KR100474253B1 - Speech recognition method using utterance of the first consonant of word and media storing thereof - Google Patents

Abstract

The present invention relates to a speech recognition method using a first consonant of a word and a recording medium storing the same. In particular, the speech recognition method includes a) receiving a voice signal corresponding to the first consonant of a target vocabulary and input by human speech. Doing; b) analyzing the received voice signal to recognize a corresponding consonant; c) reducing a vocabulary in a region divided according to the recognized consonants among all recognized vocabulary words into a recognized vocabulary; d) receiving a voice signal corresponding to a recognition target vocabulary input by human speech; And e) analyzing the voice signal received in step d) to perform the voice recognition on the reduced recognition target vocabulary in step c) and outputting a voice recognition result. According to the present invention, the recognition target vocabulary to be recognized at the same time can be significantly reduced, for example, by 1/19 on the average in inverse proportion to the number of first consonants of the recognition target word. In addition, the pronunciation of the Hangul consonants, such as a, b, c, etc., due to the nature of the pronunciation, the phoneme comes out twice for each consonant, there is an advantage in the speech recognition of the consonant.

Description

Speech recognition method using the first consonant of a word and a recording medium storing the same {SPEECH RECOGNITION METHOD USING UTTERANCE OF THE FIRST CONSONANT OF WORD AND MEDIA STORING THEREOF}

The present invention relates to speech recognition, and more particularly, to a speech recognition method for improving the success rate of speech recognition using a first consonant of a word and a recording medium storing the same.

In general, speech recognition extracts phonological and linguistic information from acoustic information included in speech, models a vocabulary registered in a speech vocabulary dictionary through a speech recognizer, and then finds and responds to the most similar data. It is a process.

Therefore, in the whole process of speech recognition, many and small words of recognition targets play an important part in determining recognition rate improvement and recognition speed.

The current speech recognition technology goes beyond the level of isolated word recognition for small vocabulary, and the research on continuous speech recognition for large vocabulary is active.

In this regard, the recognition rate of the speech is becoming a big problem as the large-scale speech vocabulary is recognized. This is because the recognition rate is relatively high when the number of words to be recognized is relatively high, but the recognition rate is markedly decreased as the number of words to be recognized is increased.

At present, the recognition rate of more than 1,000 words for speech recognition is announced as 90 ~ 95% in the refined laboratory environment, but in real life, 80% is due to the environmental noise of the surrounding environment and the channel noise in various communication networks and communication devices themselves. It is less than a level. There is a problem that successful commercial service is not possible at this level of recognition rate.

In addition, unlike the fixed word recognizer, the variable word recognizer can accommodate a vocabulary for adding and changing a recognized vocabulary. have.

Conventional technology for solving the above problems is disclosed in Korea Patent Publication No. 195-34054, characterized in that the recognition rate is improved by excluding words that are less likely to be recognition words from the target vocabulary recognition target vocabulary. In order to reduce the range of the recognition object by excluding a word that is unlikely to be a candidate word, a candidate word extracting unit must be added separately, and it takes a long time to exclude a word that is rarely a recognition word.

Accordingly, an object of the present invention is to solve the above-mentioned problem, so that the words to be recognized by the consonant to clearly know the consonant words, instead of targeting all the speech vocabulary only words that start with the consonant to the recognition target vocabulary The present invention provides a speech recognition method for improving recognition rate and recognition speed by performing recognition, and a recording medium storing the same.

In order to achieve the above object, the voice recognition method according to the characteristics of the present invention,

a) receiving a voice signal corresponding to the first consonant of the recognition target vocabulary and input by a human voice; b) analyzing the received voice signal to recognize a corresponding consonant; c) reducing a vocabulary in a region divided according to the recognized consonants among all recognized vocabulary words into a recognized vocabulary; d) receiving a voice signal corresponding to a recognition target vocabulary input by human speech; And e) analyzing the voice signal received in step d) to perform the voice recognition on the reduced recognition target vocabulary in step c) and outputting a voice recognition result.

Here, in step b), when the consonant is recognized, the phoneme of the first consonant voice is recognized, and in step c), the vocabulary area to be recognized for each vocabulary having the same first consonant as that of the first consonant voice recognized in step b) Characterized by.

At this time, the representative phonetic sounds of the first consonant vocalization is characterized in that seven of 'b', 'b', 'c', 'd', 'ㅁ', 'ㅂ' and 'o'.

