JP6820086B2 - Voice recognition device and voice recognition method - Google Patents

Description

The present invention relates to a voice recognition device and a voice recognition method, and is particularly suitable for use in a voice recognition device that does not require a trigger such as an operation of a utterance button or a specific operation during voice recognition.

Conventionally, in order to avoid one-handed driving when operating various electronic devices such as audio devices, air conditioners, and navigation devices mounted on vehicles, there is a system that enables the operation of electronic devices by voice recognition. It is provided. By using this voice recognition technology, the driver can operate various electronic devices without taking his / her hand off the steering wheel (without manually operating the operation unit such as the remote controller or the operation panel).

Normally, in a voice recognition device, a user's utterance voice input from a microphone, a specific word or compound word registered in a voice recognition dictionary, a simple command sentence, etc. (hereinafter, these are collectively referred to as a "recognition target word"). ) Is recognized based on the similarity with the voice pattern. For example, a distance value is calculated as an index indicating the degree of similarity between the user's spoken voice and the voice pattern registered in the voice recognition dictionary, and when the distance value falls below the threshold value, the recognition target corresponding to the voice pattern is calculated. Recognize that the ward has been spoken.

In the conventional voice recognition device, when the user presses the built-in utterance button, the voice recognition mode is set and the user's utterance voice input from the microphone is recognized. It is also known that instead of operating the utterance button, a specific action such as clapping a hand triggers the voice recognition mode. On the other hand, recently, voice recognition (hereinafter referred to as triggerless voice recognition) that does not require a trigger such as an operation of a utterance button or a specific action during voice recognition has been provided.

In the triggerless voice recognition device, the microphone is always on, and it is always determined whether or not the input voice corresponds to the recognition target word. In the passenger compartment, the voice input from the microphone includes various noises such as engine operation sound, running sound, audio voice, and conversation voice between passengers, in addition to the spoken voice for voice recognition. ing. Therefore, in the triggerless voice recognition device, it is necessary to devise a device for correctly performing voice recognition even in such a noisy environment.

Conventionally, a technique for suppressing erroneous recognition by audio-voice has been provided (see, for example, Patent Document 1). In the technique described in Patent Document 1, two voice recognition engines are installed, and the first voice recognition engine recognizes the voice input from the microphone and at the same time internally branches the audio voice originally output from the speaker. It is input to the second voice recognition engine, and the audio voice is recognized by the second voice recognition engine. Then, when the same word is recognized by each voice recognition engine, it is rejected as an erroneous recognition by audio voice.

Jikkenhei 7-23400

In the case of triggerless voice recognition, it is always determined whether or not the voice input from the microphone corresponds to the recognition target word, so that the processing load is always applied to the CPU of the in-vehicle device. When one voice recognition engine supports 12 words to be recognized, the CPU load increases by about 5%. When two voice recognition engines are operated at the same time as in Patent Document 1, the total number of words corresponding to 12 recognition target words is 24 words, which increases the CPU load by about 10%.

In such a situation, there arises a problem that the response performance of other processes other than the voice recognition process is deteriorated. For example, there arises a problem that the map drawing performance of the navigation device (fps = frame per second, which is the number of frames that can be drawn per second) is deteriorated. In order to suppress the deterioration of the response performance of other processes, it is necessary to limit the number of words to be recognized that are constantly waiting. However, in this way, only a very limited number of words to be recognized can be recognized by voice, and the convenience of triggerless voice recognition is impaired.

The present invention has been made to solve such a problem, without limiting the number of words to be recognized that are constantly waiting in triggerless speech recognition, and the response of processing other than speech recognition processing. The purpose is to make it possible to suppress erroneous recognition by voice generated by an in-vehicle device while suppressing deterioration of performance as much as possible.

In order to solve the above-mentioned problems, the speech recognition device of the present invention includes a first speech recognition dictionary in which the entire recognition target word is registered, and a second speech recognition dictionary in which only the latter half of the recognition target word is registered. Has. Further, the voice recognition device of the present invention is an in-vehicle device using a first voice recognition dictionary, a first voice recognition unit that performs voice recognition of externally input voice input from a microphone, and a second voice recognition dictionary. It is provided with a second voice recognition unit that performs voice recognition of the internally generated voice before being output from the speaker. The first voice recognition unit sequentially calculates the similarity in parallel with the sequential input of the external input voice, and when the calculated similarity becomes larger than the first level, the external input voice is the first half of the recognition target word. It recognizes that it corresponds to a part, and when the calculated similarity becomes larger than the second level, it recognizes that the external input voice corresponds to the whole recognition target word. The second voice recognition unit starts the recognition process when the similarity calculated by the first voice recognition unit becomes larger than the first level, and internally occurs when the calculated similarity is larger than the predetermined level. Recognize that the voice corresponds to the latter half of the recognition target word. Then, when the first voice recognition unit recognizes that the externally input voice corresponds to the entire recognition target word and the second voice recognition unit recognizes that the internally generated voice corresponds to the latter half of the recognition target word, the first 1 The recognition result by the voice recognition unit is discarded.

According to the present invention configured as described above, the second voice recognition unit is activated only when the first voice recognition unit recognizes that the external input voice corresponds to the first half of the recognition target word. The processing load can be reduced as compared with the case where the voice recognition unit is always in operation. Since the processing load is small, it is not necessary to limit the number of words to be recognized that are constantly waiting in triggerless speech recognition. When the first voice recognition unit recognizes that the externally input voice corresponds to the entire recognition target word and the second voice recognition unit recognizes that the internally generated voice corresponds to the latter half of the recognition target word. , The recognition result by the first voice recognition unit is discarded as erroneous recognition caused by the internally generated voice output from the speaker being input from the microphone. As a result, according to the present invention, the number of words to be recognized that are constantly waiting in triggerless speech recognition is not limited to a small number, and the deterioration of the response performance of other processes other than the speech recognition process is suppressed as much as possible. It is possible to suppress erroneous recognition due to the generated voice.

It is a block diagram which shows the functional structure example of the voice recognition apparatus which concerns on this embodiment. It is a figure which shows an example of the recognition target word. It is a figure which shows an example of the information which the 1st voice recognition dictionary and the 2nd voice recognition dictionary which concern on this Embodiment store. It is a graph which shows an example of the transition of the distance value calculated by the 1st voice recognition unit and the 2nd voice recognition unit which concerns on this embodiment. It is a flowchart which shows the operation example of the 1st voice recognition part. It is a flowchart which shows the operation example of the 2nd voice recognition part. It is a flowchart which shows the operation example of the recognition result discarding part.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings. FIG. 1 shows a block diagram showing a functional configuration example of the voice recognition device 100 according to the present embodiment, showing a microphone 200, a navigation device 300, an audio device 400, a camera system 500, a display device 600, and a voice output provided in the vehicle. It is a figure which shows together with the apparatus 700. The voice recognition device 100 according to the present embodiment recognizes the utterance voice (words such as a specific word, a compound word, a simple command sentence) of the occupant input from the microphone 200 provided in the vehicle, and recognizes it as a utterance command. The navigation device 300 is controlled based on the recognition result. Although the electronic device to be controlled by the voice recognition device 100 is the navigation device 300 here, it may be an audio device 400, an air conditioner, or other electronic device.

