JP3916861B2 - Voice recognition device - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a speech recognition device that specifies a character string corresponding to input speech and returns a response according to the content.
[0002]
[Prior art]
Conventionally, a voice recognition device for inputting various operation instructions and the like by voice has been put into practical use and has been adopted in various devices and systems. For example, an in-vehicle navigation device equipped with a voice recognition device can input an operation instruction such as setting a destination in route search by voice. In addition, by executing a predetermined program on a personal computer (hereinafter referred to as a “personal computer”), a speech recognition device is realized on the personal computer, and a sentence input or the like corresponding to the sound collected by the microphone is performed. Some are operating.
[0003]
By the way, generally speaking, speech recognition technology can be classified into word speech recognition technology and continuous word speech recognition technology. The former word speech recognition technology is a technology for recognizing speech uttered by separating words and specifying a character string of a corresponding word. The latter continuous word speech recognition technology is a technology for recognizing speech in which a plurality of words or the like are continuously spoken and specifying character strings of a plurality of corresponding words.
[0004]
Conventionally, speech recognition devices that employ relatively easy-to-process word speech recognition technology have been the mainstream, but from the user's perspective, it is easier to input multiple words continuously. In recent years, speech recognition apparatuses that employ continuous word speech recognition technology are becoming popular. When a speech recognition device employing such continuous word speech recognition technology is installed in a navigation device, for example, “XX prefecture XX city ΔΔ... You can input and recognize words such as prefecture names, city names, and place names in succession, so you can perform input operations more comfortably than if you enter words such as prefecture names one by one. Can do.
[0005]
[Problems to be solved by the invention]
By the way, in the speech recognition apparatus adopting the continuous word speech recognition technology described above, a state that can be regarded as almost silent when capturing the speech uttered by the user (hereinafter, this silent state is referred to as “blank”) is preset. When the predetermined time is exceeded, speech recognition processing is performed on the speech input up to that point, and the recognition result is returned to the user.
[0006]
For example, when it is detected that a blank of time T1 or more is included in the input speech, a speech synthesis process corresponding to this speech is performed and a predetermined response is output. The time required from the input of (voice excluding blank) to the output of the corresponding response is a predetermined time T obtained by adding the time T1 corresponding to the blank and the time required for the speech synthesis process. . Therefore, from the user's standpoint, this predetermined time T is an apparent blank allowed at the time of voice input, and when voice input is performed without including only a blank shorter than the predetermined time T, continuous It is normal to think that speech recognition processing is performed as a word.
[0007]
However, in the conventional speech recognition apparatus, since the speech recognition process is terminated and the response process is started when a blank of a time T1 shorter than the predetermined time T is detected, a speech is input after the elapse of this time T1. However, there is a problem that so-called “missing” occurs. In general, a word that is not used to speaking usually cannot be fluently spoken, and blanks are often included between the words, so the above-mentioned miss is likely to occur.
[0008]
For example, considering a case where an address or the like that is not commonly used is input to a navigation device equipped with a voice recognition device, the user himself / herself may enter an address such as “XX prefecture XX city ...”. In spite of intending to input continuously, in fact, blanks are inserted between each word such as between “XX prefecture” and “XX city”. The recognition process is interrupted until “XX prefecture” and a corresponding response is made, and a part of “XX city ...” uttered thereafter may be missed. In addition, when the above-mentioned loss occurs, for example, a response such as “XX prefecture. Please enter the municipality name.” Will be given. Since the voice after a certain municipality name is ignored and input is requested again, the user often feels uncomfortable with such a response.
[0009]
The present invention has been made in view of such a point, and an object of the present invention is to provide a voice recognition device that can eliminate missing voices that are input before a response is returned. Another object of the present invention is to provide a speech recognition apparatus that can return a response without a sense of incongruity.
