JP2006349871A - Device for estimating driver's burden - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a device for estimating driver's burden that estimates the burden of interaction with free utterance on a driver. <P>SOLUTION: The device which is mounted on a vehicle and estimates the burden on the driver who is driving the vehicle is equipped with a speech gathering means of gathering a speech in the cabin, a speech recognizing means of recognizing the gathered speech in word units, a calculating means of calculating appearance frequencies of nouns and verbs of the recognized words in every unit utterance time, respectively, and an interaction burden estimating means of using the sum of the calculated appearance frequencies of the nouns and verbs in every unit utterance time as a burden parameter representing the value of the burden of interaction on the driver. The interaction burden estimating means makes specified corrections of the appearance frequencies of the nouns and verbs in every unit utterance time according to reliabilities and appearance intervals of the respective recognized words. The speech gathering means includes an utterance source specifying means of specifying utterance sources of the gathered speech, and the interaction burden estimating means discriminates between words spoken by the driver and words heard by the drivers according to the specified utterance sources. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention generally relates to a driver load estimation device that is mounted on a vehicle and estimates a driver's load while driving the vehicle, and more particularly, to a driver load estimation device that estimates a driver's interaction load with respect to free speech.

Conventionally, an apparatus and a method for estimating a driver's load while driving a vehicle have been proposed (see, for example, Patent Documents 1 to 4).
JP 2003-125454 A JP 2003-150193 A JP 2004-228890 A JP 2004-272048 A

  However, none of the conventional methods disclosed in the above-mentioned Patent Documents 1 to 4 realizes accurate driver load estimation because the driver's interaction load regarding free speech is not considered or estimated.

  For example, the methods disclosed in Patent Documents 1 and 3 do not consider any interaction load.

  In the method disclosed in Patent Document 2, only a cognitive load is considered, and free dialogue is not considered at all.

  Furthermore, the technique disclosed in Patent Document 4 performs load estimation based on the utterance pitch, and it is difficult to say that the accuracy is good.

  The present invention has been made to solve the above-described problems, and a main object of the present invention is to provide a driver load estimating device that estimates the driver's interaction load with respect to free speech.

  In order to achieve the above object, a first aspect of the present invention is a driver load estimation device that is mounted on a vehicle and estimates a driver's load during driving of the vehicle. Means, speech recognition means for performing speech recognition processing on the speech collected by the speech collection means in units of words, and out of words recognized by the speech recognition means, the number of appearances per unit utterance time of nouns and verbs, respectively. Driving having calculation means for calculating, and conversation load estimation means using the sum of the number of appearances per unit utterance time of the noun and verb calculated by the calculation means as a load parameter representing the magnitude of the driver's conversation load It is a person load estimation device.

  In the first aspect, the voice recognition means performs voice recognition processing on a word-by-word basis, for example, by executing so-called n-gram type voice recognition.

  According to this first aspect, specializing in nouns and verbs that are considered to have a particularly high speech load and listening load compared to other words such as conjunctions and exclamations, the number of appearances per unit time Since the driver's interaction load is estimated, accurate estimation is possible.

  The second aspect of the present invention for achieving the above object is that, in the driver load estimating device according to the first aspect, the dialogue load estimating means determines the number of appearances per unit utterance time of the noun and verb. It is a driver | operator load estimation apparatus containing the correction | amendment means which performs this weighting process.

  In this second aspect, the correction means, for example, the appearance of the noun and the verb per unit utterance time so that the load parameter increases or decreases according to the reliability and the appearance interval of each recognized word. Weight the number of times.

  More specifically, for example, 1) The reliability is calculated for each word further processed by the voice recognition means, and the nouns and verbs with low reliability are words that do not appear much in normal conversation. The load parameter may be weighted so as to increase, or 2) the calculation means further measures the appearance interval for each word recognized by the voice recognition means, and this For nouns and verbs whose appearance interval is shorter than a predetermined time (that is, repeated frequently), weighting is performed so that the load parameter decreases by judging that both the speech load and the listening load are low, and the appearance interval is longer than the predetermined time. For nouns and verbs that are long (that is, only occasionally emitted), it is determined that both the utterance load and the listening load are high, and the load parameters May be weighted so that other increases.

