JPS5934596A - Voice recognition processing system - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION (a) Technical field of the invention The present invention relates to a speech recognition processing method, and in particular, to a speech recognition processing system for all specific speakers. This relates to a modified method for performing the recognition process.

(b) Prior Art and Problems In recent years, devices that perform speech recognition processing for a specific speaker are being used as all input media such as telephones. In this case, the specific speaker utters a predetermined threatening word through a telephone line or the like, so that the speech recognition processing device registers the utterance pattern. Thereafter, words are extracted from the voice information by comparing and matching voice information input from a specific speaker with registered speech patterns, and are used in various processes.

The problem here is that if there are fluctuations in the noise on the telephone line or the ambient noise when a specific speaker speaks, the voice information pattern at the time of registration and recognition will differ, and the voice recognition processing device (this This is because the recognition rate decreases during the recognition process.

(c) Purpose of the invention In view of the above points, the present invention takes into account all noise patterns.
The purpose is to improve the recognition rate by executing speech recognition processing.

B) Structure of the Invention In order to achieve the above object, the present invention performs a complete comparison between the voice information input by a specific speaker and the voice information of words uttered by the specific speaker whose patterns are registered in advance in a pattern dictionary. In the speech recognition processing device configured to identify the words of the speech input by the specific speaker based on the comparison result, when registering the pattern in the pattern dictionary, in addition to the regular utterance word pattern, the speech recognition processing device is configured to identify the words of the speech input by the specific speaker. From now on, when recognizing the voice information input from the above-mentioned specific speaker, all patterns of the noise that are being input at that time are created prior to the voice recognition process, and the corresponding noise patterns are registered. Already ρ in the noise pattern and the above pattern dictionary
The present invention is characterized in that the registered noise pattern is compared with the registered noise pattern, and only when the distance between both noise patterns is smaller than a predetermined value, the speech recognition process of the particular speaker is executed.

(e) Embodiment of the invention FIG. 1 is a block diagram of a speech recognition processing device according to an embodiment of the invention, in which 1 is a feature extraction section, 2 is a pattern registration section, 3 is a dictionary section, and 4 is a collation section. The determination section 5 is a control section. FIG. 2 shows an example of the format of the dictionary section 3 shown in FIG. FIG. 3 is an example of the recognition processing flow in the first embodiment.

The operation of the embodiment is as follows.

(1) Pattern registration operation First, a specific speaker is connected to the voice recognition processing device shown in FIG. , the noise that is input at step 9 in the unvoiced state before voice input from a specific speaker is input to the total feature extraction unit II. The feature extraction unit 1 extracts the no(turn) of this noise.
8, send it to the pattern registration section 2. The pattern registration NI2 stores this noise pattern, for example, at the beginning position of the dictionary section 3 as shown in FIG.

After that, a message instructing the particular speaker to say all the words is sent by means not shown, and the particular speaker sequentially says all the predetermined words. The voice information uttered by a specific speaker is sequentially extracted by the feature extraction unit II, and then stored in the dictionary unit 3 under the control of the pattern registration unit 2, as shown in FIG. 2 patterns are stored in the form of...

(2) Recognition Processing Operation Notes Similarly to the pattern registration operation, when a specific speaker becomes ready for voice input using a telephone, etc., the voice recognition processing device recognizes ζ in the silent state before voice input from the specific speaker. The noise input here is input to the feature extraction unit H. The feature extraction unit 1 extracts this noise pattern and sends it to the matching determination unit 4. Criticism 4 - The determination unit 4 compares and matches the noise pattern sent from the feature extraction unit 1 with the noise pattern already registered in the dictionary unit 3, and determines the distance (difference) between the two noise patterns. is larger or smaller than a predetermined threshold. When the distance between both noise patterns is smaller than the threshold value, it is determined that recognition processing is possible, and a message instructing the specific speaker to utter all the words is sent by means (not shown).

Thereafter, the speech information from the specific speaker is recognized by a known recognition process using the feature extraction section 1, the comparison/determination section 4, and the dictionary section 3.

On the other hand, when the distance (difference) between the two noise patterns is greater than a predetermined threshold, a re-registration operation is performed as shown in the processing flow shown in FIG.

This re-registration operation is similar to the pattern registration operation described in item (1) above, and starts with the registration operation of the noise pattern when the specific speaker is in a silent state, and then utters and registers each word. It will be done. After this re-registration operation, a recognition processing operation is executed.

(F) Effects of the Invention According to the present invention, in a specific speaker speech recognition process, pattern fluctuations between the sound pattern generation and the subsequent recognition processing operation are reduced, resulting in an improvement in the recognition rate. becomes possible.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of a speech recognition processing apparatus according to an embodiment of the present invention, FIG. 2 is an example of a format i of a dictionary section, and FIG. 3 is an example of a recognition processing flow in the embodiment. In the figure, ■ is a feature extraction section, 2 is a pattern registration section, 3 is a dictionary section, and 4 is a format foot section. Figure 1

Claims (2)

[Claims] (1) The voice information input by a specific speaker is compared with the voice information of the words uttered by the specific speaker whose pattern is registered in the pattern dictionary in advance, and based on the comparison result, the voice information input by the specific speaker is In a speech recognition processing device configured to identify words in speech, when registering patterns in the pattern dictionary, it also registers noise patterns that are input when no utterance is made in addition to the regular uttered word patterns, and from now on, the above-mentioned When recognition processing is performed on voice information input from a specific speaker, a pattern of the noise that is being input at that time is created prior to the speech recognition processing, and the noise pattern and the noise pattern are already registered in the pattern dictionary. A speech recognition process characterized in that the speech recognition process is configured such that the speech recognition process of the specific speaker is executed only when the short distance between the two noise patterns is smaller than a predetermined value. method. (2) When the distance between the two noise patterns is greater than a predetermined value, a re-registration operation is performed for the specific speaker C, including all of the noise patterns. Speech recognition processing method.