[0010]
[Means for Solving the Problems]
In order to solve the above-described problems, in the speech recognition apparatus according to the present invention, speech is collected by a microphone, speech recognition processing is performed on the collected speech by speech recognition processing means, and based on the recognized contents. When the response means generates and outputs a response sound, the interruption determination means detects a silence state included in the sound collected by the microphone, and the speech recognition is performed when the silence state continues for a time t1 or more. Determine interruption of processing. The sound pressure level detection means detects the sound pressure level of the sound collected by the microphone, and the sound pressure level detection means detects the sound pressure level during the time t2 after the time t1 has elapsed after the silent state described above. When the detected sound pressure level exceeds a predetermined value, the restart determination means determines to restart the speech recognition process.
[0011]
When a sound exceeding a predetermined sound pressure level is input until a predetermined time t2 elapses after the silent state included in the sound has elapsed after the time t1 and the predetermined response processing has started. Since the processing by the voice recognition processing means is resumed, it is possible to eliminate the missed voice that has been input before the response is returned.
[0012]
In addition, it is desirable that the above-described restart determination unit determines the restart of the process only once for the speech recognition processing unit in response to the first silent state that is included in the input speech for the time t1 or longer. In general, when the first silent state is detected and a response is returned, it is unlikely that the user will continue to input voice in parallel with this response, so only one time in response to the first silent state By simply restarting the processing of the speech recognition processing means, it is possible to eliminate almost all of the speech that has been input before the response is returned.
[0013]
In addition, it is desirable that the above-described restart determination unit sends an instruction to stop the output of the response voice to the response unit together with an operation to instruct the voice recognition processing unit to restart the process. By stopping the output to the response voice when the processing of the voice recognition processing means is resumed, it is possible to prevent the voice uttered by the user and the response voice from overlapping. In particular, since the voice recognition process is resumed without returning a response voice, the voice recognition process for continuous words can be continued until the voice uttered by the user includes a time (t1 + t2) blank, which is efficient. Voice input can be performed.
[0014]
Further, the above-described time t2 is a value substantially equal to the time from when the speech recognition process is interrupted by the interruption determination means until the response speech is output by the response means after the duration of the silent state is the time t1. It is desirable to set to. By setting the time from when the duration of the silent state reaches time t1 until the response voice by the response means is output and the above-described time t2 is substantially equal, the voice input is performed before the response voice is output. If performed, the voice recognition process can be continued in a sure manner corresponding to the voice input. Therefore, although the user himself intends to continuously input the voice, the voice recognition process is not started and the response voice is not output in response to the voice being input. A person can feel uncomfortable.
[0015]
In addition, it further comprises: a speaker that outputs response sound and sound output from other sound sources; and a removing means that removes a component that is not subject to speech recognition included in the sound collected by the microphone. It is desirable to input the speech recognition target speech output from the speech recognition processing means. Since the accuracy of speech recognition processing can be improved by removing components that are not subject to speech recognition, the speech that is not subject to speech recognition, such as when the speech recognition device of the present invention is mounted on an in-vehicle navigation device, etc. Is particularly effective in an environment in which sound is easily included in the sound collected by the microphone.
[0016]
Further, it is desirable that the response means described above generates response voices having different contents depending on the success or failure of the voice recognition processing by the voice recognition processing means resumed after detecting a silence state of time t1 or longer. Specifically, for example, when the voice recognition process after the restart is successful, a response voice based on the recognition result is output, and when the voice recognition process fails, “the presence of the input voice by the user is recognized. The voice input that you made is ignored by changing the content of the response voice according to the success or failure of the voice recognition process, such as outputting a response voice that contains the content `` but the voice recognition process failed '' It is possible to eliminate the unpleasant feeling that the user feels with respect to the response voice without giving the user a bad impression of being interrupted or being interrupted.
[0017]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, a speech recognition apparatus according to an embodiment to which the present invention is applied will be described with reference to the drawings.
