signals) in Japanese, one speaker is identified by the call function device with translation function 1a, and the voice of the identified speaker is identified. The signal is converted to an English analog audio signal. The English analog audio signal is input to the line-in of the translation function-equipped telephone 1a ′ installed in the other party's conference room through the telephone line 920, and an echo-cancelled sound is output from the reception reproduction speaker 16. Thereby, a plurality of Americans A20 and A21 can understand the content of the remarks in the Japanese meeting.
In addition, since Japanese speeches (voice signals) of a plurality of Japanese A10 and A11 are output in Japanese from the reception playback speaker 16 of the translation device with translation function 1a, other Japanese participating in the conference. People can understand what they say.

On the other hand, for a plurality of Americans A20 and A21 speaking in English (speech signals), one speaker is identified by the translation device 1a 'with translation function, and the identified speaker's speech signal is Japanese analog. It is converted into an audio signal. The Japanese analog voice signal is input to the line-in of the telephone 1a with a translation function installed in the other party's conference room through the telephone line 920, and an echo-cancelled sound is output from the reception reproduction speaker 16. Thereby, several Japanese A10 and A11 can understand the content of the remark in the American meeting.
In addition, the speech (speech signal) in English of a plurality of Americans A20 and A21 is output in English from the reception / reproduction speaker 16 of the telephone device with translation function 1a ', so that other Americans participating in the conference Can understand the remarks.

As described above, even when a plurality of persons in different languages cannot speak each other's language, a telephone conference or the like can be performed by using the telephone devices with translation functions 1a and 1a ′. .
In addition, since the speech recognition unit 401 and the translation unit 402 of the translation DSP 40 can obtain text data before and after translation, by outputting this, it can be used as minutes.
Furthermore, in the present embodiment, the call device with translation function 1a has the following effects in order to inherit the features of the interactive call device 1.
(1) Since the speaker system that is distributed evenly in all directions of 360 degrees is provided, the translated audio signal is propagated to all the conference participants with equal quality.
(2) The translated voice output from the receiving / reproducing speaker is transmitted to the conference participants as a high-quality sound through the table surface on which the communication device with translation function 1a is installed. For the ceiling direction, the phase is canceled with the sound on the opposite side, resulting in a small sound, and there is less reflected sound from the ceiling direction for the conference participants, resulting in a clearly translated audio signal being distributed to the participants. The

The present embodiment is not limited to the above-described contents, and various changes can be made.
For example, in the above description, the translation DSP 40 performs a translation process between a one-to-one language, but may be configured to be able to translate between multiple languages.
In addition, when there are many conferences in Japanese between Japanese, a switching function is added to select whether or not to execute the process in the translation DSP 40 of the telephone device with translation function 1a. You may make it make a user select by the dip switch etc. which were installed in the communication apparatus 1a with a function.

Further, in the above-described communication device with translation function 1a, the translation DSP 40 is set to the output side by line-out, but the side to be output by the reception speaker 16, that is, between the DSP 26 and the D / A converter 282. It may be set.
Then, if the translation DSP 40 is configured such that bidirectional translation functions of different languages, for example, translation processing from Japanese to English and English to Japanese can be switched to a switch or the like, one telephone with translation function , It will be possible for Japanese and Americans to talk in their native language in the same conference room.
Even in this case, since the translated speech signals are output from the receiving speaker 16, no matter where the speaker (Japanese and American) is located, with the translation function-equipped communication device as the center. There is an advantage that it is recognized as a clear sound from the feature of the communication device.

Second Embodiment Hereinafter, a second embodiment will be described.
Compared with the telephone conversation apparatus with a translation function 1a described in the first embodiment, the telephone conversation apparatus with a translation function according to the present embodiment translates and outputs only a voice signal that has been voiceprint-authenticated.
27 to 28 are an example of a block diagram showing the configuration of the translation function-equipped calling device 1b in the present embodiment.
As shown in FIGS. 27 to 28, the call device with translation function 1b according to the present embodiment includes a voiceprint data memory 44 and a translation DSP 40 as compared with the call function device with translation function 1a according to the first embodiment. A voiceprint recognition unit 404 as voiceprint recognition means is added and configured.
The voiceprint recognition unit 404 and the voiceprint data memory 44 are connected to each other.

