JP2559031B2 - Voice recognition system - Google Patents

Description

DETAILED DESCRIPTION [Outline] In a speech recognition method for a specific speaker and a discrete word, a high recognition rate cannot always be obtained when recognition is performed only from speech features, but each word is prioritized based on frequency of use. The recognition rate is further improved by registering with a degree and performing recognition by adding the frequency of use to the characteristics of the voice at the time of recognition.

[Industrial applications]

The present invention relates to a voice recognition system, and more particularly to a voice recognition system capable of improving the recognition rate of voice recognition by discrete words of a specific speaker.

[Conventional technology]

Practical application of voice recognition technology has been advanced in many fields, but in particular, there is a strong demand for downsizing of the device and improvement of the recognition rate. Among the methods, the one with the highest recognition rate and suitable for miniaturization is being put to practical use, which is a speech recognition method using a specific speaker or a discrete word.

This speech recognition method causes a specific speaker who should recognize the speech to generate several discrete word speeches, analyzes the speech, extracts features, and registers them. The speech feature of the uttered discrete word is compared with the voice feature registered in advance, and the one determined to be the closest is used as the recognition result.

This speech recognition method generally provides a fairly high recognition rate, but it is still difficult to obtain a 100% recognition rate. Especially, it depends on the environmental conditions under which the recognition device is actually used or even when the same speaker changes the speech rate. The recognition rate decreases due to the fight. There is a limit to increasing the recognition rate from only the features of voice with this voice recognition algorithm,
Therefore, there is a demand for a voice recognition method that can obtain a higher recognition rate close to 100% by adding some other means.

FIG. 4 shows a conventional discrete word / speech recognition method for a specific speaker. In FIG. 4, reference numeral 1 denotes an audio input terminal. Reference numeral 2 denotes a feature extraction unit, which extracts a voice feature by a surrounding method such as dividing an input voice into a number of bands and performing frequency analysis, and forms a parameter for each band. The voice characteristic pattern extracted at the time of voice registration is sent to the standard pattern data storage unit 3 via the route a. The standard pattern data storage unit 3 is composed of a memory and stores data as a standard pattern by adding data such as numbers to the characteristic pattern of the input voice.

At the time of voice recognition, the feature of the voice input inputted to the voice input terminal 1 is similarly extracted by the feature extraction unit 2, and the extracted feature pattern is passed through the route b to the matching unit 4
Is added to The matching unit 4 uses a well-known DP (Dynamic Pr) for the input voice feature pattern and the standard pattern read from the standard pattern data storage unit 3.
The distance is calculated for each parameter by a technique such as matching, and the recognition result output terminal 5 outputs the recognition result output terminal 5 with a number indicating the standard pattern having the closest distance to the input pattern.

As described above, in the conventional speech recognition method using a specific speaker and a discrete word, recognition is performed only by information on the characteristics of the speech. Therefore, depending on the environment in which the voice recognition device is used, the voice information may be distorted and the feature extraction may not be performed correctly, and thus the voice recognition rate is unavoidably reduced. For example, when used in an office or the like, the recognition rate is greatly reduced due to the influence of ambient noise and the like. In addition to the environment of use, there are also errors caused by the cause of the recognition method itself, which causes a problem that there is a limit to the improvement of the recognition rate.

[Problems to be Solved by the Invention]

An object of the present invention is to provide a voice recognition method having a high voice recognition rate by auxiliaryly using voice frequency information in addition to voice information to determine a voice recognition result.

[Means for solving the problem]

As shown in the principle block diagram of FIG. 1, the feature pattern of the discrete word of the specific speaker is extracted by the feature extraction unit (2), and the extracted feature pattern is stored in the standard pattern data storage unit (3).
The collation unit (4) collates the characteristic pattern extracted from the speaker's voice with the characteristic pattern of the standard pattern data storage unit (3) and outputs the information indicating the closest word as the recognition result. In the voice recognition method, the standard patterns in the standard pattern data storage unit (3) are registered with priorities according to the frequency of use, and the frequency pattern determination unit (6) and the recognition determination unit (7) are provided. Get

The use frequency determination unit (6) determines the word with the highest priority in the standard pattern data storage unit (3) when there are two or more words determined to be close by the matching unit (4).