The microphone 200 is a sound collecting device, and is provided at a position where the voice of a passenger boarding the vehicle can be picked up. The microphone 200 outputs a voice signal based on the picked-up voice to the first voice recognition unit 10 described later. Hereinafter, the voice picked up by the microphone 200 is referred to as an "external input voice", and the voice signal output by the microphone 200 to the first voice recognition unit 10 is referred to as an "external input voice signal".

A display device 600 and a camera system 500 are connected to the navigation device 300. The display device 600 is a device capable of displaying an image such as a liquid crystal display panel, and is provided in the center of a dashboard, for example. The camera system 500 includes a front camera that shoots the front of the vehicle and a rear camera that shoots the rear of the vehicle, and in response to a request from the navigation device 300, the shooting result of either camera is sent to the navigation device 300. Output.

The navigation device 300 has a function of detecting the position of the vehicle, a function of displaying a map on the display device 600 and displaying the position of the vehicle on the map, a function of searching for a route to the destination, and a map on the display device 600. It has a function to display and draw a route to the destination on a map to guide the route. Further, the navigation device 300 has a function of displaying an image showing the shooting result of the front camera or an image showing the shooting result of the rear camera on the display device 600 according to the instruction of the user.

The audio output device 700 includes a D / A converter, a volume, an amplifier, a speaker, and the like, D / A converts the input audio signal, amplifies it, and then outputs the audio through the speaker. The audio device 400 may be audio data recorded on media (CD, DVD, MD, etc.) or a memory (internal memory mounted on the audio device 400, or an external memory connected to the device. A voice signal is generated based on the voice data stored in the voice output device 700 and output to the voice output device 700. The audio device 400 may be any device that outputs a voice signal to the voice output device 700 and emits sound into the vehicle, and may be, for example, a radio receiving device. As shown in FIG. 1, the voice signal output by the audio device 400 is branched and output to the second voice recognition unit 12 described later. Hereinafter, the voice signal input by the second voice recognition unit 12 is referred to as an "internally generated voice signal", and the voice based on this internally generated voice signal is referred to as an internally generated voice. The internally generated voice is the voice generated by the audio device 400, which is one of the on-board units, and before being output from the speaker.

As shown in FIG. 1, the voice recognition device 100 according to the present embodiment includes a first voice recognition unit 10, an electronic device control unit 11, a second voice recognition unit 12, and a recognition result discard unit 13 as its functional configuration. ing. Further, the voice recognition device 100 includes a dictionary storage unit 20 as a storage medium.

The functional blocks 10 to 13 can be configured by any of hardware, DSP (Digital Signal Processor), and software. For example, when configured by software, each of the functional blocks 10 to 13 is actually configured to include a computer CPU, RAM, ROM, etc., and is a program stored in a recording medium such as RAM, ROM, hard disk, or semiconductor memory. Is realized by the operation of.

The dictionary storage unit 20 stores the first voice recognition dictionary 20A and the second voice recognition dictionary 20B. Hereinafter, the recognition target word will be described, and then the first voice recognition dictionary 20A and the second voice recognition dictionary 20B will be described.

FIG. 2 is an example of a list of recognition target words in this embodiment. The recognition target word illustrated in FIG. 2 is just an example, and it goes without saying that other recognition target words may exist. As shown in FIG. 2, in the present embodiment, nine recognition target words are prepared as recognition target words. Therefore, when the passenger utters a word corresponding to any of the nine recognition target words among the nine recognition target words, the voice recognition device 100 recognizes the spoken voice as the recognition target word and sets it as the recognition target word. The navigation device 300 is made to execute the corresponding process. Further, identification information of numbers A1 to A9 is assigned to each of the nine recognition target words.

To briefly explain the process that the voice recognition device 100 causes the navigation device 300 to execute when a word corresponding to each word is spoken for each of the recognition target words, the recognition target word of the number A1 is the current position of the vehicle. It is a word for executing the process of searching for a convenience store near. Further, the recognition target word of the number A2 is a word for executing a process of searching for a gas station near the current position of the vehicle. The recognition target word of the number A3 is a word for executing a process of guiding the route to the destination. The recognition target word of the number A4 is a word for executing a process of guiding the route to the home registered in advance. The recognition target word of the number A5 is a word for executing a process of displaying a map of a predetermined scale centered on the current position of the vehicle on the display device 600. The recognition target word of the number A6 is a word for executing a process of enlarging the scale of the displayed map. The recognition target word of the number A7 is a word for executing a process of reducing the scale of the displayed map. The recognition target word of the number A8 is a word for executing a process of displaying an image indicating the shooting result of the front camera on the display device 600. The recognition target word of the number A9 is a word for executing a process of displaying an image indicating the shooting result of the rear camera on the display device 600.

In the present embodiment, each of the recognition target words is divided into a first half portion and a second half portion in advance. In FIG. 2, the symbol “/” indicates a delimiter between the first half portion and the second half portion of each recognition target word. For example, the recognition target word of the number A1 is a recognition target word composed of the wording "near convenience store". The recognition target word of the number A1 is divided into the wording "near" and the wording "convenience store". Further, for example, the recognition target word of the number A8 is a recognition target word composed of the word "front camera". The recognition target word of the number A8 is divided into the wording "front" and the wording "camera".

FIG. 3A is a diagram schematically showing information stored in the first speech recognition dictionary 20A with contents suitable for explanation. FIG. 3B is a diagram schematically showing information stored in the second speech recognition dictionary 20B with contents suitable for explanation. As shown in FIG. 3A, in the first speech recognition dictionary 20A, for each of all the recognition target words (in this embodiment, each of the above-mentioned nine recognition target words), the recognition target word " The "whole" voice pattern is registered.

As shown in FIG. 3B, in the second voice recognition dictionary 20B, the voice pattern of the "second half" of the recognition target word is registered for each of the recognition target words. For example, for the recognition target word of the number A1, the voice pattern of the word "convenience store" in the latter half of the recognition target word is registered in the second voice recognition dictionary 20B. As shown in FIG. 2, the latter half of the recognition target words of the numbers A5, A6, and A7 are common to "display". Based on this, as shown in FIG. 3B, with respect to the recognition target words of numbers A5, A6, and A7, the voice pattern of the wording "display" is 1 as the voice pattern of the "second half" of the recognition target word. Only one is registered in the second speech recognition dictionary 20B. This also applies to the recognition target words of numbers A8 and A9 whose latter half is common to the word "camera". As a result, in the second speech recognition dictionary 20B according to the present embodiment, there are six words "convenience store", word "gas station", word "guidance", word "return", word "display", and word "camera". The voice pattern of the latter half of the recognition target word of is registered.

