JPH1049194A - Voice recognition device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To efficiently recognize voices by appropriately classifying the contents of a voice dictionary. SOLUTION: When a voice inputting is conducted in the voice recognition device, an uttering is generally conducted at a constant speed. Therefore, the uttering time is comparatively constant for a same word regardless of the speakers. Using this idea, a word dictionary is classified into plural independent dictionary groups against each word, which becomes the object of voice recognition, based on the information including the number of the characters of the word and a voice continuation time (a measurement is made for a general average value). If the dictionary is classified by the number of characters, a pattern matching section 165 specifies the number of characters of the voice inputted from a microphone 24 from the voice segment data (a voice continuation time) detected in the segment detection of a preprocessing section 161. Then, a successive pattern matching is conducted from the individual dictionary having the number of characters which is close to the number of the characters of the specified input voice with a priority.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech recognition device, and more particularly to a speech recognition device used as an input device in a vehicle navigation device.

[0002]

2. Description of the Related Art Speech recognition devices for recognizing speech spoken by humans as words have been put to practical use in various fields. This voice recognition device has been put to practical use as an input device for giving a voice instruction from a place where an instruction corresponding to various devices in a factory has been released. It is considered to be used as a voice input device when inputting. Such a speech recognition apparatus generally analyzes the frequency distribution of the speech to be recognized in advance in order to identify the input speech, and for example, analyzes the spectrum and time-series information of the fundamental frequency to obtain the characteristics of the input speech. It has a speech recognition dictionary that extracts as quantities and stores the patterns in association with each word.

When a voice to be recognized is input, the frequency pattern of the input voice is compared with the pattern of each word stored in the voice recognition dictionary by pattern matching, and the similarity for each word is calculated. I do. Next, the word having the highest calculated similarity (the word having the closest pattern) is recognized as the input voice, and the word is output. That is, the input voice is determined by checking which word pattern most closely matches the pattern of the frequency distribution of the input word.

[0004] A dictionary for speech recognition used in a speech recognition apparatus is usually composed of about 1000 words in view of the relationship with the normal matching processing time. If it is necessary to recognize more than 1000 words, it is necessary to prepare a plurality of dictionaries in which the words are divided for each group, switch the dictionaries using an application program, and perform matching. Is a problem.

[0005] A technology in which the voice recognition device is applied to a vehicle-mounted navigation device is a voice recognition device for in-vehicle information processing described in Japanese Patent Application Laid-Open No. 7-64480. In this voice recognition device, when recognizing an input word by comparing and collating with a vocabulary such as a place name or a facility name related to a display content of a map for a navigation device registered in a voice dictionary,
Even if the vocabulary registered in the speech dictionary becomes large, the speech recognition rate of the input words by speech is efficiently and quickly performed, and the probability of erroneous recognition by similar words is reduced.
For this purpose, this navigation device divides the registered contents of the voice dictionary into groups according to the region, and recognizes input words based on the distance to the current position of the vehicle obtained by the navigation device. The voice dictionary group to be used is determined with priority.

[0006]

However, in the speech recognition device described in the above publication, since the priority determination index of the speech dictionary is the current position, the distance from the current position to the position coordinates of the input word of the destination is determined. The more distant, the more times the voice dictionary is switched. Also, if the destination is a vast site such as a place name, the number of times the voice dictionary is switched may be small, but it is described only on a detailed map such as a city map such as a store or a private house. If you enter a destination that is not
It is necessary to sequentially switch the voice dictionary from a voice dictionary having a low level of detail to a voice dictionary having a high level of detail, and it takes time to search.

An object of the present invention is to provide a speech recognition device capable of efficiently recognizing speech by appropriately classifying the contents of a speech dictionary.

