JP2008203800A - Audio controller - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an audio controller capable of preventing miscontrol resulting from wrong audio recognition even in an environment wherein home audio equipment is outputting sound. <P>SOLUTION: After making an echo cancel processing unit 11 performs echo cancel processing on output sound of the home audio equipment included in sound picked up by a sound pickup unit 10, the audio controller 1 smoothes an output signal by superposing a noise signal on the output signal of the echo cancel processing unit 11, and compares the smoothed signal with an audio recognition model to recognize audio included in the output signal. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a voice controller that is installed in a building such as a house and operates equipment in the house with voice.

  2. Description of the Related Art A voice controller that is installed in a building such as a house and operates the equipment in the house with voice is provided.

  By the way, there are various devices in the house, and it is complicated to operate them with a switch or the like. For example, the kitchen space includes a hot water controller, a floor heating controller, a lighting switch, an intercom, and the bathroom space includes a bubble bathtub controller, a lighting switch, a bathroom heating ventilation dryer switch, a television, and the like.

  In addition, the presence of a plurality of these controllers and switches has a problem that the scenery in the house is damaged.

  In order to solve these problems, an integrated controller that can control a plurality of devices has been provided, and a controller that performs the operation by voice recognition has been provided. This controller is a so-called voice controller.

When there is an utterance input switch, the voice controller needs to operate the utterance input switch, so there are many cases where the utterance input switch is not provided. However, in such a voice controller, the environmental noise ( Noise) is input to the microphone for collecting sound, there is a possibility that environmental noise is recognized as the user's speech and erroneous control is performed. In particular, when there is an audio device such as a television or radio in the target space, the output sound of the audio device in which a human voice is input to a microphone is reduced by an echo canceller. For example, the voice recognition disclosed in Patent Document 1 A method using an apparatus is also conceivable.
JP 2003-241791 A

  By the way, the echo canceller is a technique for reducing the output sound of the acoustic device from wrapping around the microphone as the sound collection unit by sequentially learning an adaptive filter using the sound signal output from the acoustic device as a reference signal. The voice recognition method is a method of extracting feature amounts from speech and identifying the utterance contents from the time series change of the feature amounts. The individual difference in the utterance volume is the input level due to the fluctuation of the distance between the speaker and the microphone. In general, the daily value of the voice power is used as the feature amount without using the voice power so as not to be affected by the fluctuation of the voice.

  On the other hand, in the voice controller that operates the equipment in the house with voice, the input gain of the sound collection unit must be set high so that the voice of the user away from the microphone can be recognized, and the sound in the kitchen, bathroom, etc. In an environment where the reflection is strong, the reverberation time becomes long, so the filter length estimated by the echo canceller also becomes long, and it becomes difficult to perform highly accurate filter learning. For this reason, there are many components that cannot be suppressed.

  As described above, the conventional voice controller has a problem that the amount of suppression of the echo canceller is not sufficient, and the output sound of the acoustic device is recognized as the user's uttered voice and erroneous control is performed.

  The present invention has been made in view of the above-described points, and an object of the present invention is to prevent erroneous control due to erroneous voice recognition even in an environment where sound equipment in the house outputs sound. To provide a voice controller.

  In order to achieve the above object, according to the first aspect of the present invention, there is provided a voice controller for recognizing a voice uttered by a user and controlling an operation of a device to be controlled in accordance with a control command defined by the recognized voice. A sound collection unit that collects the sound signal in the house, an echo cancellation processing unit that performs echo cancellation processing on the output sound of the home acoustic device included in the sound collected by the sound collection unit, and voice recognition A storage unit that stores a speech recognition model and a noise signal to be superimposed, a noise superimposing unit that superimposes a noise signal on an output signal of an echo cancellation processing unit, and smoothes the output signal, and the noise signal A speech recognition unit that recognizes speech included in the output signal by comparing the output signal superimposed with the speech recognition model, and a control command corresponding to the speech recognized by the speech recognition unit Characterized by comprising a control execution unit.

  According to the first aspect of the present invention, since the noise superimposing unit is provided and the noise signal is superimposed and smoothed on the output signal of the echo canceling unit, the speech recognition is performed even in the environment where the acoustic device in the house outputs the sound. The unit can recognize the user's voice, and as a result, erroneous control due to erroneous voice recognition can be prevented.

  According to a second aspect of the present invention, in the first aspect of the invention, an output sound detection unit that detects a magnitude of an output sound from the acoustic device is provided, and the noise superimposing unit is an output sound detected by the output sound detection unit. The size of the noise signal to be superimposed on the output signal is changed according to the size of the output signal.