In addition, the representative sound 'b' includes consonants' ㄱ 'and' ㅋ ', the representative sound' b 'includes a consonant' b ', and the representative sound' c 'includes a consonant' c 'and' ㅅ ',' ㅈ ',' ㅊ ',' ㅌ 'and' ㅎ ', the representative note' ㄹ 'includes a consonant' ㄹ ', the representative note' ㅁ 'includes' ㅁ', The representative note 'ㅂ' includes 'ㅂ' and 't', and the representative note '' is characterized by including 'ㅇ'.

According to another aspect of the present invention, a computer readable recording medium includes

A computer-readable recording medium tangibly embodying a program consisting of instructions executable by a computer to perform the steps of a speech recognition method,

The method may include: a) receiving a voice signal corresponding to the first consonant of the recognition target vocabulary and input by a human voice; b) analyzing the received voice signal to recognize a corresponding consonant; c) reducing a vocabulary in a region divided according to the recognized consonants among all recognized vocabulary words into a recognized vocabulary; d) receiving a voice signal corresponding to a recognition target vocabulary input by human speech; And e) analyzing the voice signal received in step d) to perform the voice recognition on the reduced recognition target vocabulary in step c) and outputting a voice recognition result.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram of a voice recognition device according to an embodiment of the present invention.

As shown in FIG. 1, a speech recognition apparatus according to an embodiment of the present invention includes a lexical dictionary 10 to be recognized and a speech recognizer 20.

In the recognition target vocabulary dictionary 10, a large-scale speech recognition target vocabulary is HMM modeled and stored. Here, the number of words to be recognized on a large scale means that at least 1000 words are used.

The vocabulary stored in the recognition target vocabulary dictionary 10 is stored to distinguish the area according to the first consonant.

As shown in the accompanying FIG. 2, representative voice values of the first consonant vocalization are seven kinds of 'a', 'b', 'c', 'ㄹ', 'ㅁ', 'ㅂ' and 'ㅇ'. It can be represented as. For example, the consonants corresponding to the case where the representative phonetic value is 'ㄱ' include 'ㄱ' and 'ㅋ', and the voice of 'ㄱ' is the 'read', and the final syllable 'station' corresponds to the ''. 'ㅋ' vocalization is 'Ki', and the final syllable '읔' becomes 'ㄱ'. The same can be applied to the remaining six representative notes.

The vocabulary stored in the recognition target vocabulary dictionary 10 is divided and stored according to the phonetic value of the first consonant.

In addition, the recognition target vocabulary dictionary 10 stores regions according to first consonants and stores the HMM model.

Thus, the pronunciation of the consonants is because the pronunciation of the Hangul consonants is more favorable because the phoneme comes out twice for each consonant. In contrast, the English alphabet is A, B, C,…. Since the phoneme is mainly pronounced only once, such as Z, etc., it is disadvantageous for accurate speech recognition.

Meanwhile, the voice recognizer 20 receives a voice signal input from the outside, in particular, a voice signal by human voice, extracts a feature thereof, and compares it with the HMM model of the target vocabulary stored in the recognition target vocabulary dictionary 10. The recognition result is output according to the probability value.

At this time, the speech recognition preprocessor (not shown) performs a preprocessing process on the speech signal input by the human voice and quantizes it, and outputs the quantized signal to the speech recognizer 20. The speech recognition preprocessor may be included in the speech recognizer 20.

The speech recognizer 20 performs two speech recognition operations to recognize a single word.

The first voice recognition operation is to determine the phonetic value of the first consonant of the word to be recognized by referring to the lexical dictionary 10 to recognize the voice signal for the first consonant of the word to be recognized. This operation is indicated by the dotted line in FIG.

The second voice recognition operation is to determine the input word by comparing the selected recognition target vocabulary according to the phonetic value determined by the first voice recognition operation after receiving the voice signal for the recognition target word. This operation is indicated by the solid line in FIG.

Hereinafter, with reference to the accompanying Figure 3 will be described in the speech recognition method in an embodiment of the present invention.

First, when a person utters the first consonant of a word to be recognized (S1), the speech recognizer 20 recognizes the sound value of the first consonant uttered by the person (S2).