[0008]

According to the first aspect of the present invention, there is provided a word dictionary in which standard patterns of a plurality of words to be recognized are stored in a state where the number of characters can be distinguished, and a voice for inputting voice. Input means, a character number specifying means for specifying the number of characters in the voice input from the voice input means, and a word dictionary for selecting a standard pattern of words from the word dictionary according to the number of characters specified by the character number specifying means Selecting means, a feature extracting means for extracting a feature of the voice input from the voice input means, and a similarity between the feature extracted by the feature extracting means and the standard pattern selected by the word dictionary selecting means. The speech recognition apparatus comprises: a similarity calculation unit that calculates the similarity; and a determination unit that determines the input voice from the similarity calculated by the similarity calculation unit. To achieve the purpose. According to the second aspect of the present invention, a word dictionary storing standard patterns of a plurality of words to be recognized in a state where their voice durations can be distinguished, voice input means for inputting voice, and voice input means And a standard pattern of a word within a predetermined time interval between the voice duration detected by the voice duration detection means and a voice duration detected by the voice duration detection means. A word dictionary selecting means, a feature extracting means for extracting a feature of a voice input from the voice input means, a feature extracted by the feature extracting means, and a standard pattern selected by the word dictionary selecting means. The speech recognition apparatus includes: a similarity calculating unit that calculates the similarity of the speech recognition unit; and a determination unit that determines the input voice based on the similarity calculated by the similarity calculating unit. So it achieves the above object. In the invention according to claim 3, the standard pattern of a plurality of words to be recognized is
A word dictionary stored in a state where the number of characters can be distinguished,
Voice input means for inputting voice, character number specifying means for specifying the number of characters of the voice input from the voice input means, feature extracting means for extracting features of the voice input from the voice input means, A similarity calculating means for calculating the similarity between the feature extracted by the feature extracting means and the standard pattern stored in the word dictionary; and the similarity of each word calculated by the similarity calculating means, A weighting means for weighting according to the number of characters specified by the specifying means, and a determination means for determining the input voice from the similarity after weighting by the weighting means, the voice recognition device comprising: The above objective is achieved. According to the fourth aspect of the present invention, a word dictionary in which standard patterns of a plurality of words to be recognized are stored in such a manner that their voice durations can be distinguished, voice input means for inputting voice, and voice input means A voice duration detecting means for detecting a voice duration of the voice input from the voice input unit; a feature extracting means for extracting a feature of the voice input from the voice input means; and a feature extracted by the feature extracting means. A similarity calculating means for calculating a similarity with the standard pattern stored in the word dictionary; and a similarity calculated for each word calculated by the similarity calculating means, the voice detected by the voice duration detecting means. The speech recognition device includes: a weighting unit that performs weighting according to the duration; and a determination unit that determines the input voice from the similarity after weighting by the weighting unit. Te, to achieve the above purpose.

[0009]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the speech recognition apparatus according to the present invention will be described below in detail with reference to FIGS. (1) Overview of the Embodiment In the speech recognition apparatus of the present embodiment, even if the speaking speed varies in daily conversation, the voice is generally uttered at a constant speed when performing voice input. And that the utterance time is relatively constant regardless of the speaker. Utilizing the above points, for each word to be subjected to speech recognition, the word dictionary is divided into a plurality of individual words based on information such as the number of characters of the word and the speech duration (measure a general average value). Group (classify) in a dictionary. If the dictionaries are classified according to the number of characters, pattern matching is performed sequentially with priority from the individual dictionaries having the number of characters close to the number of characters of the input voice.

(2) Details of Embodiment FIG. 1 shows a system configuration when a voice recognition device according to one embodiment of the present invention is applied to a navigation device. This navigation device includes a calculation unit 10. The calculation unit 10 includes a display unit 11 including a display 11a functioning as a touch panel and an operation switch 11b provided around the display 11a, and a touch panel and a switch 1 of the display unit 11
A switch input class management unit 12 for managing the input from 1b is connected.

The switches 11b include various switches such as a switch for designating a menu screen for navigation, a switch for adjusting an air conditioner, and a switch for operating audio. When these switches are pressed, the corresponding menu screen is displayed on the display 11a.

The arithmetic unit 10 includes a current position measuring unit 13 and
The speed sensor 14, the map information storage unit 15, the voice recognition unit 16 in the present embodiment, and the voice output unit 17 are connected. The current position measuring unit 13 detects the current position where the vehicle is currently running or stopped by detecting coordinate data based on latitude and longitude. The current position measuring unit 13 has a GPS (G
(lobal Positioning System) A receiver 21, a beacon receiving device 20 for receiving position information from a beacon placed on the road, an azimuth sensor 22, and a distance sensor 23 are connected, and the current position measurement unit 13 receives information from these. Is used to measure the current position of the vehicle.