  According to this second aspect, nouns and verbs recognized by voice recognition are further corrected to increase or decrease the load parameter according to their reliability and appearance interval. The accuracy can be further improved.

  According to a third aspect of the present invention for achieving the above object, in the driver load estimating device according to the second aspect, a voice source specifying means for specifying the voice source of the voice collected by the voice collecting means is provided. The dialogue load estimating means is listened to by the driver and the words spoken by the driver and the number of appearances per unit utterance time of the noun and verb according to the utterance source specified by the utterance source specifying means. This is a driver load estimation device that calculates and divides into different words.

  In this third aspect, the utterance source specifying means determines whether the utterance is from a hands-free call based on, for example, a hands-free call function installed in the vehicle or an operating state of a mobile phone connected thereto. Or by detecting the arrival direction of the utterance using the microphone array, it is determined whether the utterance is the passenger's utterance.

  In this third aspect, the identification result by the utterance source identification unit is used for, for example, a voice recognition process by the voice recognition unit or a weighting process by the correction unit.

  More specifically, for example, 1) Based on the insight that the passenger will talk to the driver in consideration of the driver's condition and vehicle driving situation rather than the driver and the other party of the hands-free phone When the utterance source identifying means identifies whether the utterance to the driver is a utterance from a hands-free telephone or a fellow passenger, and the correction means is hands-free when the number of appearances per unit utterance time of the noun and verb is equal The number of appearances per unit utterance time of the nouns and verbs may be weighted so that the load parameter increases in the utterance from the telephone than the utterance from the passenger, and / or 2 ) Based on the insight that the driver will be able to listen to audio output such as music and radio and concentrate on driving operation, the voice recognition means will Speech that have been identified as the audio output may not be the speech recognition process.

  According to the third aspect, since the driver's listening load can be estimated in detail according to the utterance source for the utterance voice to the driver, the accuracy of the driver interaction load estimation is further improved.

  ADVANTAGE OF THE INVENTION According to this invention, the driver load estimation apparatus which estimates the driver | operator's dialog load with respect to free speech can be provided.

  Hereinafter, the best mode for carrying out the present invention will be described with reference to the accompanying drawings. Note that the basic concept of the speech recognition processing, the hardware configuration and the software structure for realizing the basic concept are known to those skilled in the art, and detailed description thereof will be omitted.

  Hereinafter, the configuration and operation of the driver load estimation device according to an embodiment of the present invention will be described with reference to FIGS. 1 and 2.

  FIG. 1 is a schematic configuration diagram of a driver load estimation device 100 according to the present embodiment. The driver load estimation device 100 mainly includes a voice collection unit 101, a voice recognition engine 102, and a dialogue load estimation unit 103.

  The sound collection unit 101 collects sound in the vehicle interior using, for example, a microphone, and detects and identifies the origin of the collected sound.

  Here, the voice in the passenger compartment can be roughly classified into A) driver utterance and B) driver listening depending on whether the utterance is made by the driver. A) Driver utterances can be further classified into a1) utterances to hands-free phones or a2) utterances to passengers (for example, passengers in the passenger seat). B) Listening to the driver is further b1) listening from a hands-free phone, b2) listening from a passenger (for example, a passenger in the passenger seat), or b3) listening to audio output such as music or radio. Can be classified.

  The voice collection unit 101 classifies the collected voice in the passenger compartment into any one of the above a1) to a2) and b1) to b3). The specific method for classifying may be any known method. For example, it is possible to determine whether or not the speech is from a hands-free call based on the hands-free call function installed in the vehicle and the operating state of a mobile phone connected thereto. Further, by detecting the direction of arrival of the utterance using a microphone array (for example, a beamform type or the like), it is possible to determine whether the utterance is a fellow passenger's utterance.

  Here, the microphone array is a sound receiving device composed of a plurality of spatially arranged microphones, and can detect the direction of arrival of sound by acquiring spatial information of sound.

  The voice collection unit 101 can also identify whether the voice is a driver's speech or a fellow passenger's speech by voiceprint authentication processing by registering the driver's voiceprint in the vehicle in advance.

  In addition, it is thought that the driver can listen to the driver and concentrate on the driving operation, that is, the driver's listening load does not occur. Therefore, the voice collection unit 101 not only identifies the audio output, but also immediately removes it from the collected voice using, for example, a LMS (Least-Mean-Squares) algorithm.

