JP2015191220A - Voice processing system, voice processing method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a voice processing system, a voice processing method, and a program which can accurately detect a voice during output of a response signal even when the response signal and the input voice are the voice.SOLUTION: A voice processing system includes: voice detection means for detecting a voice using a frequency band used for voice detection of an input voice; band estimation means for notifying a response selection part of a band excluding the frequency band used for the voice detection of the input voice; and response selection means for selecting the response voice in which the frequency band known beforehand includes a large number of components of the notified band among the known response voices.

Description

  The present invention relates to a voice processing system, a voice processing method, and a program.

  In the voice response / voice dialogue, there is a barge-in function for interrupting and inputting a voice while outputting a response signal or a response voice. The barge-in function is used, for example, when rephrasing due to misrepresentation or misrecognition of voice recognition, when performing the next input immediately without waiting for a response, or when re-executing because of a change in mind. However, there is a problem that the performance of voice detection and voice recognition deteriorates when the response signal goes around the microphone for voice input.

  On the other hand, there is a method of paying more attention to the frequency characteristics of the response signal and performing greater weighting on the frequency band with less response signal (Patent Document 1). The method described in Patent Document 1 weights each frequency band based on a system sound of multiple channels, that is, a response signal (for example, beep sound or music), and applies a smaller weight to a band that is likely to be included in the response signal. / Non-voice discrimination. Therefore, the method described in Patent Document 1 can suppress degradation in performance such as voice detection when the overlap of frequency bands between the response signal and the input voice is small.

JP2012-189907

When the response signal and the input voice are voices such as a human voice as in the voice dialogue system, the frequency bands of the response voice and the input voice overlap greatly. Therefore, the technique described in Patent Document 1 has a problem that the performance of voice detection and the like is lowered.

[Purpose of the invention]
An object of the present invention is to provide a voice processing system, a voice processing method, and a program capable of accurately detecting a voice while outputting a response signal even when the response signal and the input voice are voices.

  The present invention provides a voice detection means for performing voice detection using a frequency band used for voice detection of input voice, and a response selection of a band excluding the frequency band used for voice detection of the input voice. A voice processing system comprising: band estimation means for notifying a unit; and response selection means for selecting a response voice that contains a large amount of the notified band components among response voices whose frequency bands to be used in advance are known.

  The present invention provides a voice detection step for performing voice detection using a frequency band used for voice detection of input voice, and a response selection of a band excluding the frequency band used for voice detection of the input voice. And a response selection step of selecting a response sound including a large amount of the notified band component from response sounds whose frequency band to be used in advance is known.

The present invention provides a voice detection step for performing voice detection using a frequency band used for voice detection of input voice, and a response selection of a band excluding the frequency band used for voice detection of the input voice. This is a program for causing a computer to execute a band estimation step to be notified to a unit and a response selection step to select a response sound that contains a large amount of the notified band component from response sounds whose frequency bands to be used in advance are known.

Even when the response signal and the input voice are voices as in the voice dialogue system, voice detection can be performed with high accuracy during the output of the response signal.

It is a hardware block diagram concerning the 1st embodiment of the present invention. It is a block diagram concerning the 1st embodiment of the present invention. 3 is a flowchart according to the first embodiment of the present invention. It is a block diagram concerning the 2nd embodiment of the present invention. It is a conceptual diagram concerning the 2nd Embodiment of this invention. It is a flowchart which concerns on the 2nd Embodiment of this invention. It is a block diagram concerning the 3rd embodiment of the present invention. It is a flowchart which concerns on the 3rd Embodiment of this invention.

Embodiment 1. FIG.
FIG. 1 is a hardware configuration diagram of a speech processing system 1 according to the first embodiment of the present invention. As shown in FIG. 1, the audio processing system 1 includes a CPU 10, a memory 12, a hard disk drive (HDD) 14, a communication interface (IF) 16 that performs data communication via a network (not shown), a display device 18 such as a display, and the like. The input device 20 includes a pointing device such as a keyboard and a mouse. These components are connected to each other through the bus 22 and input / output data to / from each other. Note that the hardware configuration of the voice processing system 1 is not limited to this configuration, and can be changed as appropriate. Further, the voice processing system 1 does not need to be configured by one computer system, and may be configured by a plurality of computer systems.

  FIG. 2 is a block diagram showing the configuration of the speech processing system according to the first embodiment of the present invention.

  As shown in FIG. 2, the speech processing system according to the first embodiment includes speech detection means, band estimation means 121, and response selection means 131.