The first voice recognition unit 10 calculates the degree of similarity between the entire voice pattern of the recognition target word registered in the first voice recognition dictionary 20A and the externally input voice input from the microphone 200, and the calculated similarity. When the degree is greater than a predetermined level, it is recognized that the externally input voice corresponds to the entire recognition target word. More specifically, the first voice recognition unit 10 sequentially calculates the similarity in parallel with the sequential input of the external input voice, and when the calculated similarity becomes larger than the first level, the external input It recognizes that the voice corresponds to the first half of the recognition target word, and when the calculated similarity becomes larger than the second level, it recognizes that the external input voice corresponds to the entire recognition target word.

In the present embodiment, the first voice recognition unit 10 calculates a distance value as an index indicating the degree of similarity. The distance value takes a value in the range of the value "0" to the value "1000". The distance value related to the recognition target word 1 calculated by the first voice recognition unit 10 is such that the external input voice and the “whole” voice pattern of the recognition target word are similar to each other. It becomes smaller. Then, the first voice recognition unit 10 detects that the calculated distance value is smaller than the first threshold value T1 (> second threshold value T2 (described later)), so that the similarity becomes larger than the first level. It is recognized that the external input voice corresponds to the first half of the recognition target word at that time. The first voice recognition unit 10 detects that the calculated distance value is smaller than the first threshold value T1 and then the calculated distance value is smaller than the second threshold value T2, so that the similarity is second. It detects that the level is higher than the level, and at that point, recognizes that the externally input voice corresponds to the entire recognition target word.

Hereinafter, the processing of the first voice recognition unit 10 will be described in detail. FIG. 4A is a graph showing an example of the transition of the distance value calculated by the first voice recognition unit 10 for a certain recognition target word. In particular, FIG. 4A shows an example of the transition of the distance value when the distance value falls below the first threshold value T1 and then further falls below the second threshold value T2. The vertical axis of the graph of FIG. 4A shows the distance value, and the horizontal axis shows the passage of time. The first voice recognition unit 10 executes so-called triggerless voice recognition and constantly calculates the distance value.

As described above, the microphone 200 outputs an external input voice signal based on the collected voice to the first voice recognition unit 10. The first voice recognition unit 10 has, for each of the nine recognition target words, the voice waveform of the external input voice signal and the voice pattern registered in the first voice recognition dictionary 20A in parallel with the sequential input of the external input voice. The distance value is calculated sequentially based on the comparison with (the "whole" voice pattern of the recognition target word). As a result, as shown in FIG. 4A, the value of the distance value changes sequentially with the passage of time.

When the distance value is above the first threshold value T1, the first voice recognition unit 10 monitors whether or not the distance value has shifted to a state below the first threshold value T1. In FIG. 4A, at the timing TM1, the distance value shifts from the state where the distance value exceeds the first threshold value T1 to the state where the distance value falls below the first threshold value T1. When the first voice recognition unit 10 detects that the distance value has shifted to a state below the first threshold value T1, at that time (timing TM1 in FIG. 4A), the external input voice is the recognition target word. Recognize that it corresponds to the first half.

Here, when the microphone 200 picks up the voice of the wording corresponding to the recognition target word of 1, the distance value between the external input voice signal based on the picked up voice and the voice pattern of the recognition target word of 1 is , Gradually getting smaller. More specifically, as the amount of comparison between the external input voice signal based on the collected voice and the voice pattern of the recognition target word of 1 increases, the voice waveform of the external input voice signal and the recognition target word of 1 concerned. The matching rate with the voice pattern of is gradually increased, and the distance value is gradually decreased accordingly, and the value falls below the first threshold value T1 at a certain point in time and further falls below the second threshold value T2 at a subsequent point in time.

The value of the first threshold value T1 related to the recognition target word of 1 is the "first half portion" of the recognition target word of 1 when the microphone 200 picks up the voice of the wording corresponding to the recognition target word of 1. When the voice of the word corresponding to is picked up and the calculation of the distance value based on the external input voice corresponding to the "first half part" of the recognition target word of 1 is completed, the distance value reaches the first threshold value T1. Is set to a value like. The value of the first threshold value T1 is appropriately set for each recognition target word based on the results of prior tests and the like. Based on the above, when the first voice recognition unit 10 detects that the distance value has shifted from the state where the distance value exceeds the first threshold value T1 to the state where the distance value has fallen below the first threshold value T1, the external input voice is the recognition target word at that time. Recognize that it corresponds to the first half of.

When the first voice recognition unit 10 detects that the distance value has shifted from the state where the distance value exceeds the first threshold value T1 to the state where the distance value has fallen below the first threshold value T1, the first voice recognition unit 10 outputs a processing start notification to the second voice recognition unit 12. In the graph illustrated in FIG. 4A, the first voice recognition unit 10 outputs a processing start notification at the timing TM1. The processing start notification will be described later together with the processing of the second voice recognition unit 12.

When the distance value falls below the first threshold value T1, the first voice recognition unit 10 starts measuring the elapsed time from that time point. Then, the first voice recognition unit 10 monitors whether or not the time J1 has elapsed as the elapsed time since the distance value has fallen below the first threshold value T1, and whether or not the distance value has fallen below the second threshold value T2. To monitor.

Here, the values of the second threshold value T2 and the time J1 can be regarded as the external input voice corresponding to the entire recognition target word when the distance value falls below the second threshold value T2 without the lapse of the time J1. On the contrary, when the time J1 elapses without the distance value falling below the second threshold value T2, the value is set so that the external input voice does not correspond to the entire recognition target word. The second threshold value T2 and the time J1 are set to appropriate values for each recognition target word registered in the first speech recognition dictionary 20A based on the results of prior tests and the like. If the time J1 elapses without the distance value falling below the second threshold value T2, the distance value gradually increases after the time J1 elapses, and eventually exceeds the first threshold value T1.

Based on the above, when the time J1 elapses without the distance value falling below the second threshold value T2, the first voice recognition unit 10 stops determining whether or not the distance value falls below the second threshold value T2. It is monitored whether or not the distance value has shifted from the state where the distance value is above the first threshold value T1 to the state where the distance value is below the first threshold value T1. As described above, after the elapse of the time J1, the distance value will eventually exceed the first threshold value T1. Therefore, the first voice recognition unit 10 once the distance value exceeds the first threshold value T1. , It is monitored whether or not the transition from the state above the first threshold value T1 to the state below the first threshold value T1 has occurred. In the present embodiment, it is determined that the externally input voice does not correspond to the entire recognition target word when the time J1 elapses without the distance value falling below the second threshold value T2. , The configuration in which the external input voice does not correspond to the entire recognition target word when the distance value exceeds the first threshold value T1 without falling below the second threshold value T2 after falling below the first threshold value T1. May be.

