JP2001042886A - Speech input and output system and speech input and output method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make it possible to execute an environmental adaptation processing to speech recognition processing for making at least the human speeches the object by adopting a constitution to execute the recognition processing in accordance with the environmental learning information learned beforehand and internally held in the state that the environmental learning processing ends >=1 times when human utters voice. SOLUTION: This system has the means for executing the speech recognition processing in accordance with the environmental learning information learned beforehand and internally held in the state that the environmental learning processing ends >=1 times when the humans utter the voices. For example, a speech recognition engine 1 of single output of a speech input and output system 10 executes the environmental learning processing to the speech recognition processing for at least the human speeches simultaneously or as a post processing of the speech recognition processing. A signal switching section 2 switches the signal inputted to the speech recognition engine 1 of single output. A mixer section 4 superposes the sample patterns of the previously prepared human utterance and the speech signals inputted from outside. A speech input section 5 converts the external speeches to the speech signals.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention has an environment adaptation function for improving a recognition rate by learning background noise other than a human voice included in an input speech signal, and performs the environment learning process. The present invention relates to a speech input / output technology using a speech recognition engine that is executed simultaneously with speech recognition processing or as a post-processing of the speech recognition processing. The present invention relates to an output system and an audio input / output method.

[0002]

2. Description of the Related Art FIG. 5 shows an example of a conventional system using a one-input speech recognition engine (first prior art). Referring to FIG. 5, the first related art is a one-input speech recognition engine having only one input that expects a human voice to be inputted as a speech signal, wherein A speech recognition engine P1 having an environment learning function of learning background noise based on a signal from which speech to be recognized has been removed, a switch unit P2, and speech input means P3 for converting external speech into a signal; By not opening the switch P2 and inputting a speech signal to the recognition engine when no utterance is made, the recognition process is not executed, and by connecting the switch P2 when a human utters, the human voice is The mixed speech signal is selected, the speech signal converted by the speech input means P3 is taken into the speech recognition engine P1, and a recognition process with an environment adaptation function is executed. There.

FIG. 6 shows an example of a conventional system using a two-input speech recognition engine (second prior art). Referring to FIG. 6, the second related art has two inputs including an input expecting a human voice to be input as an audio signal and an input expecting a background noise to be input as an audio signal. Voice recognition engine, which requires two voice signal inputs, and has an environment learning function of learning background noise based on a signal obtained by removing a voice to be recognized from a voice signal fetched from outside. Recognition engine P1
And a switch unit P2, a voice input unit P3 for converting an external voice into a signal, and an environmental sound input unit P4 for converting an external environmental sound into a signal. When the switch is opened and the switch B is connected and the speech signal is not inputted to the speech recognition engine P1, the recognition process is not executed. When a human utters, the switch A of the switch section P2 is connected and the switch is connected. By releasing B, a voice signal mixed with a human voice is selected, the voice signal converted by the voice input means P3 is taken into the voice recognition engine P1, and a recognition process with an environment adaptation function is executed. I have. As another conventional technique, for example, Japanese Patent Application Laid-Open No. H08-2118
No. 88 (third prior art). That is, the third prior art aims at improving an environment adaptation method for improving the recognition rate in speech recognition using a statistical model, and aims at improving environmental noise for speech recognition using a statistical model. In the environment adaptation method of creating a non-speech model, which is a statistical model corresponding to a non-speech section, learning environment noise data for an environment in which speech recognition is to be performed before the start of speech recognition, is used as the learning speech data. Superimposed on
The speech data for the noise superimposition learning is acquired, and a non-speech model is created from the speech data for the noise superimposition learning by learning, and the non-speech model is a transition from non-speech to speech, a stationary non-speech part, and a speech to non-speech. 3 corresponding to the transition to
This is an environment adaptation method in speech recognition expressing a state.
According to such an environment adaptation method in speech recognition, a non-speech model is created by learning from learning speech data after noise of a recognition environment is superimposed, so that humans such as lip noise and breath in addition to environmental noise are uttered. Noise can be learned, the consistency in non-sound sections is improved, the speech recognition rate is improved, and the non-speech model is a transition from non-speech to speech, a stationary non-speech part, If the three states of the transition from speech to non-speech can be expressed, in addition to the noise of the environment itself and noise uttered by humans, the transition from non-speech to speech and the non-speech to speech It is also possible to learn the transition to the voice, the recognition rate is further increased, and furthermore, the voice model created from the learning voice data and the voice model created from the noise superimposition learning voice data are used. If having a selection means for-option can selectively use speech model by the nature of the environmental noise, the higher the integrity of the voice segment, the effect is disclosed such as voice recognition rate is high.