The recognition determination unit (7) outputs information indicating this when only one word is determined to be close in the collation unit (4) as a recognition result, and determines the frequency of use when there are two or more words. Information indicating a word whose priority is determined to be high in the section (6) is output as a recognition result.

In the data management based on the frequency of use in the standard pattern data storage unit (3), if the maximum number of stored pattern data is not registered, the newly registered pattern data is registered in the highest priority position, When the pattern data is registered, the lowest data is deleted and the newly registered pattern data is registered at the highest priority position. When the registered pattern data is used for recognition, one priority is set. It is characterized in that the pattern data used for the recognition is re-registered at the advanced position.

[Action]

When only one word is determined to be close to the registered standard pattern as a result of voice recognition, information indicating this is output as a recognition result.
When there are two or more, the word with the highest frequency of use is determined and the information indicating this is used as the recognition result, so voice recognition is performed by adding the frequency of use of the word to the voice information of the word, The recognition rate is improved.

〔Example〕

FIG. 2 shows an embodiment of the present invention, in which the same parts as in FIG. 4 are designated by the same numbers, and their operation is also the same. Reference numeral 6 is a usage frequency determination unit, 7 is a recognition determination unit, and includes a feature extraction unit 2, a standard pattern data storage unit 3,
The collation unit 4, the usage frequency determination unit 6, and the recognition determination unit 7 form a voice recognition unit 10. The voice recognition unit 10 exemplifies a voice recognition device in a telephone that dials by voice recognition. 11 is a processing unit (CPU), 12 is a selector, 13 is a switch, and 14 is a transmission path.

At the time of voice registration or voice recognition, the CPU 11 sets the selector 12 on the route a side. Thereby, the voice registration mode or the voice recognition mode is set.

At the time of voice registration, the uttered voice input is added to the feature extraction unit 2 from the voice input terminal 1, the features are extracted in the same manner as described with reference to FIG. 4, and the extracted feature patterns will be described later. The standard pattern is stored in the standard pattern data storage unit 3 as the standard pattern from the route c based on the designation of the priority order.

At the time of voice recognition, the feature pattern of the voice extracted by the feature extraction unit 2 is added to the matching unit 4 via the route of d, and each feature pattern registered in the standard pattern data storage unit 3 is read out by the route of e. And the number corresponding to the pattern with the smallest distance is sent to the recognition determination unit 7 through the route h. When there is only one candidate data obtained as a result of the matching in the matching unit 4, the recognition determining unit 7 sends this as a recognition result to the CPU 11 through the route k, and
11 decodes this data and, for example, dials the telephone number corresponding to this number, etc., and when connected to the other party, switches the selector 12 to the route b side and switches the selector 12 and the switch 13 to the route m. The voice input terminal 1 and the transmission line 14 are connected to each other so that the voice can be directly sent to the transmission line 14 so that the user can talk.

Further, when the result of the collation by the collating unit 4 is that the distance is small and there are two or more data to be candidates, the recognition determining unit 7 gives the information to the use frequency determining unit 6 through the route of i, and uses it. The frequency determination unit 6 makes an inquiry to the standard pattern data storage unit 3 through the routes f and g, determines the data with the highest priority in the standard pattern data storage unit 3 (that is, the data with the highest frequency of use), and makes the determination. The result is returned to the recognition determination unit 7 through the route of j. The recognition determination unit 7 passes this number or the like to the CPU 11 as a recognition result, and thereafter, processing such as dialing is similarly performed.