The azimuth sensor 22 is, for example, a geomagnetic sensor for detecting the terrestrial magnetism to determine the azimuth of the vehicle, a gas rate gyro or an optical fiber gyro for detecting the angular velocity of the vehicle and integrating the angular velocity to determine the azimuth of the vehicle. gyro,
A wheel sensor or the like is used in which left and right wheel sensors are disposed, and a displacement of the azimuth is calculated by detecting turning of the vehicle based on an output pulse difference (difference in moving distance). As the distance sensor 23, for example, various methods such as a method of detecting and counting the number of rotations of a wheel, or a method of detecting acceleration and integrating twice are used. Note that the GPS receiver 21 and the beacon receiving device 20 can perform position measurement independently,
In a place where reception by the PS receiver 21 or the beacon receiving device 20 is not possible, the current position is detected by dead reckoning navigation using both the direction sensor 22 and the distance sensor 23.

The map information storage unit 15 is, for example, a CDROM.
And the like. The map information storage unit 15 stores various data necessary for route search and route guidance, such as road data necessary for route search to the destination and map data for displaying the searched route on the display 11a. Have been. Also, the map information storage unit 1
5 stores destination data including names of various buildings and points that can be set as destinations such as public facilities, gas stations, parks, and the like, and coordinate data (latitude and longitude) indicating the positions. ing. The voice recognition unit 16 is connected to a microphone 24 to which voice is input. The audio output unit 17 includes an audio output IC 26 for outputting audio as an electric signal, a D / A converter 27 for digital-to-analog conversion of the output of the audio output IC 26, and an amplifier 28 for amplifying the converted analog signal. It has. A speaker 29 is connected to an output terminal of the amplifier 28.

The arithmetic unit 10 includes a CPU (central processing unit),
A ROM (Read Only Memory), a RAM (Random Access Memory), and the like are provided, and the CPU executes the programs stored in the ROM or the external storage device using the RAM as a working area, thereby implementing each of the above configurations. Is to be realized. That is, the operation unit 10
A map data reading unit 31 connected to the speed sensor 14 and the map information storage unit 15; a map drawing unit 32; a map management unit 33 that manages the map data reading unit 31 and the map drawing unit 32; 32, a screen management unit 34 connected to the display unit 11, an input management unit 35 connected to the switch input type management unit 12 and the voice recognition unit 16, and a voice connected to the voice output IC 26 of the voice output unit 17. Output management unit 36, communication management unit 38, and map management unit 3
3, a screen management unit 34, an input management unit 35, an audio output management unit 36, and an overall management unit 37 that manages a communication management unit 38. The communication management unit 38 can be connected to communication devices such as a car telephone, a PHS, and a mobile phone (not shown). In addition to ordinary telephone communication, multimedia communication such as facsimile communication and personal computer communication, and communication using ATIS. When various communications such as are performed, the communications are managed.

FIG. 2 is a block diagram showing the configuration of the speech recognition section 16 in FIG. As shown in the figure, the speech recognition unit 16 includes a preprocessing unit 161, a feature extraction unit 162, a word dictionary 163, a pattern matching unit 165,
A determination unit 166 is provided. The pre-processing unit 161 converts an audio signal input from the microphone 24 into a digital signal, and performs audio section detection, pre-emphasis (high-frequency emphasis), noise removal, and the like on the audio signal after the A / D conversion. Pre-processing is performed. The feature extraction unit 162
From the audio signal after the pre-processing performed by the pre-processing unit 162,
The feature of the voice is extracted.
The feature of the extracted voice is a word pattern of the word. Here, as the feature of the audio signal, for example, time-series information about a spectrum or a cepstrum obtained by fast Fourier transform (FFT) is used. The feature extraction unit 162 extracts a feature of the input speech by various analysis methods such as a multi-channel bandpass filter and a linear prediction analysis.

The word dictionary 163 stores standard patterns for all words to be subjected to speech recognition. This standard pattern is for speaker-independent recognition.
The feature of each word extracted by the same method as the speech analysis method by the feature extraction unit 162 is stored as a standard pattern. The words to be subjected to voice recognition include various designation keys displayed on the screen of the touch panel 11a and the contents of various switches of the switch 11b, destination names stored in the map information storage unit 15 and capable of setting destinations, and the like. It is.

