JP2005055667A - Audio processing device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To precisely conduct an audio recognition process even though a plurality of attendees simultaneously utter in a conference. <P>SOLUTION: A bidirectional talking device 2 receives speech commands through a plurality of directional microphones facing respective uttered person and specifies a main uttered person from audio data. An audio recognition processing device 3 conducts an audio recognition process for the audio data of the specified speaker to generate character string data. The audio recognition process is controlled by a microphone number indicating the main speaker supplied from the bidirectional talking device 2. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a voice processing apparatus that recognizes and processes commands by voices issued by, for example, a plurality of conference attendees.

  2. Description of the Related Art Conventionally, a speech processing apparatus having a function (speech recognition) for recognizing a human speech and converting it into character string data or the like (speech recognition) controls a telephone or a personal computer (hereinafter referred to as a PC) as voice input means. It is assumed that a microphone (hereinafter referred to as a microphone) connected to a device is used, and a voice processing device equipped with such a microphone, for example, an application for inquiring a bank balance by telephone or inputting a command or document to a PC, a car in a vehicle It is applied to operations using voice commands in navigation systems.

However, these conventional speech processing apparatuses are intended for speech recognition for one individual as a usage environment. Therefore, when these voice processing devices are used in a group work by a plurality of people, for example, in a meeting scene of a plurality of people, a plurality of voices are erroneously detected / recognized and cannot be used.
In other words, when two or more participants are speaking at the same time, the two utterances are mixed and input to the voice processing device through the microphone. It was impossible to identify a meeting participant and obtain an accurate speech recognition result.

Further, in view of the environment of simultaneous use by a plurality of persons as described above, it is necessary to consider the operation authority of each person in a speech processing apparatus having a speech recognition function.
In other words, in a group work, even if a conference participant who is mainly speaking can be identified from the voices of multiple people, only the conference participant who has been given the predetermined authority for a specific command in the conference In many cases, it is desirable to take a procedure such as allowing the above command for smooth progress of the conference.

  The problem to be solved is that, for example, when recognizing and processing voices uttered by a plurality of conference attendees, it is not possible to identify a conference attendee who mainly speaks from a plurality of speaker and accurately perform speech recognition processing. It is. In addition, when a conference attendant who mainly speaks is identified from a plurality of speakers and speech recognition is performed, speech recognition processing according to the authority of each conference attendee cannot be performed.

  In order to solve the above-described problems, a first aspect of the present invention is a voice processing apparatus that inputs and processes voice signals from a plurality of microphones, and selects a single microphone based on the voice signals. A selection unit; and a voice conversion unit that converts a voice signal output from the microphone selected by the microphone selection unit into character string data.

  Preferably, the microphone has a unidirectional characteristic.

  Preferably, the voice conversion means includes a buffer that holds the voice signal and a character string data memory that stores character string data corresponding to the voice signal, and is selected by the microphone selection means. The voice signal output from the microphone is taken into the buffer, and at the timing when the selected microphone changes, the voice signal taken into the buffer is checked against the character string data memory and converted to matching character string data. To do.

  According to a second aspect of the present invention, there is provided an audio processing apparatus for inputting and processing audio signals from a plurality of microphones, a microphone selecting means for selecting one microphone based on the audio signals, and the microphone selection A voice conversion unit that converts a voice signal output from the microphone selected by the unit into a command code; and a voiceprint generation that generates voiceprint data based on the voice signal output from the microphone selected by the microphone selection unit. And an instruction code processing means for processing the instruction code based on the voiceprint data.

  Preferably, the microphone has a unidirectional characteristic.

  Preferably, the voiceprint data further includes an authority data memory in which instruction authority data as to whether or not the instruction code is valid is associated and stored, and the instruction code processing means includes the voiceprint data With reference to the voiceprint data generated by the generation means and the authority data memory, if the instruction code converted by the voice conversion means is not valid for the voiceprint data, the instruction code is not processed.

  According to the first aspect of the present invention, the microphone selection means inputs a sound signal from each of the microphones through a plurality of microphones, selects one microphone based on the input sound signal, and selects a sound from the microphone. Output a signal. The voice conversion means converts the voice signal input from the microphone selection means into character string data.

  When the voice conversion means has a buffer that holds the voice signal and a character string data memory that stores character string data corresponding to the voice signal, the voice output from the microphone selected by the microphone selection means The signal is taken into the buffer, and at the timing when the microphone selected by the microphone selection means changes, the voice signal taken into the buffer is checked against the character string data memory and converted to matching character string data.

  According to the second aspect of the present invention, the microphone selection means inputs a voice signal from each of the microphones through a plurality of microphones, selects one microphone based on the input voice signal, and selects a voice from the microphone. Output a signal. The voice conversion unit converts the voice signal input from the microphone selection unit into a corresponding instruction code. The voiceprint generation unit generates voiceprint data based on the voice signal output from the microphone selection unit. The instruction code processing means processes the instruction code converted by the voice conversion means based on the voiceprint data generated by the voiceprint generation means.

  If the voiceprint data further includes an authority data memory in which the instruction authority data about whether the instruction code is valid is stored in association with each other, the instruction code processing means is generated by the voiceprint generator With reference to voiceprint data and the authority data memory, the command code is processed when the command code converted by the voice conversion means is valid for the voiceprint data.

  In the present invention, when a voice command is recognized and processed by a voice processing device, even when a plurality of commands are overlapped and input to the voice processing device, the main speaker is identified and processed. There is an advantage that voice recognition can be performed with high accuracy in the conference.

  In the first to fourth embodiments to be described later, since the two-way communication unit 2 described below is used for all, the configuration and operation of the two-way communication unit 2 are first shown for convenience of description. After detailed description with reference to 1-4, each embodiment will be described.

Two-way communication part 2
FIG. 1 is a circuit block diagram of the two-way communication unit 2.
As shown in FIG. 1, the circuit block diagram of the bidirectional communication unit 2 includes an A / D converter block 21, a DSP (Digital Signal Processor) 22, a DSP 23, a CPU (Central Processing Unit) 24, a codec 25, A D / A converter block 26 (D / A converters 261 and 262), an A / D converter 263, and an amplifier block 27 are provided.