Hereinafter, these components added to the telephone call device with a translation function 1a in the first embodiment will be described.
The voiceprint recognition unit 404 performs voiceprint authentication of the input voice signal, and outputs only the voice signal that has been voiceprint authentication to the subsequent voice recognition unit 401.
That is, the difference in individual voiceprints is determined by the difference in the volume and structure of the mouth and nostrils that arise from the person's face shape, and the difference in the vocal cords that arises from the height and gender. It is possible to identify the person by collating with the registered voiceprint data.
Such voiceprint authentication is said to have a high recognition rate specific to the person because the voiceprint waveform strength and frequency do not appear as a change even if the caller becomes faint or nose due to a cold or the like. .

Specifically, the voice print authentication performs the following processing.
(1) Frequency analysis is performed on the audio signal S26, and voiceprint data expressing the audio signal with a three-dimensional pattern of time, frequency, and sound intensity is generated.
(2) The generated voice print data is compared with the voice print data of the conference participants stored in advance and stored in the voice print data memory 44, and the voice signal is recognized only when there is matching voice print data. The signal is output to the unit 40 as a signal S404.

  As described above, the voiceprint data of the conference participant registered in advance is stored in the voiceprint data memory 44 and is referred to by the voiceprint recognition unit 404.

As described above, according to the telephone conversation device with translation function 1b in the present embodiment, one speaker who is a conference participant is identified, and the voice signal of the identified speaker is voiceprint-authenticated and registered. Since the translation processing is performed after it is confirmed that the speaker is a speaker, and the translated voice signal is output to the other party of the conference, the effect obtained by the call device with translation function 1a in the first embodiment is achieved. In addition, the following effects can be obtained.
That is,
(1) Audio signals other than those of the conference participants are not output to the other party of the conference, but only valid audio signals of only the conference participants are output to the other party of the conference, so that the conference can proceed smoothly. Be expected.
(2) Since personal authentication is possible by voiceprint authentication processing, it is possible to superimpose information on the name of the speaker who has been authenticated as necessary on the audio signal output to the other party of the conference. Become.

Third Embodiment In the first and second embodiments, a translation processing function (translation unit 402) is built in the translation DSP 40. However, in this embodiment, an external translation processing function is used. It is different.
The communication devices 1c and 1d in the present embodiment are based on the communication device with translation function 1a, but depending on whether a digital voice signal or text data is transmitted to an external translation processing function. The configuration of the telephone with translation function in the embodiment is different.

For example, in the case of transmitting a digital audio signal, the audio signal subjected to echo cancellation processing by the DSP 26 is transmitted, and the audio signal subjected to translation processing is received and output. Therefore, a component corresponding to the translation DSP 40 is not necessary. .
In addition, when transmitting text data, the speech signal subjected to echo cancellation processing by the DSP 26 is converted into text data by the speech recognition unit 401 and then transmitted, and the translated text data is converted into a speech signal by the speech synthesis unit 403. The voice recognition unit 401 is required at least because the voice signal that has been converted and output, or the voice signal that has undergone echo cancellation processing by the DSP 26 is transmitted and the voice signal that has undergone translation processing is received and output.

FIG. 29 shows an embodiment in which the communication devices 1c and 1c ′ according to the present embodiment transmit text data or voice signals to the web translation processing server 930 via the LAN devices 950 and 951, respectively. FIG.
As described above, the translation service performed by the translation processing server 930 on the web is being put into practical use by wireless communication. By using such a translation service, the telephone with translation function according to the first embodiment is used. The same effect as 1a can be obtained.

FIG. 30 is a diagram illustrating an embodiment in which the communication devices 1d and 1d ′ according to the present embodiment transmit text data or audio signals to the translation processing center 931 via the mobile phones 940 and 941, respectively. is there.
As described above, the translation service performed in the translation processing center 931 is being put into practical use by wireless communication. By using such a translation service, the translation service with the translation function 1a in the first embodiment is used. An effect can be obtained.

Further, if the above-described communication devices 1c and 1d are configured to output the translated speech signal, which is the received translation result, by the receiving speaker 16, these communication devices can be used as a translator. Become. That is, for example, a Japanese and an American can have a conversation using their native languages in the same conference room using one telephone.
Even in this case, since the translated speech signal (received signal) is output from the receiving speaker 16, even if the speaker (Japanese and American) is located at any position around the communication device, both There is an advantage that it is recognized as a clear sound from the feature of the communication device.

