JP3058125B2 - Voice recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice recognition device which can be used for searching for a place name using voice in a car navigation system.

[0002]

2. Description of the Related Art A speech recognition technique is a technique for realizing a machine interface using speech as a more natural and easy input means. When performing voice input, a silent period that is semantically or physiologically short during the utterance (hereinafter, referred to as
Called a pose. ) Is often included. For this reason, conventionally, JP-A-6-259090 and JP-A-6-202
As disclosed in Japanese Patent Application Laid-Open No. 689, there is a speech recognition device that can easily confirm the recognition result by outputting the recognition result at the time of the pause of the utterance so that the speaker can input the sound more naturally. Was.

For example, in a speech recognition apparatus disclosed in Japanese Patent Application Laid-Open No. 6-259090 , in recognizing continuous speech, a pause is used as a unit of a sentence break, and when a pause is detected, the recognition result up to that point is obtained. Display and execution of application for it. When the utterance continues, the recognition is continuously performed, and when the next pose is detected, a new recognition result is displayed.

A speech recognition apparatus disclosed in Japanese Patent Application Laid-Open No. 6-202689 sets a pause time in each of portions where a pause may appear in information defining a word order for recognizing a voice. When the pause time during utterance exceeds the set time, the recognition result up to that time is presented.
At this time, it is possible to present the missing information based on the recognition result based on the recognition result, and to prompt the user to input again.

[0005]

Problems to be Solved by the Invention Japanese Patent Laid-Open No. 6-25909
The speech recognition device disclosed in Japanese Patent Publication No. 0 , for example, published in IEICE Transactions D Vo1.J71 No.9 pp1650-1659 published in July 1988 by "DP by integrating frame synchronization, beam search, and vector quantization
A method of recognizing continuous speech by connecting word patterns according to a grammar expressed by a finite state automaton as shown in "Speeding Up Matching" (hereinafter referred to as Reference Document 1). In this case, speech recognition is performed by searching for an optimum word string with all word patterns as recognition targets. By detecting the pose,
When recognizing a uttered voice after the recognition result is output, the entire grammar including the word already recognized at the time when the recognition result is obtained is processed as a search target, so the search range of the word Cannot be reduced. For this reason, even after the recognition result is obtained and confirmed, the search range remains large, and it is not expected that the amount of calculation and the recognition performance due to the obtained recognition result are improved. In addition, in the case of allowing the utterance of the restatement in the recognition result after the pause, all combinations of words including the restatement are searched, so that the search range becomes large, a large amount of calculation is required, and the recognition performance is increased. Has the disadvantage that it cannot be increased. Furthermore, in the case of using a method that uses an overall optimal word string as a recognition result as shown in Reference 1, a recognition result is output when a pause is detected, and the recognition result already confirmed. May change to a word that gives an optimal word string as a whole when all the utterances are finished.

In the speech recognition apparatus disclosed in Japanese Patent Application Laid-Open No. 6-202689, re-entry is prompted based on the result of recognition. For example, it is necessary to newly prepare a grammar for re-entry.

SUMMARY OF THE INVENTION An object of the present invention is to reduce the amount of computation, improve recognition accuracy, and improve recognition accuracy when uttering or rephrasing subsequent to a recognition result output at a sentence break. It is an object of the present invention to provide an easy-to-use speech recognition device in which the result does not change.

[0008]

[MEANS FOR SOLVING THE PROBLEMS] A speech recognition apparatus according to the first invention comprises:
Word information on the word to be recognized, information on the connection relationship between the words, one or more pieces of start point information at which the sentence may start, and a connection point 1 at which the sentence may be separated
A dictionary recording unit that holds dictionary information including one or more pause points, and performs recognition processing on input speech from one or more start points specified in advance based on the dictionary information, and performs recognition up to the pause points A speech recognition unit for outputting a result, and a start point control unit for designating a start point of a next recognition process corresponding to a pause point according to the recognition result are provided.

A speech recognition apparatus according to a second aspect of the present invention is the speech recognition apparatus according to the first aspect, wherein a pause point is set next according to a recognition result.
And a start point control unit for designating a start point for the recognition processing.

A speech recognition apparatus according to a third aspect of the present invention is the speech recognition apparatus according to the first aspect, wherein the speech recognition apparatus detects a pause point according to a recognition result .
Accordingly, a start point control unit for designating a start point of a next recognition process and a start point included in a connection point of a word from a pause point to the beginning of a sentence as a start point for the next recognition process. It is characterized by.