  The two-way communication unit 2 inputs voice from a plurality of microphones, six unidirectional microphones MC1 to MC6 in the example of FIG. A microphone having a single directivity exhibits strong directivity in front of the position where the microphone is placed.

The CPU 24 performs overall control processing of the two-way call unit 2.
The codec 25 encodes voice.
The DSP 22 performs various signal processing, details of which will be described later, such as filter processing and microphone selection processing.
The DSP 23 functions as an echo canceller.
In FIG. 1, A / D converters 211 to 213 are illustrated as an example of the A / D converter block 21, and an A / D converter 263 is illustrated as an example of the A / D converter. D / A converters 261 to 262 are illustrated as an example of the A converter block 26, and amplifiers 271 to 272 are illustrated as an example of the amplifier block 27.

A pair of microphones MC1-MC4: MC2-MC5: MC3-MC6 is input to A / D converters 211-213, which convert 2-channel analog signals into digital signals, respectively.
The collected sound signals of the microphones MC1 to MC6 converted by the A / D converters 211 to 213 are input to the DSP 22, and various signal processing described later is performed.
As one of the processing results of the DSP 22, one of the microphones MC1 to MC6 is selected. The DSP 22 selects a microphone using the unidirectional characteristics of the microphone described above.

The processing result of the DSP 22 is output to the DSP 23 and an echo cancellation process is performed. The processing result of the DSP 23 is converted into an analog signal by the D / A converters 261 to 262. The output from the D / A converter 261 is encoded by the codec 25 as necessary, and is output via the amplifier 271.
The output from the D / A converter 262 is output as a sound from the speaker 28 of the two-way communication unit 2 via the amplifier 272. That is, a conference participant or the like who uses the two-way call unit 2 can hear the voice uttered by the speaker in the conference room via the speaker 28.
The two-way communication unit 2 inputs the other party's voice to the DSP 23 via the A / D converter 263 and performs echo cancellation processing. Further, the other party's voice is applied to the speaker 28 via a path (not shown) and output as sound.

  When a microphone with no directivity is used, all sounds around the microphone are collected, so that the S / N (Signal to Noise) between the voice of the speaker and the ambient noise is not good. In order to avoid this, in the present embodiment, the S / N with surrounding noise is improved by collecting sound with a directional microphone.

Next, processing contents performed by the DSP 22 will be described.
The main processing performed by the DSP 22 is microphone selection / switching processing. That is, when a plurality of conference participants who use the two-way call unit 2 speak at the same time, voices are mixed and difficult to hear for the other party, so only the audio signal from the selected microphone is output as the signal S271 in FIG. Is done.
In order to perform this process accurately, various signal processes exemplified below are performed.
(A) Microphone signal band separation and level conversion processing (b) Speech start / end determination processing (c) Speaker direction microphone detection processing The collected sound signal of each microphone is analyzed and facing the speaker Determine the microphone.
(D) Speaker direction microphone switching timing determination processing, and
Microphone signal selection switching process facing the detected speaker

Hereinafter, each signal processing described above will be described.
(A) Microphone signal band separation and level conversion process The start and end of speech is determined as one trigger for starting the microphone selection process. For this purpose, each microphone signal is subjected to bandpass filter (hereinafter, BPF) processing and level conversion processing.
FIG. 2 is a diagram showing only one channel (CH) among the six microphones MC1 to MC6 for the BPF process and the level conversion process.
The BPF processing and level conversion processing circuit converts the collected sound signal of the microphone into band pass characteristics of 100 to 600 Hz, 100 to 250 Hz, 250 to 600 Hz, 600 to 1500 Hz, 1500 to 4000 Hz, 4000 to 7500 Hz, respectively. ˜221f (collectively referred to as BPF block 221) and level converters 222a through 222g (collectively referred to as level conversion block 222) for converting the levels of the original microphone sound collection signal and the band-pass sound collection signal.
Each level converter includes a signal absolute value processing unit 223 and a peak hold processing unit 224. Therefore, as illustrated in the waveform diagram, the signal absolute value processing unit 223 inverts the sign and converts it into a positive signal when a negative signal indicated by a broken line is input. The peak hold processing unit 224 holds the absolute value of the output signal of the signal absolute value processing unit 223.

(B) Processing for determining start / end of speech The DSP 22 makes the sound pressure level data passing through the BPF of 100 to 600 Hz, which has been subjected to sound pressure level conversion by the microphone signal level conversion processing unit 222b illustrated in FIG. When it becomes, it determines with the speech start, and when it becomes below predetermined value for a fixed time (for example, 0.5 second), it determines with the speech end.

(C) Speaker Direction Microphone Detection Processing For detecting the speaker direction, the characteristics of the unidirectional microphone illustrated in FIG. 3 are used.
As illustrated in FIG. 3, the frequency characteristics and level characteristics of the unidirectional microphone change depending on the arrival angle of sound from the speaker to the microphone. FIG. 3 shows the result of performing FFT on the sound collected by each microphone at a fixed time interval by placing a speaker at a distance of 1.5 meters of the two-way communication unit 2. The X axis represents frequency, the Y axis represents time, and the Z axis represents signal level. A line drawn for each specific frequency on the XY plane represents the cut-off frequency of the BPF processing described with reference to FIG. 2, and the level of the frequency band sandwiched between the lines passes through the BPFs 221b to 221f in FIG. It becomes.

Appropriate weighting processing (for example, 0dBFs is set to 0 and -3dBFs is set to 3 for the 1dBFs step) is performed on the output level of the BPF in each band. This weighting step determines the processing resolution.
For each sample clock, the above weighting process is executed, the weighted scores of each microphone are added, and a microphone signal with a small (or large) total score is averaged over a certain number of samples, and the microphone facing the speaker judge. Table 1 shows an image of this result.
In the example of Table 1, it is determined that there is a sound source in the direction of microphone 1 because the smallest total score is MIC1. The result is held in the form of a sound source direction microphone number.