  The voice collection unit 101 also measures the utterance time for each utterance collected and classified in this way.

  The voice collection unit 101 sends the collected voice information to the voice recognition engine 102 and sends information related to the utterance time and the utterance source to the dialogue load calculation unit 103.

  The speech recognition engine 102 performs speech recognition processing that sequentially outputs characters to the uttered speech collected by the speech collection unit 101, and recognizes each uttered word. Here, as an algorithm capable of performing speech recognition processing in units of words, for example, n-gram (engram) type speech recognition can be cited. n-gram is a representative example of a statistical language model that treats a word string statistically and probabilistically. A sentence actually spoken by calculating (learning) the probability that n words are linked in advance. It is possible to calculate the reliability (likelihood) of.

  Since the main purpose of this device is not to interpret the semantic content of the utterance text but to estimate the driver's interaction load, the speech recognition engine 102 determines the part of speech of each recognized word after the speech recognition processing. The reliability is calculated, and these are sent to the dialogue load estimation unit 103.

  The dialogue load estimating unit 103 determines the driver's dialogue load based on the information and the utterance time regarding the utterance source acquired from the voice collecting unit 101 and the part of speech and reliability of the recognized word acquired from the voice recognition engine 102. presume. The processing in the conversation load estimation unit 103 will be described in detail in the description of FIG. 2 below.

  FIG. 2 is a flowchart illustrating a process flow of the driver load estimation device 100 according to the present embodiment.

  The apparatus 100 first monitors and stands by for voice input (S201). When a sound is input (“YES” in S201), it is determined whether or not it is an audio output (S202).

  If the input sound is an audio output such as music or radio (“YES” in S202), as described above, the data is discarded, for example, by using the LMS algorithm (S208).

  If the input voice is not an audio output, that is, if it is a human utterance (“NO” in S202), then the voice recognition engine 102 uses the n-gram type or the like to divide this voice into words. Voice recognition processing is performed (S203).

  The speech recognition engine 102 further determines the part of speech and the reliability for each recognized word (S204).

  When the part of speech and the reliability of each word recognized in this way are determined, the dialogue load estimating unit 103 then determines the speech time measured by the voice collecting unit 101 and the word recognized by the voice recognition engine 102. The number of appearances of each noun and verb per unit utterance time is calculated from the part of speech, and the sum of the number of appearances is set as a conversation load parameter indicating the magnitude of the conversation load (S205).

  Here, the reason for narrowing down to nouns and verbs is that the speaking load and listening load of nouns and verbs are considered to be particularly high compared to other words such as conjunctions and exclamations.

  Next, the conversation load estimation unit 103 corrects the conversation load parameter corresponding to the sum of the number of appearances of nouns and verbs per unit time with the reliability and the appearance interval of each recognized word (S206).

  Specifically, nouns and verbs with low reliability are determined to be words that do not appear much in normal conversation, and weighting processing is performed so that the dialogue load parameter increases.

  For nouns and verbs whose appearance interval is shorter than a predetermined time (that is, repeated frequently), it is determined that both the speech load and the listening load are low, and weighting processing is performed so that the conversation load parameter decreases. .

  Furthermore, for nouns and verbs whose appearance interval is longer than a predetermined time (that is, they are uttered only occasionally), it is determined that both the speech load and the listening load are high, and weighting processing is performed so that the conversation load parameter increases.

The increase / decrease correction based on the appearance interval is realized, for example, by multiplying the interaction load parameter by 1 / (1-e− ^τt ) as the appearance interval t.

  Next, the conversation load estimation unit 103 divides the conversation load parameter into the speech load and the listening load based on the utterance source information acquired from the voice collection unit 101. Further, the utterance to the driver who has been cared as the listening load is identified as the utterance from the passenger to the driver or the utterance from the other party of the hands-free telephone to the driver.

  Then, the conversation load estimation unit 103 is configured so that the conversation load parameter is increased in the utterance from the hands-free phone than the utterance from the passenger when the number of appearances per unit utterance time of the noun and the verb is equal. Weights the number of occurrences per verb utterance time. This is because the passenger is talking to the driver in consideration of the driver's condition and the driving situation of the vehicle rather than the call partner of the hands-free phone (for example, turning right and left at the intersection while paying attention to the pedestrian) It ’s based on the insight that sometimes it ’s not going to talk to you. As a result, the utterance from the passenger to the driver reflects the evaluation that the listening load is relatively lower than the utterance from the call partner of the hands-free phone.