When the phonetic value of the first consonant uttered by a person is recognized, the speech recognizer 20 reduces the list of recognized vocabulary words into a vocabulary divided into regions of the recognized phonetic value among all the vocabularies stored in the recognized vocabulary dictionary 10 (S3). ).

Thereafter, when a person speaks an actual recognition target word (S4), the speech recognizer 20 performs speech recognition on the reduced vocabulary list (S5).

That is, when the voice signal (voice signal quantized and input by the voice recognition preprocessor) of the actual word to be recognized is input, the feature of the received voice signal is extracted, the target word is HMM modeled from the reduced voice vocabulary list, The result is compared with the feature extraction data of the received voice signal and output as a recognition result according to the statistical probability value (S6).

In this way, the speech recognizer 20 may display the recognition result more quickly by performing the HMM modeling on only the reduced recognition target vocabulary corresponding to the first consonant of the recognition target word.

Meanwhile, the voice recognition method according to the embodiment of the present invention as described above may be implemented as a program and stored in a recording medium (CD-ROM, RAM, ROM, floppy disk, hard disk, magneto-optical disk, etc.) in a computer-readable form. Can be.

Although the present invention has been described with reference to the most practical and preferred embodiment, the present invention is not limited to the above-described embodiment, but also includes various modifications and equivalents within the scope of the following claims.

According to the present invention, the recognition target vocabulary to be recognized at the same time can be significantly reduced, for example, by 1/19 on the average in inverse proportion to the number of first consonants of the recognition target word.

In addition, the pronunciation of the Hangul consonants, such as a, b, c, etc., due to the nature of the pronunciation, the phoneme comes out twice for each consonant, there is an advantage in the speech recognition of the consonant.

1 is a block diagram of a voice recognition device according to an embodiment of the present invention.

2 is a view showing a representative sound value of the final consonants in the speech recognition apparatus according to an embodiment of the present invention.

3 is a flowchart of a voice recognition method according to an embodiment of the present invention.

<Brief description of symbols for the main parts of the drawings>

10: Vocabulary dictionary of recognition 20: Speech recognizer

Claims (6)

a) receiving a voice signal corresponding to the first consonant of the recognition target vocabulary and input by a human voice; b) analyzing the received voice signal to recognize a corresponding consonant; c) reducing a vocabulary in a region divided according to the recognized consonants among all recognized vocabulary words into a recognized vocabulary; d) receiving a voice signal corresponding to a recognition target vocabulary input by human speech; And e) performing speech recognition by analyzing the speech signal received in step d) on the reduced recognition target vocabulary in step c) and outputting a speech recognition result; Speech recognition method comprising a. The method of claim 1, In step b), when the consonant is recognized, the phonetic value of the first consonant voice is recognized. In step c), the recognition target vocab region having the same consonant as the phonetic value recognized in step b) is identified. Speech recognition method, characterized in that. The method of claim 2, Representative phonetics of the Jongseong 'a', 'b', 'c', 'ㄹ', 'ㅁ', 'ㅂ' and 'ㅇ' voice recognition method characterized in that seven. The method of claim 3, The representative note 'ㄱ' includes consonants 'ㄱ' and 'ㅋ', The representative note 'b' includes a consonant 'b', The representative phonetic value 'ㄷ' includes consonants 'ㄷ', 'ㅅ', 'ㅈ', 'ㅊ', 'ㅌ' and 'ㅎ', The representative note 'ㄹ' includes a consonant 'ㄹ', The representative note 'ㅁ' includes 'ㅁ', The representative note 'ㅂ' includes 'ㅂ' and 'ㄷ', The representative note 'ㅇ' includes 'ㅇ' Speech recognition method, characterized in that. The method of claim 1, Before step a), Speech recognition method further comprises the step of performing a pre-processing process for the speech signal input by the human voice quantization and outputting. A computer-readable recording medium tangibly embodying a program composed of instructions executable by a computer to perform steps of a speech recognition method, The method step, a) receiving a voice signal corresponding to the first consonant of the recognition target vocabulary and input by a human voice; b) analyzing the received voice signal to recognize a corresponding consonant; c) reducing a vocabulary in a region divided according to the recognized consonants among all recognized vocabulary words into a recognized vocabulary; d) receiving a voice signal corresponding to a recognition target vocabulary input by human speech; And e) performing speech recognition by analyzing the speech signal received in step d) on the reduced recognition target vocabulary in step c) and outputting a speech recognition result; Computer-readable recording medium comprising a.