On the other hand, when the distance value falls below the second threshold value T2 before the time J1 elapses, the first voice recognition unit 10 recognizes that the external input voice corresponds to the entire recognition target word. In the graph illustrated in FIG. 4A, the first voice recognition unit 10 recognizes that the external input voice corresponds to the entire recognition target word in the timing TM2. When the distance value is below the second threshold value T2, the distance value once falls below the second threshold value T2, then gradually increases, and eventually exceeds the first threshold value T1.

After the above recognition, the first voice recognition unit 10 recognizes the first voice recognition notification for notifying that the externally input voice corresponds to the entire recognition target word for any one recognition target word. Output to 13. The processing of the recognition result discarding unit 13 will be described later.

Next, the first voice recognition unit 10 monitors whether or not either the recognition destruction notification or the recognition confirmation notification is input from the recognition result destruction unit 13. The recognition discard notification indicates to discard the recognition regarding the recognition that the external input voice corresponds to the entire recognition target word (hereinafter, may be simply referred to as "recognition result of the first voice recognition unit 10"). It is a notification. The recognition discard notification will be described in detail later, but the voice emitted by the audio device 400 accidentally contains a word corresponding to the recognition target word, and the result of voice processing of an external input signal based on the voice. , The first voice recognition unit 10 is notified when it recognizes that the externally input voice corresponds to the entire recognition target word. In this case, since the passenger did not utter the word corresponding to the recognition target word, the recognition result of the first voice recognition unit 10 is discarded, and the navigation device 300 is not controlled based on the recognition result. Must be done. Further, the recognition confirmation notification will be described in detail later, but the recognition result of the first voice recognition unit 10 is not derived from the result of voice processing for the voice emitted by the audio device 400, but the voice spoken by the passenger. Is notified when it comes from the result of voice processing for. In this case, it is necessary for the navigation device 300 to execute the process corresponding to the recognition target word.

When the recognition discard notification is input, the first voice recognition unit 10 discards the recognition that the externally input voice corresponds to the entire recognition target word. In this case, the navigation device 300 is not controlled based on the voice picked up by the microphone 200. After that, the first voice recognition unit 10 monitors whether or not the distance value has shifted from the state where the distance value exceeds the first threshold value T1 to the state where the distance value has fallen below the first threshold value T1. As described above, since the distance value falls below the second threshold value T2 and then eventually exceeds the first threshold value T1, the first voice recognition unit 10 temporarily sets the distance value to the first threshold value T1. After the threshold is exceeded, it is monitored whether or not the transition from the state where the threshold value T1 is exceeded to the state where the threshold value T1 is lower is changed.

On the other hand, when the recognition confirmation notification is input, the first voice recognition unit 10 confirms the recognition that the externally input voice corresponds to the entire recognition target word. Next, the first voice recognition unit 10 notifies the electronic device control unit 11 of the recognition target word (hereinafter, referred to as “fixed recognition target word”) recognized as corresponding to the external input voice. The definite recognition target word is a recognition target word corresponding to an utterance command uttered by the passenger to cause the navigation device 300 to execute a specific process. After that, the first voice recognition unit 10 monitors whether or not the distance value has shifted from the state where the distance value exceeds the first threshold value T1 to the state where the distance value has fallen below the first threshold value T1.

When the first voice recognition unit 10 notifies the definite recognition target word, the electronic device control unit 11 outputs a control signal to the navigation device 300 to cause the navigation device 300 to execute the process corresponding to the notified definite recognition target word. To do. The navigation device 300 executes the process based on the input control signal.

The second voice recognition unit 12 calculates the similarity between the voice pattern of the latter half of the recognition target word registered in the second voice recognition dictionary 20B and the internally generated voice described above, and the calculated similarity is predetermined. When it is larger than the level, it recognizes that the internally generated voice corresponds to the latter half of the recognition target word. More specifically, the second voice recognition unit 12 calculates the distance value as an index indicating the degree of similarity as in the first voice recognition unit 10, and the calculated distance value is smaller than the third threshold value T3. By detecting, it is detected that the similarity becomes larger than a predetermined level. Then, when the calculated distance value becomes smaller than the third threshold value T3, the second voice recognition unit 12 recognizes that the internally generated voice corresponds to the latter half of the recognition target word.

Further, in the second voice recognition unit 12, when the distance value calculated by the first voice recognition unit 10 becomes smaller than the first threshold value T1 (= the similarity calculated by the first voice recognition unit 10 is first. The recognition process is started when the level becomes higher than the level of. The recognition process means the calculation of the distance value using the second speech recognition dictionary 20B and various processes using the calculated distance value. Further, when the second voice recognition unit 12 recognizes that the internally generated voice corresponds to the latter half of the recognition target word before the time J2 (described later) elapses after starting the recognition process, or the internally generated voice is heard. When the time J2 elapses without recognizing that it corresponds to the latter half of the recognition target word, the recognition process is stopped.

Hereinafter, the processing of the second voice recognition unit 12 will be described in detail. FIG. 4B is a graph showing an example of the transition of the distance value calculated by the second voice recognition unit 12 for a certain recognition target word. In particular, the graph of FIG. 4B shows the transition of the distance value when the distance value is less than the third threshold value T3 within the time J2 in the recognition process executed by the second voice recognition unit 12. The vertical axis of the graph of FIG. 4B shows the distance value, and the horizontal axis shows the passage of time. Each timing on the horizontal axis of the graph of FIG. 4 (B) corresponds to each timing of the horizontal axis of the graph of FIG. 4 (A).

In the following description, the latter half of the recognition target word in which the voice pattern is registered in the second speech recognition dictionary 20B is referred to as a "second half word". As described with reference to FIG. 3B, six latter-half partial words are prepared in the present embodiment.

As described above, the voice signal output by the audio device 400 to the voice output device 700 is branched and output to the second voice recognition unit 12. Further, as described above, when the first voice recognition unit 10 detects that the distance value for the recognition target word of 1 is less than the first threshold value T1, at that time point (timing TM1 in FIG. 4), The processing start notification is output to the second voice recognition unit 12. The second voice recognition unit 12 does not execute the recognition process (calculation of the distance value using the second voice recognition dictionary 20B and the accompanying process as described above) until the process start notification is input. Monitor whether you have entered a start notification. Then, the second voice recognition unit 12 starts the recognition process when the process start notification is input. As a result, the second voice recognition unit 12 starts the recognition process when the distance value calculated by the first voice recognition unit 10 becomes smaller than the first threshold value T1. In the graph illustrated in FIG. 4B, the second voice recognition unit 12 starts the recognition process at the timing TM1 (see also FIG. 4A).

In the recognition process, the second voice recognition unit 12 registers the voice waveform of the internally generated voice signal and the second voice recognition dictionary 20B for each of the six latter half words in parallel with the sequential input of the internally generated voice. The distance value is calculated sequentially based on the comparison with the voice pattern (the voice pattern of the "second half" of the recognition target word). As a result, as shown in FIG. 4B, the value of the distance value changes sequentially with the passage of time.