[0011]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A speech recognition apparatus according to the present invention is a method for recognizing continuous speech by connecting standard patterns in accordance with a grammar represented by a finite state automaton as shown in Reference Document 1, for example. Is controlled so that a speech can be efficiently input in the case of a subsequent utterance and a restatement utterance. Here, for a finite state automaton,
It is assumed that recognition can be started from any connected word, and the end of recognition at any word can be specified. In the speech recognition apparatus of the present invention, for example, the unit of recognition may be a section divided by a pause, which is a silent section of 300 milliseconds or longer as disclosed in JP-A-6-259090 . FIG. 3 is an example of connection of standard patterns according to a grammar expressed by a finite state automaton. Solid lines represent standard patterns in word units. The beginning of the standard pattern is defined as start points 401 to 404 at which recognition can be started, and the tail is defined as pause points 405 to 411 at which recognition can be completed. Each word is connected by connection information represented by a broken line.

Next, referring to FIG. 3, a description will be given of a method of recognizing the speech recognition apparatus of the present invention when the utterance after the pause is the continuation of the utterance before the pause, in comparison with the conventional method. . For example, if an utterance divided into a first utterance “Yokohama-shi, Kanagawa” and a second utterance “Kohoku-ku” is input by a pause, first, as a result of the recognition processing, the recognition result of the first utterance “Kanagawa-ken” Yokohama City ”and the corresponding pause point 40
6 is output. Subsequently, a process when the second utterance is input will be described. When performing recognition of a second utterance by a conventional speech recognition apparatus that performs recognition according to a finite state automaton, a method of designating all of the start points 401 to 404 as start points of recognition can be used, similarly to the first utterance. . Thus, by starting the recognition again, the second
Can recognize utterances. However, in this case, since it is necessary to always search all words, the amount of calculation does not decrease.

On the other hand, the start point control unit in the present invention first designates the start point 401 of the head of the sentence, and after the speech recognition unit outputs the above-described first utterance recognition result,
Based on the output information of the pause point 406, the start point 403 to which the pause point 406 is connected is designated as a new start point. As a result, only the grammar after the start point 403 is searched, so that the search range can be made smaller than before, and the amount of calculation can be reduced. further,
Since the number of recognition target candidates is reduced, the recognition performance may be improved. Further, in the present invention, the recognition result of the first utterance “Yokohama-shi, Kanagawa” indicates that the recognition result of the second utterance is “Kohoku-ku” or “Kanazawa-ku”. Stored unchanged. On the other hand, in a conventional speech recognition device that performs recognition according to a finite state automaton, the entire utterance of "Kohoku-ku, Yokohama-shi, Kanagawa" is converted to "Nakahara-ku, Kawasaki-shi, Kanagawa".
In the case where is close to, the recognition result of the first utterance, "Yokohama City, Kanagawa Prefecture" is changed to "Kawasaki City, Kanagawa Prefecture", so that the recognition result of "Yokohama City, Kanagawa Prefecture" is not utilized.

Next, referring to FIG. 3, a method of recognizing a speech recognition device according to a third invention of the present invention in the case where the utterance after the pause is a restatement of the utterance before the pause will be described. This will be described in comparison with the method. For example, if a pause after the first utterance “Yokohama-shi, Kanagawa” is input and “Kawasaki-shi”, which is a restatement of “Yokohama-shi”, is input in the second utterance, first,
Pause point 4 corresponding to the recognition result of utterance "Yokohama-shi, Kanagawa"
06 is output. Subsequently, when the second utterance is input, in the conventional method in which all of the start points 401 to 404 are designated as the start points of recognition, rephrasing can be recognized by searching for all words. It is possible. However, the amount of calculation does not decrease.

On the other hand, in the second start point control unit according to the present invention, the start point 101 at the head of the sentence is first designated as the start point of recognition, and the recognition result of the first utterance is output by the speech recognition unit. In this case, in addition to the start point 403 to which the pause point 406 is connected, the pause point 406
Start point 402 of the word which is a word string leading to
The start point 401 of the connection source word 2 is designated as a new recognition start point. As a result, it is not necessary to search for a word other than the designated start point such as a word starting from the start point 404, so that the search range can be made smaller than before. As a result, the amount of calculation can be reduced, and the recognition performance can be improved.