  According to the second aspect of the present invention, the ratio of the noise signal to be superimposed can be reduced when the power of the output sound is low, and the ratio of the noise signal to be superimposed can be increased when the output sound is large. An excessive noise signal is superimposed when the power of the user is low, and the user's voice signal is not smoothed by the noise signal. As a result, the user's voice cannot be recognized.

  According to a third aspect of the present invention, in the first or second aspect of the present invention, the speech recognition model is configured by previously incorporating a noise signal to be superimposed on the output signal.

  According to the invention of claim 3, since the noise signal for superimposition is incorporated in the speech recognition model, the acoustic mismatch between the speech signal on which the noise signal is superimposed by the noise superimposing unit and the speech recognition model is reduced. In other words, it is possible to reduce degradation in recognition performance due to noise superimposed on the user's voice.

  According to a fourth aspect of the present invention, in the first aspect of the invention, the noise signal stored in the storage unit is a stationary noise signal of a device installed in a home.

  According to the invention of claim 4, since the noise signal for superimposition uses a noise signal generated by a device already installed in the house, the performance degradation due to the superposition of the noise signal is caused by noise during operation of the home device. The performance is reduced by the same level, and the user experience is not unnatural.

  According to a fifth aspect of the present invention, in the fourth aspect of the invention, the speech recognition model is configured by incorporating the stationary noise signal superimposed on the output signal in advance.

  According to the invention of claim 5, since the stationary noise signal for superimposing is also incorporated into the speech recognition model, an acoustic mismatch between the speech signal on which the stationary noise signal is superimposed by the noise superimposing unit and the speech recognition model. As a result, the degradation of recognition performance due to the superimposition of noise on the user's voice can be reduced.

  According to a sixth aspect of the present invention, in the fourth or fifth aspect of the present invention, the apparatus includes a device power source detection unit that detects on / off of the power source of the device installed in the house, and the noise superimposing unit is a power source of the device. When ON is turned on, the superposition of a noise signal on the output signal is suppressed.

  According to the invention of claim 6, when the power of the device is turned on, the noise signal is not superimposed, so that the noise generated by the bathroom device 2 is mixed instead of the superimposed noise signal. In addition, the S / N ratio of the user's voice and noise is not deteriorated more than expected and the recognition performance is not deteriorated, and the noise of the bathroom device smooths the echo cancellation unsuppressed component, so that the noise signal is not superimposed However, the output sound signal of the audio equipment is smoothed and is not erroneously controlled.

  According to a seventh aspect of the invention, in any one of the first to sixth aspects of the present invention, the apparatus includes a device power source detection unit that detects power on / off of the acoustic device installed in the house, and the noise superimposing unit includes: When the power source of the acoustic device is off, superimposition of a noise signal on the output signal is suppressed.

    According to the invention of claim 7, since the noise signal is not superimposed when the power of the audio equipment is off, a sufficient S / N ratio can be obtained even if the user inputs a voice with a small voice, High recognition performance can be obtained.

  In the present invention, since the noise superimposing unit is provided and the noise signal is superimposed on the output signal of the echo canceling unit and smoothed, the voice recognizing unit can perform the user's voice even in an environment where the sound equipment in the home outputs sound. Can be recognized, and as a result, erroneous control due to erroneous voice recognition can be prevented.

  Embodiments of the present invention will be described below.

(Embodiment 1)
The voice controller 1 of the present embodiment is for controlling a bathroom device 2 installed in a bathroom as shown in FIG. 1, for example, and captures a user's uttered voice as shown in FIG. 1 to generate a sound signal (voice signal). From the sound collection unit 10 such as a microphone to be converted, the echo cancellation unit 11 that performs echo cancellation processing on the output sound of the acoustic device 3 such as a bathroom television set collected by the sound collection unit 10, and the acoustic device 3 For example, output sound detection unit 12 that measures the signal power of a sound signal of an output sound that is input as a reference signal to echo cancellation unit 11 by line input, and data of a noise signal for superimposition superimposed on the output signal of echo cancellation unit 11 , A noise superimposing unit 14 that superimposes a noise signal for superimposition on the output signal of the echo canceling unit 11, a bathroom device 2 and an acoustic device 3 that are control target devices The device power source detection unit 15 that detects power on / off, the storage unit 16 that stores the data of the voice recognition model used for the voice recognition processing, and the output signal of the noise superimposing unit 14 and the voice recognition model are compared. A voice recognition unit 17 for recognizing the speech voice, and a control execution unit 18 for executing the control command corresponding to the voice recognized by the voice recognition unit 17 to control the bathroom device 2 which is the control target device of this embodiment; Consists of. The storage units 13 and 16 may be configured using the same storage device.