FIG. 3 conceptually shows an example of the contents of the word dictionary 163. As shown in FIG. 3, the word dictionary 163 is classified into a plurality of individual dictionaries, and the number of words classified into each individual dictionary is 1000 words or less. Each individual dictionary is a two-letter word dictionary, a three-letter word dictionary,
.. Are classified according to the number of characters of each word. As described above, the words stored in each of the individual dictionaries include, as described above, various designation keys displayed on the touch panel 11a, such as "Hosei", "Golfu", "Yuenchi", and "Tamatekku" as the destination name. , "Toshimaen",
"Kiyomizu dera" and others. Each individual dictionary actually contains:
A standard pattern of the word and code information including a code string corresponding to the word are stored. As the code information of each word, the same code information as the code information of the destination name stored in the map storage unit 15 and the code information corresponding to the input content from the touch panel 11a or the like is used.

The pattern matching unit 165 compares the standard pattern of each word stored in the word dictionary 163 with the word pattern (feature) extracted by the feature extraction unit 162, and calculates the similarity between the two. It has become. The pattern matching unit 165 includes a pre-processing unit 161
The data (speech continuation time) of the speech section detected in is supplied. The pattern matching unit 165 specifies the number of characters of the uttered voice from the voice continuation time, and preferentially performs pattern matching from an individual dictionary having a number of characters close to the specified number of characters. As for the relationship between the duration of the voice and the number of characters in the word, a correspondence table is created by measuring the relationship between the two in advance. For example, a plurality of words are uttered a plurality of times by a plurality of persons for each of a plurality of words composed of three characters, and the distribution of the utterance time is measured. Similarly, the distribution of the utterance time is measured for words having other numbers of characters. From this measured value, a time zone in which the distribution of each word is large is extracted for non-overlapping, thereby creating a correspondence table of the voice duration and the number of characters. The created correspondence table can be referred to by the pattern matching unit 165.

The determination unit 166 is provided by the pattern matching unit 1
Based on the comparison result of 65, the content of the voice input from the microphone 24 is recognized, and code information corresponding to the recognized content (recognized word) is supplied to the input management unit 35 of the arithmetic unit 10. .

Next, the speech recognition operation in the speech recognition apparatus thus configured will be described. When a voice to be recognized is input from the microphone 24 to the voice recognition unit 16, the preprocessing unit 161 converts an analog signal of the input voice into a digital signal, and then detects a voice section, pre-emphasis, and noise removal. And so on. Preprocessing unit 1
61 supplies the preprocessed voice signal to the feature extraction unit 162 and also supplies voice section data (voice duration) obtained by voice section detection to the pattern matching unit 165.

The characteristic extracting section 162 analyzes the supplied audio signal to extract the characteristic of the input voice.
Then, the extracted feature is supplied to the pattern matching unit 165 as a word pattern for the input voice. In the pattern matching unit 165, the feature extraction unit 16
From 2 to the time when the word pattern is supplied, the number of characters of the input voice corresponding to the voice duration supplied from the preprocessing unit 161 is specified by the voice duration-character number correspondence table. Then, the pattern matching 165 selects an individual dictionary having a corresponding number of characters from among the individual dictionaries of the word dictionary 163. Then, when the word pattern is supplied, it is compared with each of the standard patterns in the selected individual dictionary, and the similarity for each word is calculated.
To supply.

The determination unit 166 sorts the supplied words in descending order of the degree of similarity, and on the condition that the degree of similarity exceeds a predetermined threshold value, determines a predetermined number of high-order words having a high degree of similarity. Take out N pieces. Then, the word having the highest similarity is set as the recognition word for the input voice, and the other words are set as the next candidates in descending order of the similarity, and the code information corresponding to each word is supplied to the input management unit 35 of the control unit 10.

If the number of words whose similarity supplied from the pattern matching unit 165 exceeds a predetermined threshold value is not N, the determination unit 166 sends the pattern matching unit 165 another individual word. Request the pattern matching again for the dictionary. Pattern matching unit 16
5 receives this request, switches to the individual dictionary whose voice duration is the next closest, and performs pattern matching again. For example, after performing pattern matching for the individual dictionary of five characters from the voice duration first, and then receiving a pattern matching request again from the determination unit 166, the next four voices or six characters next to the voice duration Perform pattern matching for the individual dictionary. Here, regarding the selection of four characters and six characters, six characters are selected when the audio duration is longer than the time at the center of the time range of five characters, and four characters are selected when the voice duration is short. The determination unit 166 selects the top N items exceeding the threshold value from the similarities already supplied and the similarities newly supplied by the pattern matching again, and inputs the code information to the input management unit 35. To supply.
If there are no N words exceeding the threshold value, pattern matching is requested again until N words exceed the threshold value.