[0004]

However, in the above-mentioned prior art speech recognition engine, environmental learning is performed simultaneously with or after the recognition process. Therefore, despite the fact that the speech recognition engine has an environment adaptation function, the speech recognition engine does not execute environmental learning until a human utters, so that the speech recognition engine There was a problem that the environment adaptation function could not be sufficiently exhibited for the first utterance because environmental information was not accumulated.

SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and has as its object to provide a speech input / output system and a speech input / output system capable of executing at least environment adaptation processing for speech recognition processing for human speech. The point is to provide an input / output method.

[0006]

An object of the present invention is to provide a speech input / output system capable of executing an environment adaptation process for at least a speech recognition process for a human voice, wherein the environment learning process is performed. A voice recognition engine that executes simultaneously with the voice recognition process or as a post-process of the voice recognition process, and a voice pattern prepared in advance when a human does not execute vocalization and an external input signal are superimposed. A means for inputting to a speech recognition engine and performing a recognition process in advance to learn environmental information; and a method in which when a human utters, the environment learning process is completed at least once and learned and held internally. A voice input / output system characterized by having means for executing a recognition process based on existing environmental learning information. The gist of the invention according to claim 2 is a voice input / output system capable of executing an environment adaptation process at least for a voice recognition process for a human voice, wherein the environment learning process is performed simultaneously with the voice recognition process or A voice recognition engine to be executed as post-processing of the voice recognition process, and a voice pattern prepared in advance when a human does not execute voice and a background sound from outside are superimposed and input to the voice recognition engine. Based on the means for learning the environment information by executing the recognition process in advance, and the environment learning information that is learned in advance and held internally when the human has uttered the environment learning process at least once. The speech input / output system has means for executing a recognition process. The gist of the invention according to claim 3 is that an audio input unit for converting an external audio into an audio signal and only one audio input channel for inputting the audio signal are provided except for a human voice included in the audio signal. A speech recognition engine having an environment adaptation function for improving a recognition rate by learning background noise of the above, a signal switching unit for switching a signal to be input to the one-input speech recognition engine, and a human prepared in advance. 3. The audio input / output system according to claim 1, further comprising a mixer section for superimposing a sample pattern of the utterance and an externally input audio signal. The gist of the invention described in claim 4 is that an audio input unit for converting an external audio into an audio signal and only one input terminal for the audio signal are provided, and background noise other than human voice included in the audio signal is eliminated. Said speech recognition engine having an environment adaptation function aimed at improving the recognition rate by learning,
A signal switching unit that switches a signal input to the one-input speech recognition engine; a sample speech pattern storage unit that stores a sample pattern of a human utterance prepared in advance; and a sample stored in the sample speech pattern storage unit. 3. The audio input / output system according to claim 1, further comprising a mixer section for superposing and outputting an audio pattern and an audio signal input from the audio input section. The gist of the invention according to claim 5 is that, after the start of voice input and when no human is uttering, an external input is made to the human voice sample pattern held in the sample voice pattern storage unit. By superimposing in the mixer unit and inputting to the one-input speech recognition engine and setting the signal switching unit to a predetermined side, the one-input speech recognition engine is:
5. The apparatus according to claim 4, wherein the apparatus performs recognition processing when no voice is input by a human utterance and performs one-input environment adaptation processing for learning environment adaptation. In the voice input / output system described in (1). The gist of the invention according to claim 6 is that, when voice input is started and a human utterance is made, the signal switching unit is set to another predetermined side, and the external voice input is directly input to the one input. The speech input / output system according to claim 4 or 5, characterized in that the speech recognition engine is configured to execute a one-input speech recognition process for performing a recognition process by inputting to the speech recognition engine. The gist of the invention described in claim 7 is that an audio input unit that converts an external sound into an audio signal, an environmental sound input unit that converts an external environmental sound into an environmental sound signal, and the audio signal is input. It is provided with two voice input channels for voice input and a noise input channel for environmental sound input in which the environmental sound signal is expected to be input. A two-input speech recognition engine having an environment adaptation function for improving a recognition rate by learning, a signal switching unit for switching a signal input to the two-input speech recognition engine, and the sample speech pattern storage unit And a mixer unit that superimposes and outputs the sample sound pattern stored in the unit and the environmental sound signal input from the environmental sound input unit. Or it consists in the audio input and output system according to 2. The gist of the invention described in claim 8 is that, after the start of voice input and when no human is uttering, the environmental sound signal output from the environmental sound input unit is held in the sample voice pattern storage unit. A signal generated by being superimposed on the sample voice pattern is input to the voice input channel of the two-input voice recognition engine, the signal switching unit is set to a predetermined side, and the signal input from the environmental sound input unit is set. By directly inputting the environmental sound signal to the noise input channel of the two-input speech recognition engine, the two-input speech recognition engine performs recognition processing when no human is uttering and learns environmental information. The speech input / output system according to claim 7, wherein the speech input / output system is configured to execute a two-input environment adaptation process. The gist of the invention according to claim 9 is that, when voice input is started and a human utterance is made, the signal switching unit is set to another predetermined side, and the audio signal from the audio input unit is set to the other side. The two-input voice recognition engine includes a voice actually uttered by a human by inputting to the voice input channel and inputting the environmental sound signal input from the environmental sound input unit to the noise input channel. The speech input / output system according to claim 7 or 8, wherein the speech input / output system is configured to execute a two-input speech recognition process for performing a recognition process for a speech signal as a recognition target. Claim 1