  FIG. 2 shows the configuration of the echo canceling unit 11, which is composed of a subtractor 11 a and an adaptive filter 11 b, and the user's uttered voice and acoustic device 3 with a microphone constituting the sound collecting unit 10. The sound signal consisting of the output sound reproduced by the speaker 3a is input to the subtractor 11a. The subtractor 11a subtracts the estimated echo signal synthesized by the adaptive filter 11b from the sound signal, erases the echo, and outputs the result. On the other hand, the adaptive filter 11b estimates the transfer characteristic in the room from the speaker 3a of the acoustic device 3 to the sound collection unit 10 from the error signal after the subtraction processing by the subtractor 11a, and outputs the output of the acoustic device 3 using the estimated filter. The estimated echo signal is output from the sound signal (reference signal) of the sound to the subtractor 11a.

  Next, the operation of the voice controller 1 of this embodiment will be described.

  First, when the user wants to control the bathroom device 2, the user inputs a predetermined sound to the sound collection unit 10.

  At this time, when the acoustic device 3 in the bathroom operates and a reproduction sound, that is, an output sound is output from the speaker 3a (see FIG. 2), the output sound is also input to the sound collecting unit 10. .

  FIG. 3A shows a waveform example of a sound signal output from the sound collection unit 10 when the output sound (II) of the acoustic device 3 is input to the sound collection unit 10 following the user's voice (I).

  The echo canceling unit 11 inputs the sound signal output from the sound collecting unit 10 and also inputs the output sound signal of the acoustic device 3 by line input, and the sound signal output from the sound collecting unit 10 by this output sound signal The process which suppresses the component of the output sound (II) of the audio equipment 3 contained in is performed. FIG. 3B is a waveform example of an output signal that is subjected to the suppression process and is output from the echo canceling unit 11.

  The noise superimposing unit 14 that has received the output signal of the echo canceling unit 11 superimposes the superimposing noise signal (see FIG. 3C) read from the storage unit 13 on the output signal of the echo canceling unit 11 (FIG. 3 ( Smoothing is performed as shown in d).

  Here, as the noise signal for superimposition, for example, an operation sound signal of a bathroom device 2 such as a bathroom dryer is used. The noise superimposing unit 14 controls the ratio of the noise signal to be superimposed based on a signal notified in correspondence with the signal power output from the output sound detection unit 12. The output sound detection unit 12 notifies the noise superimposing unit 14 of the signal power per unit time of the reference signal output from the acoustic device 3 to the echo cancellation unit 11 at predetermined time intervals. The noise superimposing unit 14 may be notified only when the signal power varies.

  In addition, the noise superimposing unit 14 is in a case where a signal indicating that the bathroom device 2 is turned on is output from the device power source detection unit 15 or in a case where the power source of the acoustic device 3 is turned off from the device power source detection unit 15. In addition, no noise superposition is performed.

    The speech recognition unit 17 compares the output signal from the noise superimposing unit 14 with the speech recognition model and performs speech recognition processing, and when the user's speech is not input to the sound collection unit 10. Although only the output sound of the acoustic device 3 is input to the main sound unit 10, since the noise signal is superimposed on the output signal of the echo cancellation unit 11, the speech recognition unit 17 does not recognize the sound.

  On the other hand, when the user's voice is input, the power of the user's voice is sufficiently larger than the superimposed noise, so that the voice is correctly determined and the voice recognition process is performed. At this time, noise is superimposed on the user's voice, but since the voice recognition model is also created from learning data on which noise is superimposed, there is little degradation in recognition performance due to acoustic mismatch between the voice and the voice recognition model.

  Based on the recognition result by the sound recognition unit 17, the control execution unit 18 outputs a control command to the bathroom device 2 and executes predetermined control.

    As described above, the voice controller 1 according to the present embodiment provides the noise superimposing unit 14 as described above and superimposes the noise signal on the output signal of the echo canceling unit 11 so as to be smoothed. It is no longer necessary to perform speech recognition processing by determining the output sound of the acoustic device 3 that could not be completely suppressed by the cancel unit 11 as a sound, and as a result, the output sound is determined as the user's sound and erroneous control is performed. Can be prevented.