When the code information corresponding to the recognized word is supplied to the input management unit, the calculation unit 10 controls the overall management unit 37.
However, by performing answerback by voice from the speaker 29 via the voice output management unit 36, the recognized voice is confirmed. Alternatively, a predetermined number of N words corresponding to the supplied code information is displayed on the display 11a, and the user selects the word, thereby identifying the recognition voice.

As described above, according to the present embodiment,
Paying attention to the fact that the utterance time when performing voice input is relatively constant, each word to be subjected to voice recognition is classified into an individual dictionary with the number of characters, and the duration of the input voice is used to determine the input voice. Since the number of characters is specified and the pattern matching is preferentially performed from the individual dictionaries having the number of characters close to the specified number of words, it is possible to appropriately select and switch the voice dictionary and reduce the recognition time. In addition, since appropriate classification and selection of the individual dictionaries are performed, the recognition rate can be improved.

In the embodiment described above, only one of the preferred embodiments of the present invention has been described.
The present invention can be variously modified within the scope of the invention described in the claims. For example, in the embodiment described above, pattern matching is sequentially performed from the individual dictionaries having the number of characters close to the number of characters corresponding to the voice continuation time. Although the pattern matching for the individual dictionaries is terminated, the present invention may alternatively perform pattern matching for all the individual dictionaries. in this case,
The similarity obtained as a result of pattern matching for words in each individual dictionary may be weighted according to the number of characters specified from the duration of the input voice. For example, if the input voice is specified to be five characters, the words “Yuenchi”, “Toshimaen”,.
Is weighted the most to the degree of similarity to. Next, the next largest weight is assigned to words in the individual dictionary of 4 and 6 characters that differ by 1 character. In addition, it is possible to arbitrarily select to which range (up to how many characters different characters) are to be weighted. As the weighting, it is possible to arbitrarily select whether to add a predetermined value or multiply by a predetermined coefficient, and also to add or multiply the value.

Further, in the above-described embodiment, the description has been made on the assumption that the number of circuits or chips for performing pattern matching is one, but in the present invention, a plurality of circuits or chips may be arranged. For example, a dedicated circuit or chip for pattern matching may be arranged for each of the classified individual dictionaries. In this case, weighting is performed according to the number of characters specified from the duration of the input voice. With such a configuration, high-speed recognition and a high recognition rate can be obtained.

Further, in the described information, the word dictionary 163
Are classified into individual dictionaries according to the number of characters of a word to be recognized. However, in the present invention, the classification may be performed according to the utterance time of each word. In other words, for all the words to be recognized, the utterance time (speech duration) by multiple people
Is measured, and the individual dictionaries are classified for each average time. For example, the voice continuation time is classified into an individual dictionary of the order of 0.1 seconds, an individual dictionary of the order of 0.2 seconds, an individual dictionary of the order of 0.3 seconds, ..., an individual dictionary of the order of m seconds, ... Individual dictionaries, not the number of characters,
The recognition rate can be further increased by using the speech duration obtained from the average of the actual utterance times of the recognition target words.

In the embodiment described above, as shown in FIG. 3, each word is classified into an individual dictionary according to the number of characters. However, the standard pattern data and the code of each word are classified without using the individual dictionary. In addition to the information, voice duration information of the word or character number information of the word may be stored. In this case, before performing the pattern matching, the pattern matching unit 165 selects, from all words in the word dictionary, the speech duration of the input speech supplied from the preprocessing unit 161 or the word corresponding to the number of characters specified from this. And then perform pattern matching.

In the embodiment described above, all functions of the speech recognition device are applied to the navigation device. However, in the present invention, part or all of the speech recognition device is arranged in another device outside the navigation device. You may do it. As another device, it is desirable to be an information providing station that provides information on a traveling route to a destination and the like to a vehicle by communication. In the information providing station, a word dictionary 163 classified into individual dictionaries at least for each number of characters or for each voice duration and a pattern matching unit 165 are arranged. The preprocessing unit 161, the feature extraction unit 162, It is preferable to arrange the entire voice recognition unit 16 including the determination unit 166 in the information providing station in order to reduce the device configuration on the navigation device side. When the entire voice recognition unit 16 is arranged at the information providing station, the voice of the destination or the like is input from the navigation device, and is transmitted from the car telephone or the like to the information providing station via the communication management unit 38. The information providing station performs pre-processing, feature extraction, pattern matching, and determination on the received voice, and transmits the code information for the top N recognized words having similarities to the navigation device by communication. The pattern matching processing and the determination processing by the information providing station may be the method described in the above embodiment or any of the methods described in the modified examples. In the navigation device, the communication management unit 38
This code information is received via the PC and the destination is set. The information providing station may search for a route to the destination based on the destination obtained by voice recognition, and transmit information on the searched route to the navigation device.