The gist of the invention described in claim 0 is a voice input / output method capable of executing an environment adaptation process for at least a voice recognition process for a human voice, wherein the environment learning process is performed simultaneously with the voice recognition process or the voice recognition process. A voice recognition step to be performed as a post-processing, and a voice pattern prepared in advance when a human does not execute a voice and an input signal from the outside are superimposed and input to the voice recognition step to perform a recognition process in advance. And learning the environmental information, and executing recognition processing based on environmental learning information that is learned in advance and stored internally when the environmental learning processing is completed at least once when a human utters. And a voice input / output method. Claim 11
The gist of the invention described in (1) is a voice input / output method capable of executing an environment adaptation process for at least a voice recognition process for a human voice, wherein the environment learning process is performed simultaneously with the voice recognition process or the voice recognition process. A voice recognition step to be executed as post-processing, and a voice pattern prepared in advance when a human does not execute utterance and an external background sound are superimposed and input to the voice recognition step to perform a recognition process in advance. And learning the environmental information, and executing recognition processing based on environmental learning information that is learned in advance and stored internally when the environmental learning processing is completed at least once when a human utters. And a voice input / output method. The gist of the invention according to claim 12 is that an audio input step of converting external audio into an audio signal, and that only one audio input channel for inputting the audio signal is provided, other than a human voice included in the audio signal A speech recognition step having an environment adaptation function for improving a recognition rate by learning background noise of the above, a signal switching step of switching a signal input to the one-input speech recognition step, and a human voice prepared in advance 12. The audio input / output method according to claim 10, further comprising: a mixing step of superimposing the sample pattern of (1) with an externally input audio signal. The gist of the invention according to claim 13 is that an audio input step of converting an external audio into an audio signal, and that only one input terminal for the audio signal is provided to remove background noise other than human voice included in the audio signal. The voice recognition step having an environment adaptation function for improving the recognition rate by learning; a signal switching step of switching a signal input to the one-input voice recognition step;
A sample voice pattern storage step of storing a sample pattern of a human utterance prepared in advance, and a sample voice pattern stored in the sample voice pattern storage step and a voice signal input from the voice input step superimposed and output 12. The audio input / output method according to claim 10, further comprising a mixing step. The gist of the invention according to claim 14 is that, after the start of voice input and when no human is uttering, an external input is applied to the human voice sample pattern held in the sample voice pattern storage step. By superimposing in the mixing step and inputting to the one-input speech recognition step, and setting the signal switching step to a predetermined side,
The one-input speech recognition step includes a step of performing a one-input environment adaptation process of performing an environment adaptation learning by performing a recognition process when a speech input by a human utterance is not actually performed. A voice input / output method according to claim 13. The gist of the invention according to claim 15 is that, when voice input is started and human utterance is performed, the signal switching step is set to another predetermined side,
The method according to claim 13 or 14, further comprising a step of executing a one-input speech recognition process in which a speech input from the outside is directly input to the one-input speech recognition process to perform a recognition process.
In the voice input / output method described in (1). The gist of the invention described in claim 16 is that an audio input step of converting an external audio into an audio signal, an environmental sound input step of converting an external environmental sound into an environmental sound signal, and the audio signal is input. It is provided with two voice input channels for voice input and a noise input channel for environmental sound input in which the environmental sound signal is expected to be input. The two-input speech recognition step having an environment adaptation function for improving the recognition rate by learning, a signal switching step of switching a signal input to the two-input speech recognition step, and a sample speech pattern storage step A mixing step of superimposing and outputting the stored sample sound pattern and the environmental sound signal input from the environmental sound input step. It consists in the audio input and output method according to Motomeko 10 or 11. The gist of the invention according to claim 17 is that, after the start of voice input and when no human is speaking, the environmental sound signal output from the environmental sound input step is stored in the sample voice pattern storage step. A signal generated by being superimposed on the sample voice pattern is input to the voice input channel of the two-input voice recognition step, the signal switching step is set to a predetermined side, and the environment input from the environmental sound input step is set. By directly inputting a sound signal to the noise input channel in the two-input speech recognition step, the two-input speech recognition step performs a recognition process when no human is uttering to learn environmental information. The voice input / output method according to claim 16, further comprising a step of executing an environment adaptation process. Further, the gist of the invention according to claim 18 is that, when voice input is started and human utterance is performed, the signal switching step is set to another predetermined side, and the voice signal from the voice input step is set The two-input speech recognition step is performed by inputting a sound input channel to the noise input channel and inputting the environmental sound signal input from the environmental sound input step to the noise input channel. The speech input / output method according to claim 16 or 17, further comprising a step of performing a two-input speech recognition process for performing a recognition process on a signal as a recognition target.