FIG. 3 illustrates data management based on the frequency of use in the standard pattern data storage unit 7.
Indicates the time of registration. (A) shows the case where the number of data items is less than the maximum number of items to be stored. When data items 1 to 4 are registered and there is data item A to be newly registered, the data item is registered at the highest priority position. It

(B) shows a case where the maximum number of stored data (for example 32) of data 1 to data 32 has already been stored. The newly registered data A is registered at the position with the highest priority and the lowest data 1 Will be deleted.

(2) indicates the time of recognition, and when data 1 to data 32 are registered, for example, when data 9 is used for recognition, it is indicated that the data is reregistered at a position with the priority increased by one. ing.

In this case, if the recognized one is immediately raised in priority, the priority will be disturbed and a malfunction will occur when a recognition error occurs, so wait until the recognition result is judged to be correct by the user. That is, when the dial is performed and the connection is correctly established, the CPU 11 controls the usage frequency determination unit 6 via the route 1 to raise the priority of the data.

In a telephone as shown in FIG. 2, a voice recognition dial is often used mainly for a person who is frequently used. According to the present invention, information on the frequency of use is used in addition to voice information. Since the recognition result is determined, it is possible to realize voice recognition with a higher recognition rate.

Moreover, since the standard data is managed by the method described in FIG. 3, there is little need for the user.
It also has the effect of automatically deleting data that is rarely used.

〔The invention's effect〕

As described above, according to the present invention, since the voice recognition result is discriminated by supplementarily using the information on the frequency of use in addition to the voice information, the recognition is performed using only the voice information as in the conventional case. A higher voice recognition rate can be obtained compared to the method.

[Brief description of drawings]

FIG. 1 is a diagram showing a basic configuration of a voice recognition system of the present invention, FIG. 2 is a block configuration diagram of a voice recognition system as an embodiment of the present invention, and FIG. 3 is a frequency of use in the voice recognition system of the present invention. FIG. 4 is a diagram for explaining data management based on FIG. 4, and FIG. 4 is a diagram showing a conventional specific speaker and discrete word speech recognition system. 1 ... Voice input terminal 2 ... Feature extraction unit 3 ... Standard pattern data storage unit 4 ... Verification unit 5 ... Recognition result output terminal 6 ... Use frequency determination unit 7 ... Recognition determination unit 10 ... Voice recognition Section 11 …… Processor (CPU) 12 …… Selector 13 …… Switch 14 …… Transmission path

 ─────────────────────────────────────────────────── ─── Continuation of front page (56) Reference JP-A-57-209575 (JP, A) JP-A-60-172100 (JP, A) JP-A-60-8899 (JP, A) JP-A-57- 66498 (JP, A) JP-A-62-206693 (JP, A) JIII Journal of Technical Disclosure No. 83-3134

Claims (1)

(57) [Claims] 1. A feature extraction unit for extracting a feature pattern of discrete words of a specific speaker, a standard pattern data storage unit for registering the extracted feature pattern, and a feature pattern extracted from a voice of the speaker. In a voice recognition method comprising a collating unit that collates with a characteristic pattern in the standard pattern data storage unit and outputs information indicating a word having the shortest distance as a recognition result, the standard pattern in the standard pattern data storage unit is used in frequency. The frequency of use in which the highest priority word is determined in the standard pattern data storage unit when there are two or more words determined to be close in distance by the matching unit When only one word is determined to be close in distance by the determination unit and the matching unit, information indicating this is output as a recognition result,
When there are two or more, a recognition determining unit that outputs information indicating a word whose priority is determined to be high in the use frequency determining unit as a recognition result, and data based on the use frequency in the standard pattern data storage unit. In the management, if the maximum stored number of pattern data is not registered, the newly registered pattern data is registered in the highest priority position, and if the maximum stored number of pattern data is registered, the lowest data is deleted. New registered pattern data is registered in the highest priority position, and when the registered pattern data is used for recognition, the pattern data used for the recognition is registered at a position which is moved up by one priority. A voice recognition method characterized by fixing.