FIG. 1 is a diagram illustrating the configuration of the speech recognition apparatus according to the present embodiment. The voice recognition apparatus 100 shown in the figure is used to give an operation instruction by voice to the in-vehicle navigation apparatus 300, and includes a talk switch 10, a microphone 12, a control unit 14, a delay element 16, an adaptive filter. (ADF) 17, calculation unit 18, speech recognition processing unit 20, level meter 30, speech synthesis processing unit 32, synthesis unit 34, and speaker 36 are configured. It is assumed that the speech recognition apparatus according to the present embodiment employs continuous word speech recognition technology.
[0018]
The talk switch 10 is operated before the user performs voice input, and the operation status is output to the control unit 14. The microphone 12 collects voice uttered by the user, converts it into an electric signal (voice signal), and outputs it.
The control unit 14 controls the overall operation of the voice recognition device 100 and outputs information such as a character string obtained as a result of the voice recognition processing to the navigation device 300. Details of the operation of the control unit 14 will be described later.
[0019]
The delay element 16 outputs a signal obtained by delaying the audio signal output from the microphone 12 by a predetermined time. The delay element 16 has, for example, a transfer characteristic Z ^-m This is realized by setting a filter coefficient corresponding to the delay time t to 1 and setting other filter coefficients to 0 using an FIR (Finite Impulse Response) type digital filter.
[0020]
The adaptive filter 17 is for simulating the transfer characteristic of the acoustic space in the vehicle interior, specifically, the transfer characteristic until the sound radiated from the speaker 36 reaches the microphone 12, and the filter coefficient W And a filter coefficient setting unit for setting the filter coefficient of the digital filter. For example, using a LMS (Least Mean Square) algorithm, a filter coefficient W is determined by performing an adaptive equalization process using an audio signal (described later) input to the speaker 36 as a reference signal, and the filter coefficient W is determined as an output signal of the microphone 12. Processing for removing the output sound component of the included speaker 36 is performed by the calculation unit 18.
[0021]
In this manner, in this embodiment, the output sound component of the speaker 36 is removed from the sound signal output from the microphone 12, so that the response sound during the sound recognition process, the audio sound output from the audio device 200, and the like Even when the voice overlaps with the voice input by the user, only the voice of the user can be reliably extracted, and the recognition rate of the voice recognition process can be improved.
[0022]
The speech recognition processing unit 20 performs a predetermined speech recognition process for specifying a character string corresponding to the input speech, and includes two ring buffers 22 and 24, a feature amount extraction unit 26, and a matching processing unit 28. It is configured to include.
The ring buffer 22 takes in and stores the audio signal after removal of noise components (audio sound, response sound, etc.) output from the calculation unit 18 in the order of input. The stored audio signals are read out in the order of storage and input to the feature amount extraction unit 26.
[0023]
The feature quantity extraction unit 26 extracts various voice feature quantities necessary for performing the voice recognition process. The voice feature amount extracted by the feature amount extraction unit 26 is output directly to the collation processing unit 28 or stored in the ring buffer 24 in accordance with an instruction from the control unit 14.
[0024]
The ring buffer 24 stores the audio feature amounts output from the feature amount extraction unit 26 in the order of input, and when a read request is given from the collation processing unit 28, the stored audio feature amounts are read in the storage order. .
The collation processing unit 28 prepares a standard pattern in units of phonemes or words in advance, and corresponds to the input voice by collating the voice feature amount extracted by the feature amount extraction unit 26 with this standard pattern. The character string to be specified is specified and output to the control unit 14.
[0025]
The level meter 30 measures the sound pressure level of the voice based on the voice feature amount output from the feature amount extraction unit 26 and outputs the measurement result to the control unit 14.
The voice synthesis processing unit 32 generates and outputs a voice signal for outputting a response voice corresponding to the recognition result output from the matching processing unit 28 in accordance with an instruction from the control unit 14.
[0026]
The synthesizing unit 34 synthesizes the audio signal output from the audio synthesis processing unit 32 and the audio sound signal output from the audio device 200 and outputs the synthesized audio signal to the speaker 36. The speaker 36 outputs response sound and audio sound in response to the output signal from the synthesis unit 34.