[0007]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The features of each embodiment described below are characterized in that a voice pattern prepared in advance when a human does not execute a voice and an external input (only a background voice without a human voice is included). Input) to the speech recognition engine to perform recognition processing in advance and learn environment information, so that when a human utters, the environment learning processing is completed at least once. The object of the present invention is to enable a recognition process to be executed effectively based on environmental learning information that has been learned in advance and is internally stored. Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

(First Embodiment) FIG. 1 shows a first embodiment of the present invention.
1 is a block diagram for explaining a voice input / output system 10 according to an embodiment. In FIG. 1, 1 is a one-input speech recognition engine, 2 is a signal switching unit, 4 is a mixer unit, 5 is a speech input unit, and 10 is a speech input / output system. Referring to FIG. 1, a voice input / output system 10
Is a speech input / output system capable of executing an environment adaptation process for at least a speech recognition process for a human speech, which has one external speech input and learns background noise other than a human voice included in a speech signal. Has an environment adaptation function for the purpose of improving the recognition rate by performing the environment learning process for at least the speech recognition process for human voice at the same time as the speech recognition process or as a post-process of the speech recognition process. 1-input speech recognition engine 1 to be executed
A signal switching unit 2 for switching a signal to be input to the one-input speech recognition engine 1; a mixer unit 4 for mixing (overlapping) a sample pattern of a human utterance prepared in advance with an externally input audio signal. ,
An audio input unit 5 for converting external audio into an audio signal is provided.

FIG. 3 is a specific configuration diagram of the voice input / output system 10 of FIG. 3, reference numeral 1 denotes a one-input voice recognition engine, 2 denotes a signal switching unit, 3 denotes a sample voice pattern storage unit, 4 denotes a mixer unit, 5 denotes a voice input unit (microphone), and 10 denotes a voice input / output system. I have. FIG.
Referring to FIG. 1, the voice input / output system 10 has one external voice input and executes the environmental learning process simultaneously with the voice recognition process or as a post-process of the voice recognition process.
An input speech recognition engine 1, a signal switching unit 2 for switching a signal input to the one-input speech recognition engine 1,
A sample voice pattern storage unit 3 that stores a sample pattern of a human utterance prepared in advance, and a sample voice pattern stored in the sample voice pattern storage unit 3 and a voice signal input from the voice input unit 5 are mixed. A mixer section 4 for outputting (overlapping) and an audio input section 5 for converting external audio into an audio signal and outputting the audio signal are provided.