(D) Speaker direction microphone switching timing determination processing, and
Processing for selecting and switching microphone signals opposite to the detected speaker When a speech from a certain speaker (for example, microphone MC1) ends and a new speech (for example, microphone MC2) is received, (B) As described in the start / end determination process, the microphone (MC1) signal level of the previous speaker falls below a predetermined value, and after a certain time (for example, 0.5 seconds) has passed, It is determined that the speaker's speech has ended.
When the microphone of the later speaker is started and the microphone (MC2) signal level exceeds a predetermined value, the microphone facing the later speaker is determined as the sound collecting microphone and the microphone signal selection switching process is started. To do.

In addition, when the previous speaker (Mic MC1) continues to speak and a new voice is spoken from another direction (later speaker (Mic MC2)), the subsequent speaker starts to speak ( After a certain time (for example, 0.5 seconds) has elapsed since the signal level of the microphone MC2 has reached a predetermined value or more, the microphone switching determination process is started.
The microphone switching determination process is performed as follows.
That is, if there is a speaker (microphone MC2) speaking louder than the currently selected speaker before the end of the previous speaker (microphone MC1), the sound pressure level from the microphone MC2 is (C) In the speaker direction microphone detection process, the scores in MC1 and MC2 in Table 1 are reversed, and the sound source direction microphone number is changed as microphone MC1 → 2, and at the same time, the microphone signal selection switching process Is done.

As illustrated in FIG. 4, the microphone signal selection switching process includes a 6-circuit multiplier and a 6-input adder. In order to select the microphone signal, the channel gain (CH Gain) of the multiplier to which the microphone signal to be selected is connected is set to 1, and the channel gains of the other multipliers are set to 0. The processing result of [selected microphone signal × 1] and [other microphone signal × 0] are added to output a desired microphone selection signal.
Note that the change in channel gain (for example, CH1 Gain and CH2 Gain) before and after the microphone switching is gradually performed, for example, in 10 ms.

  As described above in detail, according to the two-way communication unit 2, the voice from the speaker is collected with good S / N using the characteristics of the directional microphone, and one of the plurality of microphone signals is appropriately selected. Can be selected, and the selected microphone signal and the selected microphone information (the microphone numbers 1 to 6) are supplied to the subsequent apparatus.

First Embodiment Hereinafter, a speech processing apparatus according to a first embodiment will be described.
FIG. 5 is a block diagram of the speech processing apparatus 1 according to the first embodiment.
As shown in FIG. 5, the voice processing device 1 includes the above-described two-way communication unit 2 and the voice recognition processing device 3.
The microphone selection means in the present invention corresponds to the two-way call unit 2 in the first embodiment.
The voice conversion means in the present invention corresponds to the voice recognition processing device 3 in the first embodiment.
The character string data memory in the present invention corresponds to the voice recognition memory 324 in the first embodiment.

In the first embodiment, the audio processing device 1 is used by being set, for example, in the middle of a circular table in a conference room.
As described above, the two-way communication unit 2 includes a plurality of microphones, for example, six microphones facing the conference attendees, inputs the voice of each conference attendee, selects one microphone signal, and performs voice recognition. While outputting to the processing apparatus 3, the selected microphone number is notified. Here, the voice of each conference attendant is assumed to be a command with a voice of about 2 to 3 seconds such as “Open File”, “Next”, and the like.
The voice recognition processing device 3 performs voice recognition processing on the microphone signal selected by the two-way call unit 2 according to a predetermined procedure described later, and converts the microphone signal into character string data.

  In the bidirectional communication unit 2 of the voice processing device 1 shown in FIG. 5, the A / D converter block 21 is not a 2CH A / D converter as described with reference to FIG. It is composed of 1-CH A / D converters 211 to 216. 5 does not use the speaker 28, the peripheral portion thereof and the DSP 23 for performing echo cancellation processing are not required, and these are not described in FIG.

In the voice processing device 1 shown in FIG. 5, when the two-way call unit 2 and the voice recognition processing unit 3 are integrated, the microphone signal S261 selected by the two-way call unit 2 is recognized as an analog signal. Since it is not necessary to supply the signal to the processing device 3, it is supplied to the speech recognition processing device 3 as a digital signal S22 indicated by a dotted line. In the following description, it is assumed to be supplied as an analog signal (signal S261).
Further, as described above, the selected microphone information (the microphone numbers 1 to 6) is supplied to the voice recognition processing device 3 as MC_SEL.

The speech recognition processing device 3 includes an A / D converter 31 and a speech recognition processing unit 32. The speech recognition processing unit 32 includes a CPU 321, a buffer 322, a speech recognition unit 323, and a speech recognition memory 324. Have
The A / D converter 31 receives the microphone signal (S261), which is an analog signal selected by the bidirectional communication unit 2, and converts it into a digital signal.
The voice recognition processing unit 32 inputs the digitized microphone signal from the A / D converter 31 and the microphone signal selected by the two-way call unit 2. Therefore, when the microphone is switched in the two-way communication unit 2, the microphone signal and its microphone information (microphone numbers 1 to 6) MC_SEL that are sequentially updated are supplied to the voice recognition processing unit 32.

The CPU 321 performs overall control of the voice recognition processing unit 32, and particularly performs input signal level monitoring and switching control based on microphone information, which will be described later.
The buffer 322 includes a plurality of buffers for each microphone number, and sequentially holds the microphone signal of the bidirectional communication unit 2 captured by the A / D converter 31 in each buffer of the microphone number based on the microphone switching signal MC_SEL. To do.
The voice recognition unit 323 takes in the microphone signal held in the buffer 322 at the timing when the microphone switching signal MC_SEL is switched, and performs voice recognition processing.
The voice recognition memory 324 stores character string data corresponding to a voice command input in advance. The voice recognition unit 323 performs voice recognition processing on the input microphone signal, collates character string data corresponding to the voice command stored in the voice recognition memory 324, and selects a matching one.