[0016]

Next, an embodiment according to the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing a first embodiment of the speech recognition apparatus of the present invention. The speech recognition apparatus according to the first embodiment includes a speech recognition unit 101, a dictionary recording unit 102,
And a first starting point control unit 103. The dictionary recording unit 102 records a standard pattern representing acoustic features of a plurality of registered words to be recognized, connection information, start point information, and pause point information of each word. As an example of the standard pattern of each word, the connection information, the pause information, and the start point information, the grammar represented by the finite state automaton as described above with reference to FIG. 3 will be used. Voice recognition unit 101
Detects a voice section and a pause from the input signal S using power information as shown in JP-A-6-259090 , and converts the voice section into a reference Continuous speech recognition is performed according to a grammar described by a finite state automaton as in Document 1. When a pause is detected, a word string, a corresponding pause point, and a start point of a connection destination of the pause point are output as a recognition result R. The first start point control unit 103 includes a voice recognition unit 101
Is set as the start point of the connection destination of the pause point in the recognition result R, based on the pause point in the recognition result R output from, and the set start point is used as the start point for recognizing the next utterance. Output as information A.

Next, the operation will be described. For example, it is assumed that the first utterance “Yokohama-shi, Kanagawa” and the second utterance “Kohoku-ku” are uttered by inserting a pause between utterances. First, the start point information A specifies a start point 401 at the beginning of a sentence. The voice recognition unit 101 outputs the first utterance “Yokohama City, Kanagawa Prefecture”
Input signal S is input, and the starting point 4 which is the starting point information A
Search for a word from 01. If a pose is detected,
The word string having the highest similarity “Yokohama-shi, Kanagawa”, the corresponding pause point 406, and the start point 40 to which the pause point 406 is connected
3 is output as the recognition result R. First starting point control unit 10
Reference numeral 3 denotes a pause point in the recognition result R when the recognition result R is input.
The start point 403 which is the connection destination of 06 is designated as the start point of the search for the word of the second utterance voice recognition, and is output as start point information A. From the start point 403, the speech recognition unit 101
Search for uttered words. Thus, the word “Kohoku-ku” of the recognition result R is obtained.

Hereinafter, a second embodiment of the present invention will be described with reference to the drawings. FIG. 2 is a block diagram illustrating a voice recognition device according to the second embodiment. The speech recognition apparatus according to the second embodiment includes a speech recognition unit 101, a dictionary recording unit 102, a second
And a start point control unit 201. The voice recognition unit 101 and the dictionary recording unit 102 are common to the first embodiment. The second start point control unit 201 refers to the dictionary recording unit 102 and determines a start point for recognizing the next utterance by the recognition result R.
The connection destination of the middle pause point, the start point of the word at the same pause point, the start point of the connection source word of the same pause point, and all the start points that have been traced back to the beginning of the sentence are represented by the start point information A. Output as

Next, the operation will be described. For example, it is assumed that the first utterance “Yokohama-shi, Kanagawa” and the second utterance “Kawasaki-shi” are uttered with a pause inserted between utterances. First, the start point information A specifies a start point 401 at the beginning of a sentence. The voice recognition unit 101 outputs the first utterance “Yokohama City, Kanagawa Prefecture”
Input signal S is input, and the starting point 4 which is the starting point information A
Search for a word from 01. If a pose is detected,
The word string having the highest similarity, “Yokohama City, Kanagawa Prefecture” and the corresponding pause point 406 are output as the recognition result R. The second start point control unit 201 refers to the dictionary recording unit 102, and refers to the start point 403 to which the pause point 406 in the recognition result R is connected and the start point 4 of the word “Yokohama” of the pause point 406.
02 and the starting point 401 of the word "Kanagawa", which is the connection source of the word "Yokohama" of the pause point 406, is designated as the starting point of the search for the second speech-recognition word, and is output as the starting point information A. I do. The speech recognition unit 101 searches for a word of the second utterance from the start points 401 to 403. As a result, the word “Kawasaki City” of the recognition result R is obtained.

The embodiment according to the present invention has been described above.
The speech recognition method that can be used in the above is described in Reference 1
The ATR Advanced Technology Series Automatic Translation Telephone edited by the International Telecommunications Research Institute, Inc. (hereinafter referred to as Reference 2) ) Of 20
３８38, a method using an HMM or the like can also be used. Further, the word connection information is not limited to the above, and can be expressed by a grammar such as a context-free grammar as shown on pages 49 to 68 of Reference 2. Further, in the above description, a section divided by a pause is used as a unit of recognition, but a word string extracted by a method such as word spotting may be used as a unit of recognition. In the second start point control unit 201 of the speech recognition apparatus according to the second embodiment of the present invention, all the start points are selected as the start point information A from the pause point of the recognition result R to the beginning of the sentence. For example, it is also possible to limit the number of copies to one set up to the previous word.