Next, the operation (speech input / output method) of the speech input / output system 10 using the one-input speech recognition engine 1 will be described with reference to FIGS. FIG. 2 is a flowchart for explaining a voice input / output method (operation of the voice input / output system 10 of the first embodiment) according to the first embodiment of the present invention. Referring to FIGS. 2 and 3, after the start of voice input (step S1), if no human is uttering (no utterance in step S2), a sample of human voice held in the sample voice pattern storage unit 3 is obtained. The input from the outside is superimposed on the pattern by the mixer unit 4 (step S21), input to the one-input speech recognition engine 1, and the signal switching unit 2 is set to the A side (see FIG. 3) (step S3). , One-input speech recognition engine 1
However, learning of environment adaptation is performed by performing a recognition process (step S4) in the case where a voice input by a human utterance is not actually performed (1 input environment adaptation process).

On the other hand, if a human utterance is made (the utterance is made in step S2), the signal switching unit 2 is set to the B side (see FIG. 3) (step S22), and one external voice input is directly input. And performs recognition processing (step S4) (Speech recognition processing at one input).

As described above, according to the first embodiment, in the case of recognizing a human utterance, the environment recognition information is accumulated by executing the one-input speech recognition processing in advance and accumulating the learning information of the environment. Since the processing is performed one or more times, the recognition processing can be effectively performed. On the other hand, if the one-input speech recognition process is not performed in advance, the environmental learning process is not performed because the environment information is not stored in the one-input speech recognition engine 1. It is thought that effective recognition processing by the environment adaptation function cannot be expected.

(Second Embodiment) FIG. 4 shows a second embodiment of the present invention.
1 is a block diagram for explaining a voice input / output system 10 according to an embodiment. In FIG. 4, 2 is a signal switching unit, 3 is a sample voice pattern storage unit, 4 is a mixer unit, 5 is a voice input unit, 6 is an environmental sound input unit, 8 is a 2-input voice recognition engine, and 10 is a voice input / output. Shows the system. Referring to FIG. 4, a voice input / output system 10 according to the present embodiment is a voice input / output system capable of executing an environment adaptation process at least for a voice recognition process for a human voice, and inputs two external voices. And has an environment adaptation function for improving a recognition rate by learning background noise other than a human voice included in a voice signal, and performing the environment learning process simultaneously with the voice recognition process or the voice recognition process. A two-input speech recognition engine 8 to be executed as post-processing, a signal switching unit 2 for switching a signal input to the two-input speech recognition engine 8, and a sample speech pattern for storing a sample pattern of a human utterance prepared in advance The storage unit 3, the sample voice pattern stored in the sample voice pattern storage unit 3 and the environmental sound input from the environmental sound input unit 6 Mixer 4 for mixing (superimposing) signals with each other and outputting, an audio input unit 5 (microphone) for converting external sound into an audio signal, and an environmental sound input for converting external environmental sound into an environmental sound signal. A unit 6 (microphone) is provided. The two-input speech recognition engine 8 of the present embodiment is a speech recognition engine that executes the environmental learning process simultaneously with the speech recognition process or as a post-process of the speech recognition process. Speech recognition engine provided with a voice input (referred to as a voice input channel C) that expects to be input and an environmental sound input (referred to as a noise input channel D) that expects a surrounding (background) noise to be input. The basic functions are almost the same as those of the one-input speech recognition engine 1.

Next, the operation (speech input / output method) of the speech input / output system 10 using the two-input speech recognition engine 8 will be described with reference to FIGS. FIG. 2 is a flowchart for explaining a voice input / output method (operation of the voice input / output system 10 of the second embodiment) according to the second embodiment of the present invention. Referring to FIGS. 2 and 4, after the start of voice input (step S1), if no human is uttering (no utterance in step S2), the environmental sound input unit 6
A noise signal (environmental sound signal) of the background output from is superimposed on the sample voice pattern held in the sample voice pattern storage unit 3 (step S21), and the generated signal is input to the two-input voice recognition engine 8 Channel C
The signal switching unit 2 is set to the A side (see FIG. 4) (step S3), and the environmental sound signal input from the environmental sound input unit 6 is input to the noise input channel D of the two-input speech recognition engine 8. By directly inputting, the two-input speech recognition engine 8 performs recognition processing when no human is uttering (step S4) and learns environment information (that is, performs environment adaptation processing) (two-input speech recognition engine 8). Time environment adaptation processing).

On the other hand, when a human utterance is made (with utterance in step S2), the signal switching section 2 is set to the B side (see FIG. 4) (step S22), and the audio signal from the audio input section 5 is output. By inputting the sound signal to the sound input channel C and the environmental sound signal input from the environmental sound input unit 6 to the noise input channel D, the two-input sound recognition engine 8
Recognition processing (step S4) is performed with a speech signal including a voice actually uttered by a human being as a recognition target (2-input speech recognition processing).