The second voice recognition unit 12 starts measuring the elapsed time at the same time as the start of the recognition process. Then, the second voice recognition unit 12 monitors whether or not the distance value is below the third threshold value T3 while monitoring whether or not the time J2 has elapsed since the recognition process was started.

Here, as for the values of the third threshold value T3 and the time J2, when the distance value falls below the third threshold value T3 without the lapse of the time J2, the internally generated voice becomes the second half word (the second half of the recognition target word). On the contrary, when the time J2 elapses without the distance value falling below the third threshold value T3, the internally generated voice is set to a value that can be regarded as not corresponding to the latter half word. To. The third threshold value T3 and the time J2 are set to appropriate values for each of the latter half words registered in the second speech recognition dictionary 20B based on the results of the preliminary test and the like.

When the time J2 elapses without the distance value falling below the third threshold value T3, the second voice recognition unit 12 outputs the unrecognizable notification to the recognition result discard unit 13. The unrecognizable notification is a notification that the internally generated voice is not recognized as the latter half word in the recognition process. The processing of the recognition result discarding unit 13 will be described later. After that, the second voice recognition unit 12 stops the recognition process. After stopping the recognition process, the second voice recognition unit 12 does not start the recognition process until the above-mentioned processing start notification is input.

On the other hand, if the distance value falls below the third threshold value T3 before the time J2 elapses, the second voice recognition unit 12 recognizes that the internally generated voice corresponds to the latter half word. After the recognition, the second voice recognition unit 12 outputs the second voice recognition notification to the recognition result discarding unit 13. The second voice recognition notification is a notification that the internally generated voice has been recognized as corresponding to any one of the latter half words among the latter half words registered in the second voice recognition dictionary 20B. The processing of the recognition result discarding unit 13 will be described later. After that, the second voice recognition unit 12 stops the recognition process. After stopping the recognition process, the second voice recognition unit 12 does not start the recognition process until the above-mentioned processing start notification is input. In the graph of FIG. 4B, the second voice recognition unit 12 executes the output of the second voice recognition notification and the stop of the recognition process at the timing TM3.

The timing at which the first voice recognition unit 10 outputs the first voice recognition notification and the timing at which the second voice recognition unit 12 outputs the second voice recognition notification or the unrecognizable notification are very close in time. .. Further, when the first voice recognition notification is output to the recognition result discarding unit 13 by the first voice recognition unit 10, the second voice recognition notification or the unrecognizable notification is always sent by the second voice recognition unit 12 to the recognition result discarding unit 13. Is output to.

In the recognition result discarding unit 13, the first voice recognition unit 10 recognizes that the externally input voice corresponds to the entire recognition target word, and the second voice recognition unit 12 recognizes that the internally generated voice is the latter half word (recognition target word). When it is recognized that it corresponds to the latter half), the recognition result by the first voice recognition unit 10 is discarded.

Hereinafter, the processing of the recognition result discarding unit 13 will be described in detail. As described above, when the first voice recognition unit 10 recognizes that the external input voice corresponds to the entire recognition target word, the first voice recognition unit 10 outputs the first voice recognition notification to the recognition result discard unit 13. The recognition result discarding unit 13 monitors whether or not the first voice recognition notification is input from the first voice recognition unit 10. Then, when the first voice recognition notification is input, the recognition result discarding unit 13 inputs either the second voice recognition notification or the unrecognizable notification at a timing close in time to the first voice recognition notification.

When the notification input from the second voice recognition unit 12 is the second voice recognition notification, the recognition result discard unit 13 outputs the recognition discard notification to the first voice recognition unit 10. As described above, when the recognition discard notification is input, the first voice recognition unit 10 discards the recognition regarding the recognition that the externally input voice corresponds to the entire recognition target word. That is, when the second voice recognition notification is input, the recognition result discard unit 13 causes the first voice recognition unit 10 to discard the recognition result.

On the other hand, when the notification input from the second voice recognition unit 12 is an unrecognizable notification, the recognition result discard unit 13 outputs the recognition confirmation notification to the first voice recognition unit 10. As described above, when the recognition confirmation notification is input, the first voice recognition unit 10 notifies the electronic device control unit 11 of the confirmation recognition target word, and executes the process corresponding to the confirmation recognition target word to the navigation device 300. Let me. That is, when the recognition result discard unit 13 inputs the unrecognizable notification, the recognition result of the first voice recognition unit 10 is determined, and the processing corresponding to the navigation device 300 is executed.

Here, the sound emitted by the audio device 400 may accidentally include a wording corresponding to the recognition target word. In such a case, the first voice recognition unit 10 recognizes that the external input voice (voice including the wording corresponding to the recognition target word) collected by the microphone 200 corresponds to the entire recognition target word. It is necessary to discard such recognition so that the navigation device 300 is not controlled. This is because the external input voice is not the voice emitted by the passenger. Then, as described above, in the present embodiment, even when the first voice recognition unit 10 recognizes that the external input voice corresponds to the entire recognition target word, the second voice recognition unit 12 causes the internally generated voice. When is recognized as corresponding to the latter half word, the recognition of the first voice recognition unit 10 is discarded and the navigation device 300 is not controlled. As a result, even if the voice emitted by the audio device 400 accidentally contains a word corresponding to the recognition target word, the navigation device 300 can be prevented from being controlled. The details will be described below.

As described above, the period during which the recognition process is performed by the second voice recognition unit 12 is a predetermined period following the timing when the first voice recognition unit 10 recognizes that the external input voice corresponds to the first half of the recognition target word. is there. In this way, the first voice recognition unit 10 and the second voice recognition unit 12 receive the first voice recognition unit 10 and the second voice recognition unit 12 during a predetermined period following the timing when the first voice recognition unit 10 recognizes that the external input voice corresponds to the first half of the recognition target word. In both cases, when it is recognized that the voice corresponds to the latter half of the recognition target word, it can be said that the situation is as follows. That is, the voice emitted by the audio device 400 includes the wording corresponding to the recognition target word, and the first voice recognition unit 10 recognizes the voice emitted by the audio device 400. It is highly likely that it corresponds to the latter half of.

That is, even when the first voice recognition unit 10 recognizes that the external input voice corresponds to the entire recognition target word, the second voice recognition unit 12 recognizes that the internally generated voice corresponds to the latter half word. In this case, it is highly possible that the recognition of the first voice recognition unit 10 is not derived from the voice generated by the passenger, but from the voice emitted by the audio device 400. Based on this, even when the first voice recognition unit 10 recognizes that the externally input voice corresponds to the entire recognition target word, the second voice recognition unit 12 considers that the internally generated voice corresponds to the latter half word. When it is recognized, by discarding the recognition of the first voice recognition unit 10, even if the voice emitted by the audio device 400 contains a word corresponding to the word to be recognized, the voice becomes the voice. Based on this, it is possible to prevent the navigation device 300 from being controlled to execute the process.