  The voice detection unit 111 receives input voice and performs voice detection. When a voice is detected, the voice detection unit 111 notifies the band estimation unit 121 of detection band information indicating a frequency band in which the voice is included.

  The band estimation unit 121 selects a band excluding at least a part of the frequency band of the detection band information, and notifies the response selection unit 131 of the selected band as estimated band information.

  The response selection unit 131 selects response speech that includes a large amount of band components of the estimated band information. Note that the fact that the sound includes many components in a certain band indicates that the amount (gain) of the sound included in that band is large. The response voice is a voice output from the voice processing system. The response voice is, for example, a voice having a content that responds to a certain input voice.

  Next, the operation of the first embodiment for carrying out the present invention will be described in detail. FIG. 3 is a flowchart illustrating an example of the operation of the first embodiment.

  The voice detection unit 111 receives input voice, performs voice detection, and notifies detection band information. (Step 101). The band estimating unit 121 selects a band excluding at least a part of the frequency band of the detected band information notified from the voice detecting unit 111, and notifies the response selecting unit 131 of the estimated band information (step 102). The response selection unit 131 selects response speech that includes a large amount of band components of the estimated band information (step 103). (gain)

  According to the present embodiment, in order to select a response sound that includes many components in the band excluding the frequency band used for sound detection of the input sound (for example, immediately preceding sound detection), the response signal is being output. Even if it exists, the audio | voice detection of an input signal can be performed accurately. For example, when the input voice is a male voice and voice detection is performed in the male voice frequency band, the voice of the input signal can be accurately detected by selecting the female voice response voice.

Embodiment 2. FIG.
FIG. 4 is a block diagram showing a configuration of a voice processing system according to the second embodiment of the present invention. The speech processing system according to the second embodiment of the present invention corresponds to the first example. Note that the voice processing system according to the second embodiment of the present invention does not necessarily include the voice recognition unit 142 described later.

  The voice processing system according to this embodiment includes a voice response device 2, a response voice storage unit 212, an output unit 162, and an input unit 172. The voice response device 2 includes a voice detection unit 112, a band estimation unit 122, a response selection unit 132, a voice recognition unit 142, and a voice reproduction unit 152.

  The voice detection unit 112 receives an input voice from the input unit 172 described later and performs voice detection. The voice detection unit 112 notifies the band estimation unit 122 of detected band information when a voice is detected. The voice detection unit 112 may notify the voice recognition unit 142 of the voice detection result.

  The voice detection unit 112 may receive band estimation information described later from the band estimation unit 122 and perform voice detection in a frequency band specified by the band information. When the same user performs voice input continuously, the voice can be detected with higher accuracy by detecting the voice in the frequency band in which the previous voice or the past voice is detected. Note that there may be no band information of past detection results at the start of the conversation. Therefore, the voice detection unit 112 may perform voice detection for the entire frequency band, for example, at the start of voice detection.

  The voice detection unit 112 may divide a frequency band that is a target of voice detection into a plurality of subbands (partial bands) in advance, and notify the band estimation unit 122 of the subband including the voice as detection band information. Further, the voice detection unit 112 performs weighting according to the amount (gain) of the input voice included in each subband, and notifies the band estimation unit 122 of the detection band information with a larger weight for the subband having a larger amount. May be. Since the technique for performing voice detection by dividing into subbands is a known technique, the description thereof is omitted here.

  The band estimation unit 122 selects a band excluding at least a part of the frequency band, and notifies the response selection unit 132 as estimated band information. Further, the band estimation unit 122 may notify the voice detection unit 112 of the estimated band information. In that case, the voice detection means 112 may detect the voice at the frequency excluding the frequency of the estimated band information for the next input voice. The band estimation unit 122 may notify the estimated band information estimated from the immediately preceding sound detection result, or notify the frequency band obtained by smoothing the frequency band estimated from the most recent sound detection results as estimated band information. May be.

  The response selection unit 132 selects response speech that is appropriate as a response and includes a large amount of band components of the estimated band information notified from the band estimation unit 122 from the response voice storage unit 212 described later. For example, as illustrated in FIG. 5, the response selection unit 132 may select a response voice mainly including a component of the band of the estimated band information (a band excluding the band used for voice detection). The response selection unit 132 notifies the audio reproduction unit 152 of the selected response voice. The response selection unit 132 may select a response voice that includes many estimated band information components from the plurality of corresponding responses. Further, the response selection unit 132 may select a response voice based on a voice recognition result of the voice recognition unit 142 described later.