(A) is a figure which shows the outline | summary of the conference system as an example to which a two-way communication apparatus is applied, (B) is a figure which shows the state in which the telephone apparatus in (A) is mounted, (C) is a diagram showing the arrangement of the communication device placed on the table and the conference participants. It is a perspective view of a two-way communication device. FIG. 2 is an internal sectional view of the two-way communication device illustrated in FIG. 1. FIG. 2 is a plan view of a microphone / electronic circuit housing unit from which an upper cover of the two-way communication device illustrated in FIG. 1 is removed. It is a figure which shows the connection state of the main circuit of a microphone and an electronic circuit accommodating part, and has shown the connection state of the connection of a 1st digital signal processor (DSP1) and a 2nd digital signal processor (DSP2). FIG. 5 is a characteristic diagram of the microphone illustrated in FIG. 4. (A)-(D) is a graph which shows the result of having analyzed the directivity of the microphone which has the characteristic illustrated in FIG. It is a partial block diagram of the deformation | transformation aspect of a two-way communication apparatus. It is a graph which shows the outline | summary of the whole processing content in a 1st digital signal processor (DSP1). It is a flowchart which shows the 1st form of the noise measuring method of a two-way communication apparatus. It is a flowchart which shows the 2nd form of the noise measuring method of a two-way communication apparatus. It is a flowchart which shows the 3rd form of the noise measuring method of a two-way communication apparatus. It is a flowchart which shows the 4th form of the noise measuring method of a two-way communication apparatus. It is a flowchart which shows the 5th form of the noise measuring method of a two-way communication apparatus. It is drawing which shows the filtering process in the telephone apparatus of this invention. It is a frequency characteristic figure which shows the process result of FIG. It is a block diagram which shows a band pass filtering process and a level conversion process. It is a flowchart which shows the process of FIG. It is a graph which shows the process which determines the speech start and completion | finish of a two-way communication apparatus. It is a graph which shows the flow of the normal process of a two-way communication apparatus. It is a flowchart which shows the flow of the normal process of a two-way communication apparatus. It is the block diagram which illustrated the microphone switching process of a two-way communication apparatus. It is the block diagram which illustrated the method of the microphone switching process of a two-way communication apparatus. It is an example of the block diagram of the communication apparatus with a translation function which is the 1st Embodiment of this invention. It is an example of the block diagram of the communication apparatus with a translation function which is the 1st Embodiment of this invention. It is a figure which illustrates the example of application of the telephone apparatus with a translation function of 1st Embodiment. It is an example of the block diagram of the telephone apparatus with a translation function which is the 2nd Embodiment of this invention. It is an example of the block diagram of the telephone apparatus with a translation function which is the 2nd Embodiment of this invention. It is a figure which illustrates the example of application of the telephone apparatus with a translation function of 3rd Embodiment. It is a figure which illustrates the example of application of the telephone apparatus with a translation function of 3rd Embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Two-way communication apparatus 1, MC1-MC6 ... Microphone, 16 ... Reception speaker, 23 ... Microprocessor, 24 ... Codec, 25 ... 1st digital signal processor (DSP1), 26 ... 2nd digital signal processor (DSP2) ), 27... A / D converter block, 28... D / A converter block, 29... Amplifier block, 1a, 1a ', 1b .. communication device with translation function, 1c, 1c', 1d, 1d '. 40 ... Translation DSP, 401 ... Speech recognition unit, 402 ... Translation unit, 403 ... Speech synthesis unit, 404 ... Voiceprint recognition unit, 41 ... Speech recognition memory, 42 ... Translation dictionary memory, 43 ... Phoneme segment memory, 44 ... Voiceprint data memory, 920 ... 930 ... Translation processing server, 931 ... Translation processing center, 940, 941 ... Mobile phone, 950, 951 ... LA Apparatus.

Claims (4)

Multiple microphones,
Microphone signal selection means for selecting one of the collected sound signals of the plurality of microphones;
Voice recognition means for generating first text data based on a microphone sound collection signal selected by the microphone signal selection means;
Text translation means for translating the first text data generated by the speech recognition means to generate second text data;
A speech processing apparatus comprising: speech synthesis means for generating a speech signal based on the second text data generated by the text translation means. Voice print recognition means for performing voice print authentication as to whether or not the selected microphone sound collection signal matches a voice print registered in advance;
The voice recognition means
Only when the voiceprint recognition means determines that the voiceprint obtained from the microphone sound collection signal selected by the microphone signal selection means matches the voiceprint registered in advance.
The speech processing apparatus according to claim 1, wherein the first text data is generated based on the microphone sound collection signal. Selecting one of the collected signals of the plurality of microphones;
Generating first text data based on the selected microphone sound collection signal;
Translating the generated first text data to generate second text data;
A voice processing method comprising: generating a voice signal based on the generated second text data. The first text data is generated based on the microphone sound collection signal only when it is determined that the voice print obtained from the selected microphone sound collection signal matches the voice print registered in advance. The voice processing method described.

Images