Further, in the present embodiment, the second voice recognition unit 12 executes the recognition process only when the first voice recognition unit 10 recognizes that the external input voice corresponds to the first half of the recognition target word. As a result, the first voice recognition unit 10 recognizes that the external input voice corresponds to the first half of the recognition target word, and then the first voice recognition unit 10 determines that the external input voice corresponds to the "whole" of the recognition target word. Only in the situation where there is a possibility of recognition, the second voice recognition unit 12 can perform the recognition process, and the recognition result of the first voice recognition unit 10 can be accurately discarded when necessary.

In this way, the second voice recognition unit 12 executes the recognition process only when the first voice recognition unit 10 recognizes that the external input voice corresponds to the first half of the recognition target word, and the second voice recognition unit 12 executes the recognition process. The maximum period during which the voice recognition unit 12 executes the recognition process is time J2. With such a configuration, the period for executing the recognition process is limited and the processing load can be reduced as compared with the case where the second voice recognition unit 12 constantly executes the recognition process. Since the processing load is small, when other processing other than the voice recognition processing by the first voice recognition unit 10 and the second voice recognition unit 12 is executed, the deterioration of the response performance of the other processing is suppressed. Therefore, in consideration of adverse effects on other processing (for example, a significant decrease in CPU allocation to other processing) due to an increase in the processing load related to the recognition processing of the second voice recognition unit 12, triggerless by the first voice recognition unit 10. It is not necessary to limit the number of recognition target words (recognition target words registered in the first speech recognition dictionary 20A) that are constantly waiting in speech recognition.

Further, in the present embodiment, the wording of the latter half of the plurality of recognition target words may be common, and in such a case, one second half word of the common wording is registered in the second speech recognition dictionary 20B. To. As a result, the number of the latter half words registered in the second voice recognition dictionary 20B can be reduced, and the processing load of the second voice recognition unit 12 can be reduced more effectively.

FIG. 5 is a flowchart showing an operation example of the first voice recognition unit 10 according to the present embodiment. FIG. 6 is a flowchart showing an operation example of the second voice recognition unit 12 according to the present embodiment. FIG. 7 is a flowchart showing an operation example of the recognition result discarding unit 13 according to the present embodiment. The processing of each flowchart of FIGS. 5, 6 and 7 is appropriately executed after the power of the voice recognition device 100 is turned on and the start of triggerless voice recognition is instructed.

In the following description, it is assumed that the distance value calculated by the first voice recognition unit 10 exceeds the first threshold value T1 at the start of the flowchart of FIG. Further, although not particularly described, the first voice recognition unit 10 continuously calculates the distance value while the processing of the flowchart of FIG. 5 is performed.

As shown in the flowchart of FIG. 5, the first voice recognition unit 10 determines whether or not the distance value calculated at the present time has shifted from the state where the distance value exceeds the first threshold value T1 to the state where the distance value falls below the first threshold value T1. (Step SA1). In this step SA1, the first voice recognition unit 10 does not determine whether or not the distance value is below the first threshold value T1, but from the “state where the distance value is above the first threshold value T1”, the “first threshold value T1”. It is determined whether or not there has been a change in the state to "a state below." The first voice recognition unit 10 continuously and repeatedly executes the process of step SA1 until it detects that the distance value has shifted from the state where the distance value exceeds the first threshold value T1 to the state where the distance value has fallen below the first threshold value T1.

When it is detected that the distance value shifts from the state where the distance value exceeds the first threshold value T1 to the state where the distance value falls below the first threshold value T1, the first voice recognition unit 10 determines that the external input voice corresponds to the first half of the recognition target word. Recognize (step SA2). Next, the first voice recognition unit 10 outputs a processing start notification to the second voice recognition unit 12 (step SA3). Next, the first voice recognition unit 10 starts measuring the elapsed time (step SA4).

Next, the first voice recognition unit 10 determines whether or not the time J1 has elapsed (step SA5), and determines whether or not the distance value has fallen below the second threshold value T2 (step SA6).

When the time J1 elapses without the distance value falling below the second threshold value T2 (step SA5: YES), the first voice recognition unit 10 shifts the processing procedure to step SA1. On the other hand, when the distance value falls below the second threshold value T2 before the time J1 elapses (step SA6: YES), the first voice recognition unit 10 recognizes that the external input voice corresponds to the entire recognition target word (step SA6: YES). Step SA7). Next, the first voice recognition unit 10 outputs the first voice recognition notification to the recognition result discard unit 13 (step SA8).

After outputting the first voice recognition notification, the first voice recognition unit 10 determines whether or not either the recognition destruction notification or the recognition confirmation notification has been input from the recognition result discard unit 13 (step SA9). The first voice recognition unit 10 continuously and repeatedly executes the process of step SA9 until any notification is input. When any of the notifications is input (step SA9: YES), the first voice recognition unit 10 determines whether or not the input notification is a recognition discard notification (step SA10).

When the input notification is a recognition discard notification (step SA10: YES), the first voice recognition unit 10 discards the recognition result (step SA11). After that, the first voice recognition unit 10 shifts the processing procedure to step SA14.

On the other hand, when the input notification is not the recognition discard notification (step SA10: NO), the first voice recognition unit 10 confirms the recognition that the externally input voice corresponds to the entire recognition target word (step SA12). .. Next, the first voice recognition unit 10 notifies the electronic device control unit 11 of the definite recognition target word (step SA13). After that, the first voice recognition unit 10 shifts the processing procedure to step SA14.

In step SA14, the first voice recognition unit 10 determines whether or not the end of triggerless voice recognition is instructed. When the end of the triggerless voice recognition is not instructed (step SA14: NO), the first voice recognition unit 10 shifts the processing procedure to step SA1. When the end of the triggerless voice recognition is instructed (step SA14: YES), the first voice recognition unit 10 ends the process. In the flowchart of FIG. 5, for convenience of explanation, the first voice recognition unit 10 is configured to determine whether or not the end of triggerless voice recognition is instructed in step SA14, but the first voice recognition unit 10 continuously monitors whether or not the end of triggerless speech recognition is instructed while the processing of the flowchart is being executed, and if instructed, executes the necessary end processing and then ends the processing. To do.

As shown in the flowchart of FIG. 6, the second voice recognition unit 12 determines whether or not the processing start notification has been input (step SB1). When the process start notification is input (step SB1: YES), the second voice recognition unit 12 starts the recognition process (step SB2). Next, the second voice recognition unit 12 starts measuring the elapsed time according to the start of the recognition process (step SB3). Next, the second voice recognition unit 12 determines whether or not the time J2 has elapsed since the start of the recognition process (step SB4), and determines whether or not the distance value is below the third threshold value T3 (step SB4). Step SB5).

When the time J2 elapses without the distance value falling below the third threshold value T3 (step SB4: YES), the second voice recognition unit 12 outputs the unrecognizable notification to the recognition result discard unit 13 (step SB6). Next, the second voice recognition unit 12 stops the recognition process (step SB7). After that, the second voice recognition unit 12 shifts the processing procedure to step SB11.