[0027]
The speech recognition processing unit 20 described above is the speech recognition processing unit, the speech synthesis processing unit 32, the speaker 36 is the response unit, the collation processing unit 28 is the interruption determination unit, the control unit 14 is the restart determination unit, and the level meter 30 is Each corresponds to the sound pressure level detection means. Further, the delay element 16, the adaptive filter 17, and the calculation unit 18 correspond to a removing unit.
[0028]
The voice recognition apparatus of this embodiment has such a configuration, and the operation thereof will be described next.
[First operation procedure]
FIG. 2 is a flowchart showing a first operation procedure in the speech recognition apparatus 100. In the following description, it is assumed that a destination or the like is set in the navigation apparatus 300. As an operation instruction, a voice “XX prefecture XX city Δ △ ……”, that is, “prefecture name” and “city / town / village” In the following description, it is assumed that speech recognition processing is performed on a continuous word speech composed of a plurality of words following “name”, “place name”,.
[0029]
The control unit 14 determines whether or not the talk switch 10 has been pressed (step 100), and outputs an activation instruction to the voice recognition processing unit 20 when the talk switch 10 is pressed.
When the user performs voice input to the microphone 12 after the voice recognition processing unit 20 is activated (step 101), a predetermined voice recognition process is performed by the voice recognition processing unit 20 in response to the voice input. (Step 102). Specifically, based on the audio signal stored in the ring buffer 22, the audio feature amount is extracted by the feature amount extraction unit 26, and the collation processing unit 28 performs collation processing between the audio feature amount and the standard pattern. Thus, character strings (words) corresponding to the input voice are sequentially identified.
[0030]
Next, the collation processing unit 28 in the speech recognition processing unit 20 determines whether or not a blank (silenced state) of time t1 or more is included in the input speech (step 103). If a blank longer than time t1 is not included, a negative determination is made in step 103, the process returns to step 102, and a predetermined voice recognition process is continued.
[0031]
If the input voice includes a blank for time t1 or more, an affirmative determination is made in step 103, and the collation processing unit 28 determines to interrupt the voice recognition processing by the voice recognition processing unit 20. The recognition result for the voice up to the time of blank detection is output to the control unit 14.
[0032]
The control unit 14 outputs the response voice corresponding to the voice up to the time of blank detection by outputting the recognition result received from the collation processing unit 28 to the voice synthesis processing unit 32 (step 104).
In parallel with the processing shown in step 104, the control unit 14 checks whether or not the output signal from the level meter 30 has exceeded a predetermined value, so that it is within time t2 from the time of blank detection at time t1 or more. It is determined whether or not voice input has been performed (step 105). In the following description, the voice input performed within the time t2 after the blank detection at the time t1 or more is referred to as “additional voice input”.
[0033]
If an additional voice input has been performed, an affirmative determination is made in step 105, and the control unit 14 determines to restart the voice recognition process, and outputs an activation instruction again to the voice recognition processing unit 20. In accordance with this activation instruction, the predetermined voice recognition processing by the voice recognition processing unit 20 is resumed (step 106), and the voice shown in step 106 is included until the input voice includes a second blank of time t1 or more (step 107). The recognition process continues.
[0034]
If the input voice includes a second blank of time t1 or more, an affirmative determination is made in step 107, and the collation processing unit 28 outputs the recognition result for the voice up to the time of blank detection to the control unit 14. . Based on the recognition result output from the matching processing unit 28, the control unit 14 determines whether or not the additional voice input has been normally recognized (step 108). Specifically, if any character string (word) can be specified in response to the additional voice input, the character string and the character string corresponding to the additional voice input could not be specified. In this case, since an error notification indicating that the recognition could not be normally performed is output from the collation processing unit 28, the control unit 14 determines whether or not there is an error notification from the collation processing unit 28. It is determined whether or not the additional voice input has been successfully recognized.