As described above, according to the second embodiment, when recognizing a human utterance, a two-input speech recognition process is executed in advance to accumulate learning information of the environment, thereby enabling environmental learning. Since the processing is performed one or more times, the recognition processing can be effectively performed. On the other hand, if the speech recognition processing at the time of two inputs is not executed in advance, the environment learning processing is not executed because the environment information is not accumulated inside the two-input speech recognition engine 8, so that the speech input by human utterance is recognized. It is thought that effective recognition processing by the environment adaptation function cannot be expected.

Finally, technical differences and effects of the present embodiment and the above-mentioned conventional technology will be described. The third technique described in the above-mentioned Japanese Patent Application Laid-Open No. H8-212888.
In the prior art, since the path for inputting pseudo data obtained by superimposing the learning voice and the environmental noise to the environmental learning means is completely different from the path for performing the normal recognition processing, the environment learning is performed during the normal recognition processing. It is considered difficult.

On the other hand, the present invention has the above-mentioned environment adaptation function for speech recognition processing for at least human speech, and executes the environment learning processing simultaneously with the speech recognition processing or as a post-processing of the speech recognition processing. A speech recognition system that includes a speech recognition engine that performs a speech recognition process on pseudo-prepared speech data instead of speech data of a true speaker, so that an environment learning process can be performed in advance. By providing a configuration in which voice data for learning and utterance data of a speaker are switched on the same route, the environment learning process can be always performed together with the recognition process, and the system can be simplified. Therefore, it is considered that the configuration and the effect are different from those of the related art.

As described above, according to each of the above embodiments, when recognizing a human utterance, the one-input speech recognition process or the two-input speech recognition process is executed in advance to obtain the learning information of the environment. By accumulating the information, the environmental learning process is executed one or more times, so that the recognition process can be effectively performed. On the other hand, if the one-input speech recognition process or the two-input speech recognition process is not performed in advance, the environment learning process is not executed by not storing the environmental information inside the one-input speech recognition engine 1, so that the human voice is used. It is thought that effective recognition processing by the environment adaptation function cannot be expected when recognizing speech input.

It should be noted that the present invention is not limited to the above embodiments, and each embodiment can be appropriately modified within the scope of the technical idea of the present invention. Further, the number, position, shape, and the like of the constituent members are not limited to the above-described embodiment, and can be set to numbers, positions, shapes, and the like suitable for carrying out the present invention. In each drawing, the same components are denoted by the same reference numerals.

[0021]

Since the present invention is constructed as described above, according to the above-described embodiment, when recognizing a human utterance, one-input speech recognition processing or two-input speech recognition processing is performed. Since the environment learning process is executed one or more times by executing the voice recognition process in advance and accumulating the learning information of the environment, there is an effect that the recognition process can be effectively performed. On the other hand, if the one-input speech recognition process or the two-input speech recognition process is not performed in advance, the environment learning process is not performed because the environment information is not stored in the speech recognition engine. It is thought that effective recognition processing by the environment adaptation function cannot be expected during processing.

[Brief description of the drawings]

FIG. 1 is a block diagram for explaining a voice input / output system according to a first embodiment of the present invention.

FIG. 2 is a flowchart illustrating a voice input / output method (operation of a voice input / output system) according to the first and second embodiments of the present invention.

FIG. 3 is a specific configuration diagram of the audio input / output system of FIG. 1;

FIG. 4 is a block diagram for explaining a voice input / output system according to a second embodiment of the present invention.

FIG. 5 is a block diagram illustrating a configuration of a conventional system.

FIG. 6 is a block diagram illustrating a configuration of a conventional system.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... 1-input speech recognition engine 2 ... Signal switching unit 3 ... Sample speech pattern storage unit 4 ... Mixer unit 5 ... Speech input unit (microphone) 6 ... Environmental sound input unit (microphone) 8 ... 2-input speech recognition engine 10 ... Sound input / output system C ... Sound input channel D ... Noise input channel

Claims (18)