On the other hand, when the distance value falls below the third threshold value T3 before the time J2 elapses (step SB5: YES), the second voice recognition unit 12 recognizes that the internally generated voice corresponds to the latter half word (step SB8). ). Next, the second voice recognition unit 12 outputs the second voice recognition notification to the recognition result discard unit 13 (step SB9). Next, the second voice recognition unit 12 stops the recognition process (step SB10). After that, the second voice recognition unit 12 shifts the processing procedure to step SB11.

In step SB11, the second voice recognition unit 12 determines whether or not the end of triggerless voice recognition is instructed. When the end of the triggerless voice recognition is not instructed (step SB11: NO), the second voice recognition unit 12 shifts the processing procedure to step SB1. When the end of the triggerless voice recognition is instructed (step SB11: YES), the second voice recognition unit 12 ends the process. The flowchart of FIG. 6 is configured such that the second voice recognition unit 12 determines whether or not the end of triggerless voice recognition is instructed in step SB11 for convenience of explanation. However, the second voice recognition unit 12 continuously monitors whether or not the end of triggerless speech recognition is instructed while the processing of the flowchart is being executed, and if instructed, executes the necessary end processing and then ends the processing. To do.

As shown in the flowchart of FIG. 7, the recognition result discarding unit 13 determines whether or not the first voice recognition notification has been input from the first voice recognition unit 10 (step SC1). The first voice recognition unit 10 continuously and repeatedly executes the process of step SC1 until the first voice recognition notification is input. When the first voice recognition notification is input (step SC1: YES), the recognition result discarding unit 13 performs the second voice recognition notification and the unrecognizable notification at a timing close in time to the first voice recognition notification input in step SC1. One of the above is input (step SC2).

Next, the recognition result discarding unit 13 determines whether or not the notification input in step SC2 is the second voice recognition notification (step SC3). When the notification input in step SC2 is the second voice recognition notification (step SC3: YES), the recognition result discard unit 13 outputs the recognition discard notification to the first voice recognition unit 10 (step SC4). After that, the recognition result discarding unit 13 shifts the processing procedure to step SC6.

When the notification input in step SC2 is not the second voice recognition notification (= unrecognizable notification) (step SC3: NO), the recognition result discard unit 13 outputs the in-type confirmation notification to the first voice recognition unit 10. (Step SC5). After that, the recognition result discarding unit 13 shifts the processing procedure to step SC6.

In step SC6, the recognition result discarding unit 13 determines whether or not the end of triggerless speech recognition is instructed. When the end of the triggerless voice recognition is not instructed (step SC6: NO), the recognition result discarding unit 13 shifts the processing procedure to step SC1. When the end of the triggerless voice recognition is instructed (step SC6: YES), the recognition result discard unit 13 ends the process. The flowchart of FIG. 7 has a configuration in which the recognition result discarding unit 13 determines whether or not the end of triggerless speech recognition is instructed in step SC6 for convenience of explanation. , While the flow chart process is being executed, it continuously monitors whether or not the end of triggerless speech recognition is instructed, and if instructed, executes the necessary end process and then ends the process.

As described in detail above, the voice recognition device 100 according to the present embodiment has a first voice recognition dictionary 20A in which the entire recognition target word is registered, and a second voice recognition in which only the latter half of the recognition target word is registered. It has a dictionary 20B. In addition, the voice recognition device 100 according to the present embodiment uses the first voice recognition dictionary 20A to perform voice recognition of externally input voice input from the microphone 200, and a second voice recognition dictionary. The 20B is provided with a second voice recognition unit 12 that performs voice recognition of the internally generated voice generated by the audio device 400, which is an in-vehicle device, before being output from the speaker. The first voice recognition unit 10 sequentially calculates the similarity (calculation of the distance value) in parallel with the sequential input of the external input voice, and when the calculated similarity becomes larger than the first level (calculated distance). When the value falls below the first threshold value T1), it is recognized that the externally input voice corresponds to the first half of the recognition target word. The first voice recognition unit 10 determines that the external input voice corresponds to the entire recognition target word when the calculated similarity becomes larger than the second level (when the distance value falls below the second threshold value T2). recognize. The second voice recognition unit 12 sets the first threshold value T1 when the similarity calculated by the first voice recognition unit 10 becomes larger than the first level (the distance value calculated by the first voice recognition unit 10 sets the first threshold value T1. When the recognition process is started (when it falls below) and the calculated similarity is greater than the predetermined level (when the calculated distance value falls below the third threshold T3), the internally generated voice is in the latter half of the recognition target word. Recognize that it is equivalent. Then, when the first voice recognition unit 10 recognizes that the external input voice corresponds to the entire recognition target word, and the second voice recognition unit 12 recognizes that the internally generated voice corresponds to the latter half of the recognition target word. , The recognition result by the first voice recognition unit 10 is discarded.

According to the above configuration, since the second voice recognition unit 12 is activated only when the first voice recognition unit 10 recognizes that the external input voice corresponds to the first half of the recognition target word, the second voice recognition unit 12 The processing load can be reduced as compared with the case where is constantly operating. Since the processing load is small, it is not necessary to limit the number of words to be recognized that are constantly waiting in triggerless speech recognition. Then, when the first voice recognition unit 10 recognizes that the externally input voice corresponds to the entire recognition target word, and the second voice recognition unit 12 recognizes that the internally generated voice corresponds to the latter half of the recognition target word. The recognition result by the first voice recognition unit 10 is discarded as erroneous recognition caused by the internally generated voice output from the speaker being input from the microphone 200. As a result, according to the present invention, the number of words to be recognized that are constantly waiting in triggerless speech recognition is not limited to a small number, and the deterioration of the response performance of other processes other than the speech recognition process is suppressed as much as possible. It is possible to suppress erroneous recognition due to the generated voice.

In the above-described embodiment, the recognition result discarding unit 13 recognizes that the externally input voice corresponds to the entire recognition target word by the first voice recognition unit 10, and the internally generated voice is generated by the second voice recognition unit 12. When it was recognized that it corresponds to the latter half of the recognition target word, the recognition discard notification was output to the first voice recognition unit 10, and the recognition result by the first voice recognition unit 10 was discarded. In this regard, the following configuration may be used. That is, in the recognition result discarding unit 13, the first voice recognition unit 10 recognizes that the external input voice corresponds to the entire recognition target word, and the second voice recognition unit 12 recognizes that the internally generated voice is the latter half of the recognition target word. If it is recognized that it corresponds to, the following processing is further executed. That is, whether or not the recognition result discarding unit 13 has the same second half of the recognition target word recognized by the first voice recognition unit 10 and the second half of the recognition target word recognized by the second voice recognition unit 12. Is determined. Then, when the latter half of the recognized recognition target word is the same, the recognition result discard unit 13 outputs a recognition discard notification to the first voice recognition unit 10 and discards the recognition result by the first voice recognition unit 10. To do. According to this configuration, the following effects are obtained.