6A to 6D are timing charts illustrating the control operation performed by the voice recognition processing unit 32. FIG.
FIG. 6A is a timing chart of the microphone switching signal MC_SEL supplied from the two-way communication unit 2. For example, when “# 4” is described, it indicates that the microphone number 4 is currently selected. Yes.
FIG. 6B is a timing chart of the microphone signal S <b> 261 supplied from the two-way call unit 2. The microphone signal S261 is an audio signal corresponding to the microphone number indicated by the microphone switching signal MC_SEL in FIG. 6A, converted into digital by the A / D converter 31, and input to the speech recognition processing unit 32. The microphone signal S261 is a voice signal of a command such as “OpenFile” and “Next” as described in FIG.
FIG. 6C is a timing chart showing a processing process performed by the speech recognition processing unit 32 based on the information obtained in FIGS. It consists of buffering of each voice data and voice recognition processing after buffering.
FIG. 6D is a timing chart of the character string data S32 sequentially output as a result of the speech recognition process shown in FIG.

Although details will be described later with reference to a flowchart, the outline of the operation of the speech recognition processing unit 32 will be described with reference to FIG.
First, as shown in FIG. 6A, the microphone number initially selected by the two-way communication unit 2 is # 4, and the voice data S261 “Open File” of the microphone number 4 is input to the voice recognition processing device 3. (FIG. 6B). The voice recognition processing unit 32 inputs the voice digital signal of the signal S261 via the A / D converter 31, starts buffering as shown in FIG. 6C, and the voice data is the microphone number of the buffer 322. It is held in a buffer corresponding to # 4.

Thereafter, when the selected microphone is changed in the two-way communication unit 2 and the microphone number is changed from # 4 to # 1, the microphone switching signal MC_SEL = 1.
As shown in FIG. 6B, the voice data of the microphone number # 1 is voice data corresponding to “Next”, and the voice recognition processing unit 32 ends the buffering of the microphone number # 4 and starts a new microphone number. While starting the buffering of # 1, the voice recognition unit 323 performs voice recognition processing in parallel based on the voice data of the microphone number # 4 held in the buffer.
In the voice recognition processing, the voice data of the microphone number # 4 is subjected to voice recognition processing, collated with a command group of character string data stored in the voice recognition memory 324, and a matching one is selected, and “ “Open File” is output as a signal S32 as shown in FIG.

Thereafter, the same applies even if the microphone number changes from # 1 to # 2.
The control operation outlined above will be further described with reference to the flowchart.

A flowchart of control performed by the CPU 321 will be described below with reference to FIGS. FIG. 7 is a diagram illustrating a main flow of control performed by the CPU 321.
In FIG. 7, first, for example, a 2 kHz T1 timer is started, and a transition to a T1 timer interrupt shown in FIG. 8 is made every 50 μs. If there is a voice input of a certain level or higher (step ST11), the process proceeds to step ST12. It goes without saying that this constant level threshold value can be appropriately set according to the application.
Since the microphone switching signal MC_SEL is supplied to the voice recognition processing device 3, if there is a voice input at a certain level or higher in step ST11, the voice recognition processing apparatus 3 knows the microphone number (1 to 6) of the voice. Accordingly, in step ST12, sampling of the input voice data is started, and the voice data is held in a buffer corresponding to the microphone number (1 to 6) of the voice.
If there is no voice input above a certain level, nothing is done in step ST12.

FIG. 10 is a diagram showing an interrupt flow when the microphone selection information is changed in the control of the main flow shown in FIG. That is, in the main flow which is a normal control operation, an interrupt flow is generated when the microphone number selected by the two-way communication unit 2 changes and the information is notified through the microphone switching signal MC_SEL. For example, when audio data of microphone number 4 (microphone switching signal MC_SEL = 4) is sampled and stored in the buffer of microphone number 4 before this interruption, the microphone switching signal MC_SEL changes from 4 to 1. This is the case.
If audio sampling has been performed in step ST40 of FIG. 10, no more audio data is stored in the buffer.
In this case, it is considered that the speech input from the currently performed microphone number 4 is completed, and the sampling is terminated (step ST41).
Furthermore, the voice data of microphone number 4 for which sampling has been completed is delivered to the voice recognition unit 323, and voice recognition processing is performed (step ST42). In the example of FIG. 6, the voice recognition unit 323 recognizes the voice data of the microphone number 4 as “Open File” and outputs the character string data to the outside of the voice processing device 1 as the signal S32.

As described above, in step ST10 of the main flow in FIG. 7, the T1 timer is started, and for example, the T1 timer interrupt flow shown in FIG. 8 is started every 50 μs (20 kHz). In the T1 timer interruption, it is monitored whether there is an audio input every 5 μs and whether there is an audio input exceeding a certain level, and appropriate measures are taken. First, it is checked in step ST20 whether audio sampling has been performed.
If voice sampling has been performed, it is further checked whether or not there is a certain level of voice input (step ST21). If there is a certain level of voice input, the T2 timer described later stops. The T2 timer monitors a state where there is no utterance. When there is no utterance for a certain period of time, the T2 timer automatically shifts to speech recognition, which is the next phase.
If the utterance, that is, the voice input is above a certain level, it is considered that the utterance is continuing, and the T2 timer is reset in step ST22.
Further, although voice sampling is performed in step ST20, if there is no voice input of a certain level or more, the current utterance may be terminated. Therefore, in order to monitor the duration time when there is no utterance, T2 A timer is started (step ST23).
Even if there is no voice input of a certain level or higher in step ST21, speech sampling is continued because there is a possibility that speech is resumed (step ST24).

If audio sampling is not performed in step ST20, it is checked in step ST25 whether there is an audio input of a certain level or higher. Thereby, it is checked whether or not the utterance has started, and if there is an audio input of a certain level or more, it is assumed that the utterance has started and audio sampling is started in the buffer corresponding to the newly selected microphone ( Step ST26).
If there is no voice input exceeding a certain level in step ST25, nothing is done and the next valid utterance is awaited.