[Claims] 1. A speech input / output system capable of executing an environment adaptation process at least for a speech recognition process for a human speech, wherein the environment learning process is performed simultaneously with the speech recognition process or after the speech recognition process. A speech recognition engine to be executed as, and a speech pattern prepared in advance when a human does not execute vocalization and an external input signal are superimposed and input to the speech recognition engine to execute a recognition process in advance. A means for learning environmental information; and a means for executing recognition processing based on environmental learning information that has been learned in advance and held internally when the environmental learning processing has been completed at least once when a human utters. A voice input / output system characterized by the following. 2. A speech input / output system capable of executing an environment adaptation process for at least a speech recognition process for a human speech, wherein the environment learning process is performed simultaneously with the speech recognition process or after the speech recognition process. A speech recognition engine to be executed as, and a speech pattern prepared in advance when a human does not execute utterance and an external background sound are superimposed and input to the speech recognition engine to execute a recognition process in advance. Means for learning the environmental information by means of the method, and means for executing the recognition processing based on the environmental learning information which is learned in advance and stored therein when the human utters the environment learning processing at least once. A voice input / output system characterized by having. 3. An audio input unit for converting external audio into an audio signal, and only one audio input channel for inputting the audio signal is provided, and background noise other than human voice included in the audio signal is learned. The speech recognition engine having an environment adaptation function for improving the recognition rate by the above, a signal switching unit for switching a signal to be inputted to the speech recognition engine of one input, a sample pattern of a human utterance prepared in advance and an external 3. The audio input / output system according to claim 1, further comprising a mixer unit that superimposes an audio signal input from the audio input device. 4. An audio input unit for converting an external audio into an audio signal, and having only one input terminal for the audio signal, learning a background noise other than a human voice included in the audio signal, thereby reducing a recognition rate. A speech recognition engine having an environment adaptation function for the purpose of improvement, a signal switching unit for switching a signal inputted to the speech recognition engine of one input, and a sample speech pattern for storing a sample pattern of a human utterance prepared in advance 3. The storage device according to claim 1, further comprising: a storage unit; and a mixer unit configured to superimpose and output the sample voice pattern stored in the sample voice pattern storage unit and the voice signal input from the voice input unit. The voice input / output system according to 1. 5. After the start of voice input and when no human is uttering, the mixer unit superimposes an external input on the sample pattern of the human voice stored in the sample voice pattern storage unit. By inputting to the one-input speech recognition engine and setting the signal switching unit to a predetermined side, the one-input speech recognition engine does not actually perform a speech input by human utterance. 5. The speech input / output system according to claim 4, wherein the one-input environment adaptation process for performing the environment adaptation learning by performing the recognition process is performed. 6. When a human utterance is made after the start of voice input, the signal switching unit is set to another predetermined side, and an external voice input is directly input to the one-input voice recognition engine. The voice input / output system according to claim 4, wherein the voice input / output system is configured to execute a one-input voice recognition process for performing a recognition process. 7. An audio input unit for converting an external sound into an audio signal, an environmental sound input unit for converting an external environmental sound into an environmental sound signal, and an audio input unit expecting the audio signal to be input. And two noise input channels for inputting environmental sound, which are expected to receive the environmental sound signal, and learning background noise other than the human voice included in the audio signal to reduce the recognition rate. A two-input speech recognition engine having an environment adaptation function for the purpose of improvement, a signal switching unit for switching a signal input to the two-input speech recognition engine, and a sample speech stored in the sample speech pattern storage unit 3. The sound according to claim 1, further comprising a mixer that superimposes and outputs a pattern and the environmental sound signal input from the environmental sound input unit. 4. Voice input / output system. 8. After the start of voice input and when no human is uttering, the environmental sound signal output from the environmental sound input unit is superimposed on the sample voice pattern held in the sample voice pattern storage unit. The generated signal is input to the voice input channel of the two-input voice recognition engine, the signal switching unit is set to a predetermined side, and the environmental sound signal input from the environmental sound input unit is input to the 2nd voice recognition engine.
By directly inputting to the noise input channel of the input speech recognition engine, the two-input speech recognition engine