That is, during the execution of the recognition process performed after the first voice recognition unit 10 recognizes that the external input voice corresponds to the first half of one recognition target word, the internally generated voice accidentally becomes one recognition target word. Is not entirely possible to contain the second half of other recognized words that are different. In this case, since the external input voice recognized by the first voice recognition unit 10 as corresponding to the first half of the recognition target word is not the voice emitted by the audio device 400, the recognition result of the first voice recognition unit 10 is discarded. must not. However, in the above-described embodiment, whether or not the latter half of the recognition target word recognized by the first voice recognition unit 10 and the second half of the recognition target word recognized by the second voice recognition unit 12 are the same. Since the determination is not performed, the recognition result of the first voice recognition unit 10 is discarded. On the other hand, according to the above configuration, the recognition result discarding unit 13 includes the latter half of the recognition target word recognized by the first voice recognition unit 10 and the second half of the recognition target word recognized by the second voice recognition unit 12. Only when is the same, the recognition result by the first voice recognition unit 10 is discarded. Therefore, during the execution of the recognition process performed after the first voice recognition unit 10 recognizes that the externally input voice corresponds to the first half of the recognition target word, the internally generated voice accidentally differs from the recognition target word. When the latter half of the recognition target word is included, it is possible to prevent the recognition result of the first voice recognition unit 10 from being discarded.

Further, in the above-described embodiment, a distance value having a value in the range of the value "0" to the value "1000" is used as the degree of similarity between the registered voice pattern and the input voice. However, as the degree of similarity, an index other than the distance value may be used. That is, existing techniques can be widely used for determining the degree of similarity.

In addition, the above embodiments are merely examples of embodiment of the present invention, and the technical scope of the present invention should not be construed in a limited manner. That is, the present invention can be implemented in various forms without departing from its gist or its main features.

10 1st voice recognition unit 12 2nd voice recognition unit 13 Recognition result discard unit 20 Dictionary storage unit 20A 1st voice recognition dictionary 20B 2nd voice recognition dictionary 100 Voice recognition device 200 Microphone

Claims (5)

It is a voice recognition device that constantly determines whether or not the voice input from the microphone corresponds to the word to be recognized.
A dictionary storage unit that stores a first voice recognition dictionary in which the entire voice pattern of the recognition target word is registered, and a second voice recognition dictionary in which the voice pattern of only the latter half of the recognition target word is registered.
When the similarity between the entire voice pattern of the recognition target word registered in the first voice recognition dictionary and the externally input voice input from the microphone is calculated, and the calculated similarity is greater than a predetermined level. In addition, the first voice recognition unit that recognizes that the external input voice corresponds to the entire recognition target word,
The similarity between the voice pattern of the latter half of the recognition target word registered in the second voice recognition dictionary and the internally generated voice before being output from the speaker generated by the in-vehicle device is calculated, and the calculated similarity is calculated. A second voice recognition unit that recognizes that the internally generated voice corresponds to the latter half of the recognition target word when the degree is greater than a predetermined level.
The first voice recognition unit recognizes that the external input voice corresponds to the entire recognition target word, and the second voice recognition unit recognizes that the internally generated voice corresponds to the latter half of the recognition target word. In this case, a recognition result discarding unit for discarding the recognition result by the first voice recognition unit is provided.
The first voice recognition unit sequentially calculates the similarity in parallel with the sequential input of the external input voice, and when the calculated similarity becomes larger than the first level, the external input voice is released. It recognizes that it corresponds to the first half of the recognition target word, and when the calculated similarity becomes larger than the second level, it recognizes that the external input voice corresponds to the entire recognition target word.
The second voice recognition unit is a voice recognition device characterized in that the recognition process is started when the similarity calculated by the first voice recognition unit becomes larger than the first level. The second voice recognition unit is characterized in that the recognition process is stopped when it recognizes that the internally generated voice corresponds to the latter half of the recognition target word, or when a predetermined time has elapsed from the start of the recognition process. The voice recognition device according to claim 1. The first voice recognition unit calculates a distance value as an index representing the similarity, and when the calculated distance value becomes smaller than the first threshold value, the external input voice is the first half of the recognition target word. When the calculated distance value becomes smaller than the second threshold value, it is recognized that the external input voice corresponds to the entire recognition target word.
The second voice recognition unit starts the recognition process when the distance value calculated by the first voice recognition unit becomes smaller than the first threshold value, and then calculates the distance value as an index indicating the similarity. The voice recognition according to claim 1 or 2, wherein when the calculated distance value becomes smaller than a predetermined threshold value, the internally generated voice is recognized as corresponding to the latter half of the recognition target word. apparatus. The above recognition result discarding part
The first voice recognition unit recognizes that the external input voice corresponds to the entire recognition target word, and the second voice recognition unit recognizes that the internally generated voice corresponds to the latter half of the recognition target word. In that case, whether or not the latter half of the recognition target word recognized by the first voice recognition unit and the latter half of the recognition target word recognized by the second voice recognition unit are the same. The voice recognition device according to any one of claims 1 to 3, wherein the recognition result by the first voice recognition unit is discarded when the judgment is made and the same is determined. It is a voice recognition method that constantly determines whether or not the voice input from the microphone corresponds to the word to be recognized.
The first voice recognition unit of the voice recognition device sequentially calculates the similarity between the entire voice pattern of the recognition target word registered in the first voice recognition dictionary and the external input voice sequentially input from the microphone. The first step and
The first voice recognition unit determines whether or not the similarity calculated in the first step is higher than the first level, and when it is determined that the similarity is higher than the first level, the external input voice is the recognition target. The second step of recognizing that it corresponds to the first half of the word,
When it is determined that the similarity calculated in the first step is higher than the first level, the third step of activating the second voice recognition unit of the voice recognition device and the third step.
At the same time, the first voice recognition unit sequentially calculates the similarity between the entire voice pattern of the recognition target word registered in the first voice recognition dictionary and the external input voice continuously input from the microphone. , The degree of similarity between the voice pattern of the latter half of the recognition target word registered in the second voice recognition dictionary by the second voice recognition unit and the internally generated voice before being output from the speaker generated by the in-vehicle device. And the fourth step of sequentially calculating
The first voice recognition unit determines whether or not the similarity calculated in the fourth step is higher than the second level, and when it is determined that the similarity is higher than the second level, the external input voice is the recognition target. The fifth step of recognizing that it corresponds to the whole word,
The second voice recognition unit determines whether or not the similarity calculated in the fourth step is higher than the predetermined level, and when it is determined that the similarity is higher than the predetermined level, the internally generated voice is the recognition target word. The sixth step, which is recognized as corresponding to the second half,
In the fifth step, the recognition result discarding unit of the voice recognition device recognizes that the external input voice corresponds to the entire recognition target word by the first voice recognition unit, and in the sixth step, the above. When the second voice recognition unit recognizes that the internally generated voice corresponds to the latter half of the recognition target word, it has a seventh step of discarding the recognition result by the first voice recognition unit. Voice recognition method.