  The dialogue load parameter after such a series of correction (weighting) processing is used for estimating the driver load as a final representation of the driver's dialogue load.

  As described above, according to the present embodiment, when speech recognition processing capable of recognition in units of words is used, the speech load and the listening load are particularly high compared to other words such as conjunctions and exclamations. Specializing in possible nouns and verbs, and estimating the driver's interaction load based on the number of appearances per unit time, accurate estimation is possible.

  Further, according to this embodiment, the noun and the verb recognized by the speech recognition are further weighted to increase / decrease the interaction load parameter according to the reliability and appearance interval of each, so that the driver interaction load estimation is performed. The accuracy can be further improved.

  Furthermore, according to the present embodiment, it is possible to precisely estimate the driver's listening load according to the utterance source for the utterance voice to the driver, so that the accuracy of the driver interaction load estimation is further improved.

  INDUSTRIAL APPLICABILITY The present invention can be used in a driver load estimation device that is mounted on a vehicle and estimates a driver's load while driving the vehicle. The appearance, weight, size, running performance, etc. of the vehicle to be mounted are not limited.

It is a schematic block diagram of the driver | operator load estimation apparatus which concerns on one Example of this invention. It is a flowchart which shows the flow of a process of the driver | operator load estimation apparatus which concerns on one Example of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 Driver load estimation apparatus 101 Speech collection part 102 Speech recognition engine 103 Dialogue load estimation part

Claims (8)

A driver load estimation device that is mounted on a vehicle and estimates a driver's load while driving the vehicle,
A voice collecting means for collecting voice in the passenger compartment;
Voice recognition means for performing voice recognition processing on a voice basis by the voice collected by the voice collection means;
Of the words recognized by the speech recognition means, calculating means for calculating the number of appearances per unit utterance time of nouns and verbs,
A dialogue load estimating means that uses a sum of the number of appearances per unit utterance time of the noun and verb calculated by the calculation means as a load parameter that represents the magnitude of the driver's dialogue load. Person load estimation device. The driver load estimating device according to claim 1,
The said driver | operator load estimation means contains the correction | amendment means which performs a predetermined weighting process to the frequency | count of appearance per unit utterance time of the said noun and a verb, The driver | operator load estimation apparatus characterized by the above-mentioned. The driver load estimating device according to claim 2,
The speech recognition means further calculates a reliability for each word subjected to speech recognition processing,
The correcting means weights the number of appearances per unit utterance time of the noun and verb so that the load parameter increases for the noun and verb with low reliability, the driver load estimating device, . The driver load estimation device according to claim 2 or 3,
The calculation means further measures an appearance interval for each word recognized by the voice recognition means,
The correction means weights the number of appearances per unit utterance time of the noun and verb so that the load parameter decreases for the noun and verb whose appearance interval is shorter than a predetermined time. Load estimation device. The driver load estimation device according to any one of claims 2 to 4,
The calculation means further measures an appearance interval for each word recognized by the voice recognition means,
The correction means weights the number of appearances per unit utterance time of the noun and verb so that the load parameter increases for the noun and verb whose appearance interval is longer than a predetermined time. Load estimation device. The driver load estimating device according to any one of claims 2 to 5,
The voice collecting means includes a voice source specifying means for specifying a voice source of the collected voice,
The dialogue load estimating means includes a word uttered by the driver and a word heard by the driver based on the utterance source specified by the utterance source specifying means, and the number of appearances per unit utterance time of the noun and verb. A driver load estimation device characterized by being calculated separately. The driver load estimating device according to claim 6, wherein
The utterance source specifying means identifies whether the utterance to the driver is an utterance from a hands-free telephone or from a passenger,
The correction means is configured so that when the number of appearances per unit utterance time of the noun and the verb is equal, the load parameter is increased for the utterance from the hands-free phone than the utterance from the passenger. A driver load estimation device, characterized by weighting the number of appearances per unit utterance time. The driver load estimation device according to claim 6 or 7,
The driver recognition apparatus according to claim 1, wherein the voice recognition means does not perform voice recognition processing on the voice identified as the audio output by the voice source identification means.

Images