The speech input / output system according to claim 7, wherein the speech input / output system is configured to execute a two-input environment adaptation process of learning environment information by performing a recognition process when no human is speaking. 9. When a human utterance is performed after the start of voice input, the signal switching unit is set to another predetermined side, and a voice signal from the voice input unit is input to the voice input channel. By inputting the environmental sound signal input from the environmental sound input unit to the noise input channel, the two-input voice recognition engine recognizes a voice signal including a voice actually uttered by a human. 9. The voice input / output system according to claim 7, wherein the voice input / output system is configured to execute a two-input speech recognition process for performing a recognition process. 10. A speech input / output method capable of executing environment adaptation processing at least for speech recognition processing for human speech, wherein said environment learning processing is performed simultaneously with said speech recognition processing or post-processing of said speech recognition processing. A voice recognition step to be executed as, and a voice pattern prepared in advance when a human does not execute vocalization and an external input signal are overlapped and input to the voice recognition step to perform a recognition process in advance. A step of learning environmental information, and a step of executing a recognition process based on environmental learning information that is learned in advance and held internally in a state where the environmental learning process has been completed at least once when a human utters. A voice input / output method characterized by the above-mentioned. 11. A speech input / output method capable of executing an environment adaptation process for at least a speech recognition process for a human speech, wherein the environment learning process is performed simultaneously with the speech recognition process or after the speech recognition process. A voice recognition step to be executed as, and a voice pattern prepared in advance when a human does not execute vocalization and an external background sound are overlapped and input to the voice recognition step to execute a recognition process in advance. Learning the environmental information, and executing the recognition process based on the environmental learning information that is learned in advance and stored internally when the human utters the environmental learning process at least once. A voice input / output method comprising: 12. A voice input step of converting external voice into a voice signal, and comprising only one voice input channel for voice signal input and learning background noise other than human voice included in the voice signal. The speech recognition step having an environment adaptation function for improving the recognition rate by the following: a signal switching step of switching a signal to be input to the one-input speech recognition step; a sample pattern of a human utterance prepared in advance and input from outside 12. The audio input / output method according to claim 10, further comprising a mixing step of superimposing the audio signal to be performed. 13. A speech inputting step of converting an external speech into a speech signal; and providing a speech signal input terminal only and learning a background noise other than a human voice included in the speech signal to reduce a recognition rate. The voice recognition step having an environment adaptation function for the purpose of improvement, a signal switching step of switching a signal input to the one-input voice recognition step, and a sample voice pattern storage step of storing a sample pattern of a human voice prepared in advance 12. A mixing step of superimposing and outputting a sample voice pattern stored in the sample voice pattern storage step and a voice signal input from the voice input step. Voice input / output method. 14. After the start of voice input and when no human is speaking, an external input is superimposed on the sample pattern of the human voice held in the sample voice pattern storage step in the mixing step. By inputting the signal to the one-input speech recognition step and setting the signal switching step to a predetermined side, the one-input speech recognition step performs a recognition process in a case where a human voice is not actually input. 14. The voice input / output method according to claim 13, further comprising a step of performing a one-input environment adaptation process for performing environment adaptation learning. 15. When a human utterance is made after the start of voice input, the signal switching step is set to another predetermined side, and an external voice input is directly input to the one-input voice recognition step. 14. The method according to claim 13, further comprising a step of performing a one-input speech recognition process for performing a recognition process.
Or the voice input / output method according to 14. 16. An audio input step of converting an external sound into an audio signal, an environmental sound input step of converting an external environmental sound into an environmental sound signal, and a sound input for expecting the audio signal to be input. And two noise input channels for inputting environmental sound, which are expected to receive the environmental sound signal, and learning background noise other than the human voice included in the audio signal to reduce the recognition rate. A two-input speech recognition step having an environment adaptation function for the purpose of improvement, a signal switching step of switching a signal to be input to the two-input speech recognition step, and a sample speech pattern stored in the sample speech pattern storage step 11. A mixing step of superimposing and outputting the environmental sound signal input from the environmental sound input step and the environmental sound signal input from the environmental sound input step. Voice input and output method according to. 17. After the start of voice input and when no human is speaking, the environmental sound signal output from the environmental sound input step is superimposed on the sample voice pattern stored in the sample voice pattern storage step. Inputting the signal generated together to the voice input channel of the two-input voice recognition step, setting the signal switching step to a predetermined side,
By directly inputting the environmental sound signal input from the environmental sound input step to the noise input channel of the two-input voice recognition step, the two-input voice recognition step performs a recognition process when no human is speaking. 17. The method according to claim 16, further comprising the step of: performing a two-input environment adaptation process of learning environment information by performing the following. 18. When a human utterance is made after the start of voice input, the signal switching step is set to another predetermined side, and a voice signal from the voice input step is input to the voice input channel; By inputting the environmental sound signal input from the environmental sound input step to the noise input channel, the two-input voice recognition step recognizes a voice signal including a voice actually uttered by a human as a recognition target. 18. The voice input / output method according to claim 16, further comprising a step of performing a two-input speech recognition process for performing a process.