[0035]
If the additional speech input can be recognized, an affirmative determination is made in step 108, and the control unit 14 outputs the recognition result character string received from the collation processing unit 28 to the speech synthesis processing unit 32. Thus, a response voice corresponding to the additional voice input is output (step 109).
[0036]
If the additional voice input cannot be recognized, a negative determination is made in step 108, and the control unit 14 sends an instruction to the voice synthesis processing unit 32 to recognize that there is an additional voice input. A response voice including the message is output (step 110).
[0037]
Specifically, for example, as described above, the user tried to input “XX prefecture XX city ……”, but the response voice corresponding to “XX prefecture” is processed by the processing shown in step 104. If the input of “XX city ……” is stopped halfway because it has been output, and additional voice input cannot be recognized normally, “Recognize until XX prefecture” A response voice with the content “Please give me the post of XX prefecture again” is output. As described above, by outputting a response voice including the fact that the presence of the additional voice input is recognized, it is possible to reduce the user's discomfort and discomfort even when prompting the voice input again.
[0038]
If no additional voice input has been made, a negative determination is made in step 105 described above, and the control unit 14 outputs a response prompting additional voice input as necessary (step 111). Specifically, for example, when only “XX prefecture” is input by the user, a response voice such as “XX prefecture.
[0039]
[Second operation procedure]
By the way, in the first operation procedure shown in FIG. 2 described above, when the input voice includes a blank longer than the predetermined time t1, this is detected, and the voice is input again within the subsequent time t2. The voice recognition process is resumed only once, but the voice recognition process is resumed each time when the voice input is performed again within the time t2 after the blank time t1 or more is detected. You may make it do.
[0040]
FIG. 3 is a flowchart showing a second operation procedure in the speech recognition apparatus 100, and shows an operation procedure when the speech recognition process is restarted each time a speech input is made again within a predetermined time t2. Has been. In the following description, it is assumed that the destination is set in the navigation device 300, and the operation instruction is “XX prefecture XX city Δ △ ……”, that is, “prefecture name” and “ The description will be made on the assumption that speech recognition processing is performed on a continuous word speech composed of a plurality of words following “city name”, “location name”,. Further, in the second operation procedure shown in FIG. 3, since there are many portions that overlap with the operation in the first operation procedure shown in FIG. 2 described above, the overlapped portion will be simplified and described as appropriate.
[0041]
The control unit 14 determines whether or not the talk switch 10 has been pressed (step 200). When the talk switch 10 is pressed, the control unit 14 outputs an activation instruction to the voice recognition processing unit 20.
When the user performs voice input to the microphone 12 after the voice recognition processing unit 20 is activated (step 201), a predetermined voice recognition process is performed by the voice recognition processing unit 20 in response to the voice input. (Step 202).
[0042]
Next, the collation processing unit 28 in the speech recognition processing unit 20 determines whether or not the input speech includes a blank of time t1 or more (step 203). If a blank longer than time t1 is not included, a negative determination is made in step 103, the process returns to step 102, and a predetermined voice recognition process is continued.
[0043]
If the input voice includes a blank for time t1 or more, an affirmative determination is made in step 203, and the collation processing unit 28 determines to interrupt the voice recognition processing by the voice recognition processing unit 20. The recognition result for the voice up to the time of blank detection is output to the control unit 14.
[0044]
Next, the control unit 14 stores the voice feature amount in the ring buffer 24 by sending an instruction to the feature amount extraction unit 26 in the voice recognition processing unit 20 (step 204). By outputting the recognition result acquired from the collation processing unit 28 to the speech synthesis processing unit 32, it is instructed to start output processing of response speech corresponding to speech up to the time of blank detection (step 205).
[0045]
Next, the control unit 14 determines, based on the output signal of the level meter 30, whether or not voice input has been performed within a predetermined time t2 from a blank detection time of time t1 or more (step 206).