  Furthermore, when the estimated band information is weighted by subband, the response selection unit 132 may select the response sound according to the weight. For example, the response selection unit 132 can perform voice detection of the input signal with higher accuracy by selecting a response such that a response including many components of a large weighted subband has a low priority. As an example, among subbands B1 to B8 divided into eight in the frequency direction, B1 has a weight of 0, B2 to B3 have a large weight, B4 to B5 have a small weight, B6 has a large weight, and B7 to B8 have A case where the weight is medium will be described. In this example, the response selection unit 132 preferentially selects a response that has few components in the subbands B2 to B3 and B6 and contains many components in the subbands B4 to B5 among the plurality of response speech candidates. That's fine.

  The voice recognition unit 142 performs voice recognition using the result of voice detection performed by the voice detection unit 112.

  The sound reproduction means 152 causes the output means 172 to reproduce the response sound selected by the response selection means 132. The sound reproducing means 152 may be notified of the timing when the sound detecting means 112 starts the sound detection from the sound detecting means. The audio reproduction means 152 may stop the audio reproduction when receiving the notification. Thereby, since the reproduction of the response voice stops immediately when the voice detection is performed, the voice detection unit 112 can perform subsequent voice detection with higher accuracy.

  The output unit 162 is a unit that outputs a signal such as voice. The output unit 162 may be a speaker, for example.

  The input means 172 is a means for inputting a signal such as voice. The input unit 172 may be a microphone, for example.

  The response voice storage unit 212 stores a plurality of responses corresponding to a predetermined input signal, for example.

  Next, the operation of the second embodiment for carrying out the present invention will be described in detail. FIG. 6 is a flowchart illustrating an example of the operation of the second embodiment.

  The voice detection unit 112 receives input voice, performs voice detection, and notifies detection band information (step 201). The band estimation unit 122 selects a band excluding at least a part of the band of the detected band information, and notifies the response selection unit 132 of the estimated band information (step 202). The response selection unit 132 selects a response sound that is appropriate as a response and includes many components of the estimated band information from the response sound storage unit 212 described later (step 203). The sound reproducing means 152 causes the output means 172 to reproduce the response sound selected by the response selecting means 132 (step 204). The voice detection unit 112 performs voice detection of the next input voice, and when the voice is detected, notifies the band estimation unit 122 of the detected band information, and returns to step 202 (step 205).

  Note that after step 202, the voice recognition unit 142 may perform voice recognition using the voice detection result. In that case, in step 203, the response selection unit 132 may select a response voice based on the voice recognition result.

  According to the present embodiment, it is possible to accurately detect an input voice while outputting a response voice.

  In addition, when the same user performs voice input continuously, voice detection can be performed with higher accuracy by performing voice detection in the frequency band in which the previous or previous voice is detected.

  In addition, the voice processing system according to the present embodiment can select response voices including a band having a small weight, that is, a gain of the input voice, by weighting each subband at the time of voice detection. Therefore, it is possible to expand the variation of response voices while suppressing deterioration in accuracy of voice detection.

Embodiment 3. FIG.
FIG. 7 is a block diagram showing the configuration of a speech processing system according to the third embodiment of the present invention. The speech processing system according to the present embodiment corresponds to the second example.

  The voice processing system of this embodiment includes a voice response device 3, a response voice storage unit 213, an output unit 163, and an input unit 173. The voice response device 3 includes voice detection means 113, band estimation means 123, response selection means 133, voice recognition means 143, voice reproduction means 153, and scenario reference means 183.

  The voice detection unit 113, the band estimation unit 123, the response selection unit 133, the voice reproduction unit 153, the output unit 163, the input unit 173, and the response voice storage unit 213 are respectively a voice detection unit 112, a band estimation unit 122, and a response selection unit. 132, the sound reproduction means 152, the output means 162, the input means 172, and the response sound storage means 212 have the same functions, and thus description thereof is omitted.

  The voice recognition unit 143 recognizes the voice input from the input unit 173 using the voice detection result notified from the voice detection unit 113. The voice recognition unit 143 notifies the scenario reference unit 183 of the recognition result.

  The scenario reference unit 183 refers to the scenario storage unit 223 and notifies the response selection unit 133 of a scenario corresponding to the recognition result notified from the voice recognition unit 143.

  The response selection unit 133 selects a response voice corresponding to the notified scenario.

  The scenario storage unit 223 stores a scenario indicating the content of a response corresponding to the voice recognition result. The scenario may be specified at a text level, or may be described in a meta expression that allows the wording and vocabulary freedom even if the content is the same.