In step ST23 of FIG. 8, for example, a 2 Hz T2 timer is started and a certain time has elapsed, that is, audio sampling is performed (step ST20), but there is no audio input above a certain level for a certain time. In this case, since it is useless to continue the audio sampling, the process proceeds to the T2 timer interrupt flow shown in FIG.
That is, the voice sampling performed at that time is terminated (step ST30), and the process proceeds to the voice recognition process (step ST31).
After shifting to the voice recognition process, in step ST32, the T2 timer is reset for the next utterance process.

The first embodiment of the speech processing apparatus according to the present invention has been described above.
According to the voice processing device 1 in the first embodiment, even when a plurality of people overlap each other and issue commands to the voice processing device 1 by voice through microphones facing each of a plurality of conference attendees, The sound pressure level of each voice band is analyzed according to the characteristics of the two-way communication unit 2 of the voice processing apparatus 1, the main speaker is identified, and the voice signal is delivered to the voice recognition processing apparatus 3. Therefore, in the voice recognition processing device 3, even when a plurality of voice commands are input at the same time, the possibility of erroneous recognition processing can be avoided as much as possible. Is possible.
The voice recognition processing device 3 buffers the delivered voice command signal, performs voice recognition processing on the buffered voice signal, collates with the command character string data stored in the voice recognition memory, and matches the matching character string data. Select and process.
In addition, the voice recognition processing device 3 is sequentially notified of the microphone number selected by the two-way call unit 2. Therefore, when the selected microphone number is switched, the buffering is stopped, the voice signal that has been buffered up to that point is subjected to voice recognition processing, and the voice command signal from the updated microphone number is buffered. Since it starts, the accuracy of speech recognition is improved.

Second Embodiment Next, a second embodiment will be described with reference to FIG.
FIG. 11 is a block diagram of the speech processing apparatus 10 according to the second embodiment.
In FIG. 11, the voice processing device 10 includes a two-way call unit 2 and a voice recognition processing device 4.
In the second embodiment, the audio processing device 10 is used by being set, for example, in the middle of a circular table in a conference room.
As in the first embodiment, a two-way call unit 2 is used, and the two-way call unit 2 includes a plurality of microphones, for example, six microphones facing the conference attendees, and inputs the voices of the conference attendees. The main speaker is identified by the method described above, one microphone signal is selected, and the microphone signal is supplied to the subsequent speech recognition processing device 4 together with the voice signal.
The voice recognition processing device 4 recognizes and judges a speaker by voice recognition and voiceprint recognition of the microphone signal selected by the two-way call unit 2, and based on the role setting of each conference attendee, Authentication processing is performed for the instruction command.
Further, the voice recognition processing device 4 can interactively set the role setting of each conference attendee.

  The two-way communication unit 2 is the same as the two-way communication unit 2 described in detail in the first embodiment. In the speech processing apparatus 1 according to the first embodiment, the DSP 23 having the echo cancel function is not mounted because the two-way communication unit 2 does not use the speaker 28. However, in the present embodiment, the speaker 28 is used. Therefore, the two-way communication unit 2 having the same configuration as the two-way communication unit 2 described with reference to FIG. 1 is illustrated in FIG.

Accordingly, the difference between the speech processing apparatus 10 in the present embodiment and the speech processing apparatus 1 in the first embodiment resides in the speech recognition processing apparatus 4, and therefore, the speech recognition processing apparatus 4 will be mainly described in the following description.
Hereinafter, the configuration of the speech recognition processing device 4 will be described.
The voice recognition processing device 4 includes an A / D converter 41, a voice / voiceprint recognition unit 42, a control processing unit 43, a voice prompt storage unit 44, a voiceprint / role data storage unit 45, and a voice synthesis processing unit 46. And a D / A converter 47.
The voice / voiceprint recognition unit 42 includes a voice recognition unit 421 and a voiceprint recognition unit 422.

The microphone selection means in the present invention corresponds to the two-way communication unit 2 of the present embodiment.
The voice conversion means in the present invention corresponds to the voice recognition unit 421 of the present embodiment.
The voiceprint generation means in the present invention corresponds to the voiceprint recognition unit 422 of the present embodiment.
The instruction code processing means in the present invention corresponds to the control processing unit 43 of the present embodiment.
The authority data memory in the present invention corresponds to the voiceprint / role data storage unit 45 of the present embodiment.
The voice output means in the present invention corresponds to the voice synthesis processing unit 46 of the present embodiment.

The A / D converter 41 inputs the microphone signal of the microphone number selected by the two-way call unit 2 as an analog signal and performs digital conversion.
The voice / voiceprint recognition unit 42 takes in the voice data generated by the A / D converter 41, performs voice recognition and voiceprint recognition, and outputs character string data to the control processing unit 43 by voice recognition processing. The voiceprint data obtained by the voiceprint recognition process is stored in the voiceprint / role data storage unit 45. Speech recognition and voiceprint recognition are handled by the speech recognition unit 421 and the voiceprint recognition unit 422, respectively.
The control processing unit 43 incorporates a CPU and controls the voice recognition processing device 4 as a whole. In addition, a voice command execution authority authentication process, which will be described later, is performed.
The voice command execution authority authentication process is, for example, the role assigned to each conference participant by the voiceprint data assignment process and the authority corresponding to the role in each meeting when the role and the authority corresponding thereto are defined in advance. Based on this, the authentication process is performed on the command spoken by the conference participant, and the command is processed only when it is permitted.

The voice prompt storage unit 44 stores voice prompt data in units such as phonemes and syllables for the voice processing device 10 to generate a voice uttered to the conference participant through the speaker 28.
The voiceprint / role data storage unit 45 stores role table data and voiceprint table data, which will be described later. As a result, the control processing unit 43 can identify the conference attendee who has issued the input voice command, and can refer to the role of each conference participant and the command authority corresponding thereto.
In the voice synthesis processing unit 46, voice prompt data in units such as phonemes and syllables stored in the voice prompt storage unit 44 is supplied via the control processing unit 43, and natural accents and intonations obtained by synthesizing them are added. A voice signal is generated.
The D / A converter 47 converts the analog voice signal generated by the voice synthesis processing unit 46 into a digital signal, and outputs the digital signal to the bidirectional communication unit 2 as a signal S47.
The voice signal generated by the voice synthesis processing unit 46 is subjected to echo cancellation processing by the DSP 23 of the two-way call unit 2 and then transmitted to the conference participant through the speaker 28.