Here, the times t1 and t2 in the second operation procedure will be described. FIG. 4 is a diagram for explaining times t1 and t2 in the second operation procedure. As shown in FIG. 5A, when the first input voice includes a blank longer than time t1, this blank is detected, and a predetermined response voice corresponding to the voice inputted so far is detected. Since it is output, the time required from when the blank starts to when the response voice is output (hereinafter referred to as “response time”) t is output as the time t1 corresponding to the blank and the response voice. Equal to the total time required for the processing (response processing). As described above, from the user's point of view, this response time t corresponds to a blank allowed at the time of voice input, that is, an apparent blank, and includes only a blank shorter than the response time t. When input is performed, it is often recognized that speech recognition processing is performed as a continuous word.
[0046]
Therefore, in the present embodiment, the time t2 for determining whether or not voice input has been performed after detecting a blank of time t1 or longer is set to a value that is substantially equal to the time required for response processing. Thereby, as shown in FIG. 4B, after the first voice input (voice input 1) is performed and a blank is detected, before the response voice corresponding to the first voice input is output. That is, when the second voice input (voice input 2) is performed before the time t2 elapses, the response process corresponding to the first voice input is interrupted to correspond to the second voice input. The voice recognition process is started. That is, when speech input is performed without including only a blank shorter than the response time t, speech recognition processing can be performed as a continuous word, so that apparent blanks recognized by the user and speech recognition can be performed. The actual allowable blank time in the apparatus 100 can be made approximately equal.
[0047]
If a voice input is made within time t2, an affirmative determination is made in step 206, and the control unit 14 sends an instruction to the voice synthesis processing unit 32, and a response corresponding to the voice input up to the time of blank detection. The process of outputting the voice is stopped, the resumption of the voice recognition process is decided, a predetermined activation instruction is sent to the voice recognition processing section 20 and the collation processing section 28 is activated (step 207).
[0048]
Upon receiving the activation instruction, the collation processing unit 28 reads out the voice feature quantity stored in the ring buffer 24 (step 208), and then returns to step 202 to perform a predetermined voice recognition process based on the read voice feature quantity. Do.
If no voice is input within the time t2 from the time of blank detection at the time t1 or longer, a negative determination is made in step 206 described above, and the control unit 14 instructs the feature amount extraction unit 26. The operation of sending and storing the audio feature quantity in the ring buffer 24 is stopped (step 209).
[0049]
Further, in parallel with the determination process shown in step 206, the control unit 14 determines whether or not the input voice has been normally recognized based on the recognition result output from the matching processing unit 28. If the speech can be normally recognized, an affirmative determination is made in step 210, and the recognition result character string received from the collation processing unit 28 is sent to the speech synthesis processing unit 32. By outputting, a response sound corresponding to the input sound is output (step 211).
[0050]
Specifically, the collation processing unit 28 outputs a character string when a certain character string (word) can be specified corresponding to all of the input speech, and a part of the speech or If the character strings corresponding to all of them cannot be specified, the character strings corresponding to the fact (error notification) can be output. Therefore, the control unit 14 determines whether or not the voice has been normally recognized based on the presence / absence of an error notification from the verification processing unit 28.
[0051]
If part or all of the speech cannot be recognized, a negative determination is made in step 210, and the control unit 14 sends an instruction to the speech synthesis processing unit 32, and a response corresponding to the recognized speech. Then, a response indicating that other voices (other voices) have been input is output (step 212). Specifically, the user inputs “XX prefecture XX city ……”, but only “XX prefecture” can be recognized, and later “XX city ……” can be recognized. If it was not possible, a response such as “I was able to recognize XX prefectures. In this way, by responding that the presence of the voice that could not be recognized is recognized, the user's discomfort and discomfort when prompting voice input again can be reduced.
[0052]
As described above, in the speech recognition apparatus 100 according to the present embodiment, after the blank included in the speech passes the time t1 and the predetermined response process is started, the speech is not generated until the predetermined time t2 elapses. In the case of input, since the predetermined voice recognition process is resumed, it is possible to eliminate the missed voice that has been input before the response voice is returned. If the voice recognition process fails in response to the success or failure of the voice recognition process restarted after detecting a blank of time t1 or longer, “there is a voice recognition process that recognizes the presence of a voice input by the user. Since the response voice including the content of "failed to" is output, it does not give the user a bad impression that the voice input is ignored or blocked in the middle It is possible to prevent a person from feeling uncomfortable with the response.