  When the scenario is specified at the text level, the response voices of the same text may be response voices in which the voice quality and way of speaking of the speakers are different and the frequency bands do not overlap each other. On the other hand, when the scenario is described in a meta-expression, the response voices showing the same contents are the response voices that do not overlap each other's frequency bands by using the difference of the wording and vocabulary in addition to the voice quality difference. Can be. For example, there are cases where the frequency band to be used mainly changes between “please do” and “please do” as expressions for “request”.

  Next, the operation of the third embodiment for carrying out the present invention will be described in detail. FIG. 8 is a flowchart illustrating an example of the operation of the third embodiment.

  First, the band estimation unit 123 performs the first band estimation and notifies the response selection unit 132 of the estimated band information (step 301). Next, the response selection means 133 selects a response voice (step 302). Next, the sound reproduction means 153 causes the output means 163 to reproduce the response sound notified from the response selection means 133 (step 303). The voice detection unit 113 performs voice detection on the input voice received from the input unit 173 (step 304). The voice detection means 113 notifies the voice recognition means 143 of the detection result when the voice is detected (step 305). The band estimation unit 123 performs the second band estimation and notifies the response selection unit 133 of the estimated band information (step 306). The voice recognition unit 143 performs voice recognition using the voice detection result and notifies the scenario reference means 183 of the voice recognition result (step 307). The scenario reference unit 183 refers to the scenario storage unit 223 (step 308). If there is a scenario corresponding to the recognition result notified from the voice recognition unit 143, the scenario reference unit 183 notifies the response selection unit 133 of the corresponding scenario and returns to step 302 (step 309).

In the plurality of flowcharts used in the above description, a plurality of processes are described in order, but the execution order of the processes executed in each embodiment is not limited to the description order. In each embodiment, the order of the illustrated steps can be changed within a range that does not hinder the contents. Moreover, each above-mentioned embodiment and modification can be combined in the range with which the content does not conflict.

DESCRIPTION OF SYMBOLS 1 Voice processing system 2 Voice response apparatus 3 Voice response apparatus 10 CPU
12 Memory 14 HDD
16 Communication IF
18 Display device 20 Input device 22 Bus 111, 112, 113 Audio detection means 121, 122, 123 Band estimation means 131, 132, 133 Response selection means 142, 143 Audio recognition means 152, 153 Audio reproduction means 162, 163 Output means 172 , 173 Input means 183 Scenario reference means 212, 213 Response voice storage means 223 Scenario storage means

Claims (10)

Voice detection means for performing voice detection using the frequency band used for voice detection of the input voice;
Band estimation means for notifying a response selection unit of a band excluding a frequency band used for voice detection of the input voice;
A voice processing system comprising: a response selection unit that selects a response voice including a large number of components in the notified band among response voices whose frequency bands to be used in advance are known.
The voice detection unit performs voice detection on a frequency band that is divided into a plurality of partial bands in advance, and the band estimation unit determines a partial band that is not a partial band used for voice detection among the plurality of partial bands. The voice processing system according to claim 1 which notifies.
3. The voice detection unit performs weighting according to an amount of input voice included in each partial band, and the response selection unit selects a response voice that contains less voice in the partial band with a large weight. Voice processing system.
The voice processing system according to any one of claims 1 to 3, wherein the response selection unit selects the response voice so that a large number of frequency bands including the immediately previous response voice are included.
A voice detection step for performing voice detection using the frequency band used for voice detection of the input voice;
A band estimation step of notifying a response selection unit of a band excluding a frequency band used for voice detection of the input voice;
And a response selection step of selecting a response sound including a large number of components of the notified band among response sounds whose frequency bands to be used in advance are known.
The voice detection step performs voice detection for a frequency band that is divided into a plurality of partial bands in advance,
The voice processing method according to claim 5, wherein the band estimation step notifies a partial band excluding a partial band used for voice detection among the plurality of partial bands.
The voice detection step performs weighting according to the amount of input voice included in each partial band,
The voice processing method according to claim 6, wherein the response selection step selects a response voice that contains a small amount of voice in a partial band having a large weight.
A voice detection step for performing voice detection using the frequency band used for voice detection of the input voice;
A band estimation step of notifying a response selection unit of a band excluding a frequency band used for voice detection of the input voice;
A program that causes a computer to execute a response selection step of selecting a response sound that includes a large number of components of the notified band from response sounds whose frequency band to be used in advance is known.
The voice detection step performs voice detection for a frequency band that is divided into a plurality of partial bands in advance,
The program according to claim 8, wherein the band estimation step notifies a partial band of the plurality of partial bands excluding a partial band used for voice detection.
The voice detection step performs weighting according to the amount of input voice included in each partial band,
The program according to claim 9, wherein the response selecting step selects a response voice that includes a small amount of voice in a partial band having a large weight.

Images