  In the case where the two-way communication unit 2 and the voice recognition processing device 4 are integrally configured, it is not necessary to convert the voice data into an analog signal, so that the D / A converters 261 and A in the two-way communication unit 2 are not required. The / D converter 263 and the A / D converter 41 and the D / A converter 47 in the speech recognition processing device 4 are not necessary.

Hereinafter, the operation of the voice processing device 10 will be described with reference to an example of how to proceed in an actual conference.
In the present embodiment, for example, a command group described in the command column in Table 1 below is a voice print / role data of the voice recognition processing device 4 as a command to the voice processing device 10 issued by a conference participant. It is assumed that it is defined in the storage unit 45.

In Table 2, the common, chairperson, etc. listed in the role column are the roles assigned to each conference attendee. A command group permitted only for each role is set. That is, each command is valid only for the attendees who are assigned the respective roles.
Thus, for example, in Table 2, the attendees assigned to the assistant are “Light”
It is valid only when the commands “On” and “Light Off” are instructed, and the meeting attendees assigned to the chairperson are “Next Presenter”, “Start”.
Valid only when the command “Record” and “Stop Record” are specified. “Common” in the role column means that the authority is given to all the attendees of the conference and any attendance of the conference Even if the user utters the command “Start Meeting”, it is treated as effective.
The contents column of Table 2 shows the contents that each command means. Is it output to the outside? In the column of, a command described as Yes means that a command script for the command is output to the outside, and a command of No means that the command is not output to the outside. For example, when a “Light On” command is instructed, this command is a command to be output to the outside. Therefore, the command script is output to the outside of the sound processing apparatus 10 and an external device that controls indoor lighting (not shown). In contrast, this is a control command for turning on the room lighting.

Next, the operation of the display control unit 10 in the present embodiment will be described.
In the present embodiment, commands such as “Light On” and “Next Presenter” are issued by voice, for example, by a conference attendant, and are input to the six microphones MIC 1 to 6 included in the interactive communication unit 2 in FIG. Is done.
In the two-way communication unit 2, the input voice command is processed, and an analog signal S 261 is supplied to the voice recognition processing device 4 via the D / A converter 261. In addition, as described above, even when a plurality of voice commands are input to the six microphones MIC1 to MIC6 in the two-way communication unit 2, the sound pressure level for each band is evaluated and the voice level is relatively set. One high microphone is selected, and an input voice command from the microphone is supplied to the voice recognition processing device 4.

The analog signal S261 of the voice command input to the voice recognition processing device 4 is converted into a digital signal by the A / D converter 41 of the voice recognition processing device 4, and is taken into the voice / voiceprint recognition unit.
The voice / voiceprint recognition unit 42 performs voice recognition of the digitized voice command signal, converts it into command data, and outputs it to the control processing unit 43. Further, the voice / voiceprint recognition unit 42 performs voiceprint recognition processing on the input voice command signal and outputs the voiceprint data to the control processing unit 43.

The control processing unit 43 first collates the voiceprint data input from the voice / voiceprint recognition unit 42 with the data stored in the voiceprint / role data storage unit 45.
Here, in the present embodiment, voice print data of conference attendees is stored in advance as voice print table data in the voice print / role data storage unit 45 in a form corresponding to the role. Therefore, the control processing unit 43 can identify the conference attendee who issued the command by comparing the input voice print data with the voice print data stored in the voice print / role data storage unit 45.

The voiceprint / role data storage unit 45 further includes the voiceprint data of the conference attendees, the roles of each conference attendee, and a command group having authority corresponding to each role. Stored.
Therefore, by referring to the voiceprint / role data storage unit 45, the control processing unit 43 can know not only the conference participant who issued the command but also a command group permitted to the conference attendee in advance. Therefore, it can be determined whether or not the command is valid. When the command is valid for the identified conference attendee, the voice / voiceprint recognition unit 42 processes the command data converted by voice recognition, and when the command is not valid, the voice / voiceprint recognition unit The command data converted by voice recognition at 42 is ignored.
Furthermore, the role table data stored in the voiceprint / role data storage unit 45 includes a flag indicating whether or not to output to the outside of Table 2 (“Yes”, “No”), and is defined for each command. ing. The control processing unit 43 refers to this flag, and when the command is valid and the external output is a flag meaning “Yes”, the command data is output to the outside, and the command is valid but the external When the output is “No”, the command data is processed internally without outputting to the outside.

The above processing will be further described using a specific example.
For example, assume that Mr. A to Mr. F are holding a meeting, and Mr. A is an assistant. The voiceprint data of Mr. A is acquired in advance and stored in the voiceprint / role data storage unit 45.
Here, when Mr. A issues a command “Next Presenter” (voice), this command is input to the voice processing apparatus 10 via the microphone facing Mr. A, and is taken into the voice / voiceprint recognition unit 42. The voice print data and the voice print data recognized by the data converted to “Next Presenter” (command data) are supplied to the control processing unit 43.
The control processing unit 43 collates the supplied voiceprint data with the voiceprint table data stored in the voiceprint / role data storage unit 45. As a result, the input voice command is by Mr. A, and Mr. A is an assistant. I know that there is. Then, referring to the role table data, since the command of “Next Presenter” is not included in the authority of the assistant role, it is determined that the command is out of authority, and the voice / voiceprint recognition unit 42 recognizes the voice and generates it. “Next Presenter” (command data) is not processed.