[0032]

According to the present invention, since the contents of the speech dictionary are appropriately classified, speech can be recognized efficiently.

[Brief description of the drawings]

FIG. 1 is a configuration diagram when a voice recognition device according to the present invention is applied to a navigation device.

FIG. 2 is a configuration diagram of a voice recognition unit.

FIG. 3 is an explanatory diagram conceptually showing an example of the contents of a word dictionary.

[Explanation of symbols]

 Reference Signs List 10 arithmetic unit 11 display unit 11a display 13 current position measurement unit 15 map information storage unit 16 speech recognition unit 161 preprocessing unit 162 feature extraction unit 163 word dictionary 165 pattern matching unit 166 determination unit 17 audio output unit 24 microphone 33 map management unit 34 Screen management unit 35 Input management unit 37 Overall management unit 38 Communication management unit

──────────────────────────────────────────────────続 き Continuation of the front page (51) Int.Cl. ⁶ Identification number Agency reference number FI Technical display location // G01C 21/00 G01C 21/00 H G09B 29/10 G09B 29/10 A

Claims (4)

[Claims] 1. A word dictionary in which standard patterns of a plurality of words to be recognized are stored in such a manner that the number of characters can be distinguished, voice input means for inputting voice, and voice input from the voice input means. Character number specifying means for specifying the number of characters of the word, word dictionary selecting means for selecting a standard pattern of words from the word dictionary according to the number of characters specified by the character number specifying means, and voice input from the voice input means A feature extraction unit for extracting a feature of the feature extraction unit; a similarity calculation unit for calculating a similarity between the feature extracted by the feature extraction unit and the standard pattern selected by the word dictionary selection unit; Determining means for determining the input voice from the similarity calculated by the means. 2. A word dictionary in which standard patterns of a plurality of words to be recognized are stored in such a manner that their voice durations can be distinguished, voice input means for inputting voice, and voice input means. Voice duration detecting means for detecting a voice duration of voice, and a word dictionary selection for selecting a standard pattern of a word within a predetermined time interval from the voice duration detected by the voice duration detecting means from the word dictionary. Means, a feature extracting means for extracting a feature of the voice input from the voice input means, and a similarity between the feature extracted by the feature extracting means and the standard pattern selected by the word dictionary selecting means. A voice characterized by comprising: a similarity calculation means for calculating; and a determination means for determining an input voice from the similarity calculated by the similarity calculation means. Recognition device. 3. A word dictionary in which standard patterns of a plurality of words to be recognized are stored in such a manner that the number of characters can be distinguished, voice input means for inputting voice, and voice input from the voice input means. A character number specifying unit for specifying the number of characters of the character string; a characteristic extracting unit for extracting a characteristic of the voice input from the voice input unit; a characteristic extracted by the characteristic extracting unit; and a standard stored in the word dictionary. A similarity calculating means for calculating a similarity with the pattern; a weighting means for weighting the similarity of each word calculated by the similarity calculating means in accordance with the number of characters specified by the character number specifying means; A speech recognition device comprising: a determination unit configured to determine an input voice from the similarity after weighting by the weighting unit. 4. A word dictionary in which standard patterns of a plurality of words to be recognized are stored in a state in which their voice durations can be distinguished, voice input means for inputting voice, and voice input means. Voice duration detection means for detecting the voice duration time of voice; feature extraction means for extracting features of voice input from the voice input means; features extracted by the feature extraction means; A similarity calculating means for calculating the similarity with the standard pattern stored in the memory; and a similarity of each word calculated by the similarity calculating means, in accordance with the voice duration detected by the voice duration detecting means. Weighting means for performing weighting, and determination means for determining an input voice from the similarity after weighting by the weighting means. Identification equipment.