[0053]
In addition, this invention is not limited to the said embodiment, A various deformation | transformation implementation is possible within the range of the summary of this invention. For example, in the above-described embodiment, the example in which the voice recognition device 100 to which the present invention is applied is used in combination with the in-vehicle navigation device 300 has been described. However, the scope of the present invention is limited to in-vehicle use. In addition, the present invention can also be applied to various devices and systems such as a speech recognition device realized by using a home personal computer or the like.
[0054]
Further, in the above-described embodiment, since the in-vehicle use is assumed, the speech recognition apparatus 100 including the removing unit for removing the audio sound or the like has been described. However, in the home use, the speech recognition is performed. If the sound other than the sound to be processed has little influence, the removal means may be omitted to simplify the configuration and reduce the cost.
[0055]
【The invention's effect】
As described above, according to the present invention, the predetermined sound pressure is maintained until the predetermined time t2 elapses after the silent state included in the sound has elapsed after the time t1 and the predetermined response process is started. When the voice exceeding the level is input, since the processing by the voice recognition processing means is resumed, it is possible to eliminate the missed voice input until the response is returned. Moreover, since the contents of the response voice are made different depending on the success or failure of the voice recognition process restarted after the silence state of time t1 or more is detected, the user's voice input is ignored or interrupted in the middle. It does not give a bad impression to the user such as being uncomfortable and can eliminate the uncomfortable feeling that the user feels for the response.
[Brief description of the drawings]
FIG. 1 is a diagram illustrating a configuration of a speech recognition apparatus according to an embodiment.
FIG. 2 is a flowchart showing a first operation procedure in the speech recognition apparatus.
FIG. 3 is a flowchart showing a second operation procedure in the speech recognition apparatus.
FIG. 4 is a diagram for explaining times t1 and t2 in the second operation procedure.
[Explanation of symbols]
10 Talk switch
12 Microphone
14 Control unit
16 Delay element
17 Adaptive filter (ADF)
18 Calculation unit
20 Speech recognition processor
22, 24 Ring buffer
26 Feature Extraction Unit
28 Verification processing section
30 Level meter
32 Speech synthesis processor
34 Synthesizer
36 Speaker
100 Voice recognition device
200 audio equipment
300 Navigation device

Claims (3)

A microphone that collects audio,
Speech recognition processing means for performing speech recognition processing on speech including a plurality of words collected by the microphone;
Response means for generating and outputting a response voice based on the content recognized by the voice recognition processing means;
Interruption determining means for detecting a silent state included in the voice collected by the microphone and determining interruption of the voice recognition processing when the silent state continues for a time t1 or more;
Sound pressure level detection means for detecting the sound pressure level of the sound collected by the microphone;
When the sound pressure level detected by the sound pressure level detection means exceeds a predetermined value during the time t2 after the silence t has passed the time t1, the voice recognition processing means Resumption determining means for determining resumption ,
The restart determination means sends an instruction to stop the output of the response voice to the response means together with an operation to instruct the voice recognition processing means to restart the process,
The time t2 is the time from when the duration of the silent state is the time t1 until the interruption determination unit interrupts the speech recognition process and until the response unit outputs the response voice. A speech recognition apparatus characterized by being set to substantially equal values . In claim 1,
A speaker for outputting the response sound and sound output from other sound sources;
Removing means for removing a component other than a voice recognition target included in the voice collected by the microphone;
The speech recognition apparatus, further comprising: a speech recognition target speech output from the removing device is input to the speech recognition processing device. In claim 1 or 2,
The response means generates the response voice having different contents according to the success or failure of the voice recognition processing by the voice recognition processing means restarted after the silence state of the time t1 or more is detected. apparatus.

Images