[0021]

As described above, according to the present invention, in order to recognize the next utterance after a pause or the like, the starting point or the starting point necessary for the recognition of the utterance following the recognition result before the pause is made. ,
By specifying the starting point required for restatement utterance recognition for the recognition result before the pause, the search range of words when performing utterance recognition after the pause is reduced. Performance can be improved. In addition, when performing subsequent recognition, the already confirmed recognition result does not change, so that an easy-to-use voice recognition device can be provided.

[Brief description of the drawings]

FIG. 1 is a block diagram for explaining a first embodiment of the present invention.

FIG. 2 is a block diagram for explaining a second embodiment of the present invention.

FIG. 3 is a diagram illustrating an example of a finite state automaton.

[Explanation of symbols]

 Reference Signs List 101 speech recognition unit 102 dictionary recording unit 103 first start point control unit 201 second start point control unit S input signal R recognition result A start point information

Continuation of front page (56) References JP-A-6-202689 (JP, A) JP-A-59-62899 (JP, A) JP-A-61-240296 (JP, A) JP-A-61-245198 (JP, A) JP-A-4-310000 (JP, A) JP-A-7-104782 (JP, A) JP-A-7-121192 (JP, A) JP-A-10-78961 (JP, A) 7-1437 (JP, B2) JP 7-43599 (JP, B2) Information Processing Society of Japan Research Report [Natural Language Processing] Vol. 91, No. 80, NL-85-7, “Use of speech information in sequential analysis” p. 49-56 (1991/9/20) Information Processing Society of Japan Research Report [Speech and Language Information Processing] Vol. 95, No. 51, SLP-6-5, “Language phenomena of natural language utterances and Japanese grammar for voice recognition” p. 27-34, (May 25, 1995) ICOT Research Bulletin TM-0489, "Method of estimating sentence structure of spoken dialogue sentences using prosodic information", p. 1-33, April 1988 ICOT research bulletin TM-1123, "Use of speech information in sequential analysis", p. 1-10, October 1991 (58) Fields investigated (Int. Cl. ⁷ , DB name) G10L 15 / 18 G10L 15/00 G10L 15/28 JICST file (JOIS)

Claims (7)

(57) [Claims] 1. A connection in which a sentence may be separated from word information on a word to be recognized, information on a connection relationship between the words, and information on one or more start points which are connection points where a sentence may start. A dictionary recording unit that holds dictionary information including information of one or more pause points that are points, and a recognition process for speech input from one or more start points specified in advance based on the dictionary information. A speech recognition unit that outputs a recognition result up to a pause point, and outputs a recognition result to the pause point according to the recognition result .
And a start point control unit for designating a start point of the next recognition process . 2. The speech recognition device according to claim 1, further comprising: a start point control unit that specifies the pause point as a start point for a next recognition process according to the recognition result. 3. A speech recognition apparatus according to claim 1, wherein a start point of a next recognition process corresponding to said pause point and a word from said pause point to the beginning of a sentence are determined in accordance with said recognition result. A speech recognition apparatus comprising: a start point control unit that specifies a start point included in a connection point as a start point for a next recognition process. 4. A start point for the next recognition process, wherein the start point control unit determines a start point coincident with the pause point and a start point included in a connection point of a word from the pause point to the beginning of a sentence. The speech recognition device according to claim 2, further comprising a start point control unit designated as: 5. The dictionary information includes word information including a voice pattern related to a word to be recognized, position information indicating a connection position of the word, and one or more connection points at which a sentence may start. 2. The speech recognition apparatus according to claim 1, comprising: dictionary information including start point information and one or more pause point information that is a connection point at which a sentence may be separated. 6. The speech recognition according to claim 1, wherein the start point corresponding to the pause point is a start point of a plurality of pieces of the word information which may be connected to one pause point of arbitrary word information. apparatus. 7. The start point according to claim 2, wherein the start point coincident with the pause point is a start point selected from a plurality of pieces of word information connected to one pause point of arbitrary word information. Voice recognition device.