  In addition, when a plurality of people among Mr. A to Mr. F are issuing commands, the sound pressure level for each voice band is analyzed according to the characteristics of the two-way communication unit 2, and one talk The person is identified and the voice signal is delivered to the voice recognition processing device 4. Therefore, in the voice recognition processing device 4, it is possible to avoid the possibility of malfunction even if a plurality of voice commands are input at the same time, and it is possible to appropriately determine and process.

The configuration and operation of the audio processing device 10 have been described above using specific examples.
The present embodiment is not limited to the above contents and can be variously modified.
For example, although role table data and voiceprint table data are stored in advance in the voiceprint / role data storage unit 45, they can be added / changed / deleted as appropriate through the control processing unit 43. Similarly, the voice prompt storage unit 44 can add / delete voice prompt data.

As described above, according to the voice processing apparatus 10, a voice command is input, voice recognition and voiceprint recognition are performed on the voice command, and a speaker of the voice command is set based on preset voiceprint table data. The role of the speaker and the role of the command are identified by the role table data, so that the processing according to the authority of the voice command issued by the speaker can be performed.
Furthermore, since the voice processing device 10 has the two-way communication unit 2, as described above, even when a plurality of voice commands are input, erroneous detection / malfunction can be avoided as much as possible, and an appropriate determination is made.・ Processing can be performed.

Third Embodiment Next, a third embodiment will be described.
The voice processing apparatus 10 according to the third embodiment can be applied as it is with the voice processing apparatus 10 according to the second embodiment.
In the third embodiment, the voiceprint table data stored in the voiceprint / role data storage unit 45 in the voice recognition processing device 4 described in the second embodiment is used as, for example, bi-directional communication between a conference attendee and the voice processing device 10. Generate in an interactive exchange.
Accordingly, the voice prompt storage unit 44, the voice synthesis processing unit 46, the D / A converter 47, and the like that are not used in the second embodiment are used in the present embodiment.

  As already described, the voice prompt storage unit 44 stores voice prompt data in units of phonemes and syllables for the voice processing device 10 to generate voices uttered to the conference participants through the speaker 28, and voice synthesis. The processing unit 46 is supplied with voice prompt data in units such as phonemes and syllables stored in the voice prompt storage unit 44 via the control processing unit 43. The voice synthesis processing unit 46 synthesizes voice prompt data to generate a voice signal with natural accents and intonations.

Further, the present embodiment is characterized in that the control processing unit 43 performs voiceprint data allocation processing with respect to the control processing unit 43 in the second embodiment.
The voiceprint data allocation process is, for example, voiceprint data allocation based on voice data spoken by the conference participant and role data of the conference participant in the conference. The voice recognition processing device 4 supplies the voiceprint data through the signal S261. The role of the conference participant who is the source of the audio signal is automatically set from the audio signal from the interactive communication unit 2 to be performed.
Hereinafter, the operation of the voice processing apparatus 10 in the interactive exchange will be described using an example of an exchange between an actual conference attendee and the voice processing apparatus 10.

First, a command serving as a trigger for starting interactive exchange is set in advance and stored in the role table data of the voiceprint / role data storage unit 45. For example, in Table 2 above, the command “Start Meeting” corresponds. If the role having the authority of this command is made common and the role is set to be common to all the meeting attendees in the voiceprint table data, for example, any meeting attendee will emit “Start Meeting” (voice). However, the start of interactive processing can be permitted.
When a command serving as a trigger for such interactive exchange is issued from a conference participant, the voice recognition processing device 4 captures the voice command as a command signal through the two-way call unit 2. The voice / voiceprint recognition unit 42 performs voice recognition and voiceprint recognition on the command signal, and supplies command data and voiceprint data to the control processing unit 43.
Based on the input voiceprint data, the control processing unit 43 refers to the voiceprint / role data storage unit 45 to confirm its role and authority. Since a command as a trigger such as “Start Meeting” is permitted to all as described above, “Start Meeting” (command) is executed.

When the “Start Meeting” command is executed, the control processing unit 43 generates voice data such as “Who is the chairperson?”, For example, so that the phoneme data corresponding to the voice data is generated from the voice prompt storage unit 44. The voice synthesis processing unit 46 is controlled so as to extract voice prompt data such as syllable data and synthesize them to create voice data.
The voice synthesis processing unit 46 performs voice synthesis based on the voice prompt data supplied via the control processing unit 43 and generates a digital signal corresponding to “Who is the chairperson?”. In this case, processing is performed so that natural accents and intonation are added.
The digital signal corresponding to "Who is the chairperson?" Is directly converted into an analog signal as a digital signal or by the D / A converter 47 and the D / A converter 263, and the two-way communication unit 2 The DSP 23 performs echo cancellation processing. Then, it is amplified by an amplifier circuit (not shown), and the speaker 28 asks the conference attendee, “Who is the chairman? ”Is transmitted.

Assume that one of the attendees, for example, Mr. B, responds to this voice as follows: “Let me do the facilitator today.” This voice is also processed in the same way as “Start Meeting”, but in this case, based on the voiceprint recognition processing of the voice / voiceprint recognition unit 42, it is supported from the remarks of “Let me do the facilitator today” Voice print data is obtained.
Here, the control processing unit 43 associates the voice print data of Mr. B with the role of the chairman. That is, the voice print table data in the voice print / role data storage unit 45 is written, and the settings shown in Table 3 to be described later are made to associate Mr. B's voice print data with the role of chairman.
Therefore, from now on, among the commands remarked by Mr. B, the processing is effective for the commands for which the authority is given to the chairperson in advance.

  When the chairperson is set, the voice recognition processing device 4 sets, for example, “Who will support today's meeting? "Who is the chairman?" ], The voice prompt data in the voice prompt storage unit 44 is synthesized by the voice synthesis processing unit 46 and created.

  By performing the processing described above, for example, the roles of each person and the voiceprint data are sequentially associated with the voiceprint table data through the interactive exchange between the voice processing apparatus 10 and the meeting attendee as described below, for example.