  In addition, since the output sound detection unit 12 that detects the magnitude of the output sound from the acoustic device 3 is provided, the ratio of the noise signal to be superimposed is reduced when the power of the output sound is low, and is superimposed when the output sound is high. Since the ratio of the scrambled noise signal is increased, an excessive noise signal is superimposed when the power of the output sound of the acoustic device 3 is small, and the user's voice signal is not smoothed by the noise signal. The problem of being unable to recognize the voice of can be avoided.

  Note that when the output sound is loud, the ratio of the superimposed noise increases, but the volume of the user's uttered voice increases due to the volume of the reproduced sound of the audio equipment, so it is smoothed by the noise. There is no end to it.

  Furthermore, since the noise signal for superimposition uses the noise signal generated by the bathroom device 2 already installed in the bathroom, the performance degradation due to the superposition of the noise signal is the performance due to the noise during operation of the bathroom device 2. This is almost the same as the decrease, and the user experience is not unnatural.

  In addition, when the bathroom device 2 is turned on, no noise signal is superimposed, so that noise generated by the bathroom device 2 is mixed in place of the superimposed noise signal. Since the S / N ratio of noise does not deteriorate more than expected and the recognition performance does not deteriorate, and the noise of the bathroom device 2 smooths the echo canceling unsuppressed component, the acoustic device can be used without superimposing the noise signal. The output sound signal 3 is smoothed and no malfunction occurs.

  Even when the power of the acoustic device 3 is off, no noise signal is superimposed, so that a sufficient S / N ratio can be obtained even when the user inputs a voice with a small voice, and high recognition performance is obtained. Can do.

  If a stationary noise signal, for example, a noise signal for superimposition <noise of bathroom equipment 2> is incorporated into the speech recognition model stored in the storage unit 16, the speech signal and the speech recognition model on which the noise signal is superimposed by the noise superimposing unit Will be reduced, and the degradation of the recognition performance caused by superimposing noise on the user's voice can be reduced. In order to incorporate the noise for superimposition into the speech recognition model, the speech recognition model may be created after the noise is superimposed on the speech data for learning of the speech recognition model.

It is a block diagram of one Embodiment. It is a block diagram of the echo cancellation part used for one Embodiment. It is a wave form chart for operation explanation of one embodiment.

Explanation of symbols

1 Voice controller
DESCRIPTION OF SYMBOLS 10 Sound collection part 11 Echo cancellation part 12 Output sound detection part 13 Storage part 14 Noise superimposition part 15 Device power supply detection part 16 Storage part 17 Voice recognition part 18 Control execution part 2 Bathroom equipment 3 Acoustic equipment

Claims (7)

A voice controller for recognizing a voice uttered by a user and controlling an operation of a device to be controlled according to a control command defined by the recognized voice;
A sound collection unit that collects sound signals in the house;
An echo cancellation processing unit for performing echo cancellation processing on the output sound of the in-house audio equipment included in the sound collected by the sound collection unit;
A storage unit for storing a speech recognition model to be a reference for speech recognition and a noise signal to be superimposed;
A noise superimposing unit that smoothes the output signal by superimposing a noise signal on the output signal of the echo cancellation processing unit;
A speech recognition unit that recognizes speech included in the output signal by comparing the output signal on which the noise signal is superimposed and the speech recognition model;
A voice controller comprising: a control execution unit that executes a control command corresponding to the voice recognized by the voice recognition unit. An output sound detection unit for detecting the magnitude of output sound from the acoustic device;
The voice controller according to claim 1, wherein the noise superimposing unit changes a magnitude of a noise signal to be superimposed on the output signal according to a magnitude of the output sound detected by the output sound detecting unit. The voice controller according to claim 1, wherein the voice recognition model is configured by incorporating a noise signal to be superimposed on the output signal in advance. The voice controller according to claim 1, wherein the noise signal stored in the storage unit is a stationary noise signal of a device installed in a home. The voice controller according to claim 4, wherein the voice recognition model is configured by incorporating the stationary noise signal to be superimposed on the output signal in advance. A device power supply detection unit that detects on / off of the power supply of the device installed in the house, and the noise superimposing unit is configured to output a noise signal to the output signal when the device is turned on. 6. The audio controller according to claim 4, wherein superposition is suppressed. A device power source detection unit that detects on / off of the power source of the acoustic device installed in the house, and the noise superimposing unit is configured to detect noise in the output signal when the power source of the acoustic device is off. 7. The voice controller according to claim 1, wherein superposition of signals is suppressed.

Images