<Mr. A>"StartMeeting"
<Equipment> “Who is the chairperson? 』
<Mr. B>"Let's do the facilitator today"
→ The chairperson is automatically assigned to Mr. B's voiceprint data.
<Equipment> “Who will assist today's meeting? 』
<Mr. C> “I will be in charge”
→ An assistant role is automatically assigned to Mr. C's voiceprint data.
<Mr. B> “Next Presenter”
<Equipment> “Who is going to explain? 』
<Mr. D>"I will explain"
→ An explanatory character is automatically assigned to Mr. D's voiceprint data.

  Through the interactive exchange between the voice processing apparatus 10 and the meeting attendee described above, the voiceprint / role data storage unit 45 generates voiceprint table data as shown in Table 3 below.

Once the voiceprint table data is created, the control processing unit 43 performs command authority authentication according to each role by the same operation as in the second embodiment, as shown in the following example. Note that the contents of the commands in the following example are the same as those shown in Table 2.

<Mr. D>"Open File Urage"
→ Uriage file is opened on PC as explanatory material.
<Mr. C> “Light Off”
→ Indoor lighting is turned off.
<Mr. B>"StartRecord"
→ Recording of minutes starts.
<Mr. B> “Light On”
→ Mr. B has no authority to execute this command and is ignored.

In the above example, the audio processing apparatus 10 is connected to a PC installed in the conference, and the URI file is opened as a result of the PC being controlled by the “Open File URI” command with valid external output.
Further, for example, when a plurality of people among Mr. A to F who are meeting attendees are issuing commands, the sound pressure level for each voice band is analyzed according to the characteristics of the two-way communication unit 2. Thus, one speaker is specified and the voice signal is delivered to the voice recognition processing device 4. Therefore, in the voice recognition processing device 4, it is possible to avoid the possibility of malfunction even if a plurality of voice commands are input at the same time, and it is possible to appropriately determine and process.

Heretofore, the configuration and operation of the speech processing apparatus 10 in the present embodiment have been described using specific examples.
As described above, according to the speech processing apparatus 10, the role of the speaker can be set in the interactive exchange with the speaker.
Further, according to the voice processing device 10, a voice command is input, voice recognition and voiceprint recognition are performed on the voice command, and the voice command table data set in the interactive exchange is used to speak the voice command. The role of the speaker and the speaker and the role table data can be identified from the role table data, so that processing according to the authority of the voice command issued by the speaker, for example, can be performed.
Furthermore, since the voice processing device 10 has the two-way communication unit 2, as described above, even when a plurality of voice commands are input, erroneous detection / malfunction can be avoided as much as possible, and an appropriate determination is made.・ Processing can be performed.

Fourth Embodiment Hereinafter, a speech processing apparatus according to a fourth embodiment will be described.
The speech processing apparatus in the fourth embodiment has the same configuration as the speech processing apparatus 10 in the second and third embodiments.
In the fourth embodiment, an interactive role setting process between the speaker and the speech processing device 10 is performed by further combining speech recognition processing in the speech / voiceprint recognition unit 42 of the speech processing device 10. In addition, it is possible to assign and end roles from natural conversations.

For example, in the example of Table 4, “I want to start the meeting” “Start meeting” and “I will explain” “Explanation” keywords such as voice / voiceprint recognition By recognizing the part 42, the chairperson and the explanation role, which are roles corresponding to those keywords, are assigned to the respective speakers. That is, the voiceprint data acquired from the speaker is associated with the role in the conference and stored in the voiceprint table data of the voiceprint / role data storage unit 45.
In addition, the keyword “End Conference” in “Now the conference is ended” is recognized by voice, and for example, the grant of the authority of the chairperson is ended.
As described above, in the present embodiment, it is possible to develop the voice processing device 10 in the third embodiment and assign and end a role from a natural conversation.

3 is a block diagram of a two-way call unit 2. FIG. 3 is a block diagram of a DSP 22 of a two-way call unit 2. FIG. It is a figure which shows the FFT result of the directional microphone of the bidirectional | two-way call part. It is a figure which shows the output of the selection microphone signal of the two-way call part. 1 is a block diagram of a voice processing device 1. FIG. 4 is a timing chart for explaining the operation of the audio processing device 1. 4 is a flowchart for explaining the operation of the audio processing device 1. 4 is a flowchart for explaining the operation of the audio processing device 1. 4 is a flowchart for explaining the operation of the audio processing device 1. 4 is a flowchart for explaining the operation of the audio processing device 1. 1 is a block diagram of a voice processing device 10. FIG.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1,10 ... Voice processing apparatus, 2 ... Two-way communication part, 21 ... A / D converter block, 22 ... 1st digital signal processor (DSP), 23 ... 2nd digital signal processor (DSP), 24 ... CPU, 25 ... codec, 26 ... D / A converter block, 27 ... amplifier block, 28 ... speaker, 3 ... voice recognition processing device 3, 31 ... A / D converter, 32 ... voice recognition processing unit, 321 ... CPU 322: Buffer, 323: Voice recognition unit, 324 ... Voice recognition memory, 4 ... Voice recognition processing device, 41 ... A / D converter, 42 ... Voice / voice print recognition unit, 421 ... Voice recognition unit, 422 ... Voice print recognition Units 422, 43 ... control processing unit, 44 ... voice prompt storage unit, 45 ... voiceprint / role data storage unit, 46 ... voice synthesis processing unit, 47 ... D / A converter

Claims (3)

An audio processing device that inputs and processes audio signals from a plurality of microphones,
A voice processing apparatus comprising: a microphone selecting unit that selects one microphone based on the voice signal; and a voice converting unit that converts a voice signal output from the microphone selected by the microphone selecting unit into character string data. . The sound processing apparatus according to claim 1, wherein the microphone has a unidirectional characteristic. The voice conversion means is
A buffer that holds the audio signal; and a character string data memory that stores character string data corresponding to the audio signal;
The audio signal output from the microphone selected by the microphone selection means is taken into the buffer,
The speech processing apparatus according to claim 1, wherein at the timing when the selected microphone changes, the speech signal captured in the buffer is collated with the character string data memory and converted to matching character string data.

Images