JPS6048099A - Voice recognition equipment - 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 Field of the Invention The present invention relates to a speech recognition device that recognizes input speech using standard patterns of pre-registered words and monosyllabic speech.

2. Description of the Related Art Structures and Problems Therein In recent years, voice has been attracting attention as an input means for human-machine systems, and various voice recognition devices have been commercialized. These speech recognition devices include those that perform word-by-word recognition, and
There are speech recognition devices that perform recognition in units of monosyllables, and in particular, speech recognition devices that are capable of recognizing units in monosyllables do not have a VC limit on the number of words, so it is possible to input any sentence. At this time, if frequently used words such as numbers can be recognized in the same way as monosyllables, inputting sentences or data will be even easier.

Therefore, speech recognition devices that recognize both words and monosyllables have also been commercialized.

A conventional speech recognition device for recognizing words and monosyllables will be explained below with reference to FIG. 3 is an A/D converter that digitizes the voice features extracted by the feature extractor 2; 4 is a digitizer that is digitized by the A/D converter 3; Reference numeral 5 denotes an input pattern storage unit that temporarily stores the characteristics of the voice that has been picked up as an input pattern, and a word standard word storage unit 5 that stores in advance the characteristics of the sounds of a plurality of words to be recognized as standard patterns. 6 is a monosyllabic standard pattern storage section in which characteristics of a plurality of monosyllabic sounds to be recognized are stored in advance as a standard pattern; 7 is a standard S;
This is a mode changeover switch that connects the turn storage section 5.6 or the input pattern storage section 4 and the above-mentioned to/D conversion section 3. It is a mode changeover switch that connects the turn storage section 5. This is means for selecting a recognition mode in which the characteristics of the input voice are stored in the turn storage section 4. Reference numeral 8 denotes a speech length detecting section which detects the length of the input speech from the input kabata turn and determines whether to compare the incoming kabata turn with a standard word turn or a standard monosyllabic pattern; 9 is controlled by the speech detecting section 8; Gate 1Q compares the input cover turn stored in the input cut turn storage unit 4 with a plurality of word standard turn turns stored in the word standard turn storage unit 5, and converts the input speech into a specific word. The monosyllabic recognition processing unit 11 compares the incoming crinkle turn with a plurality of monosyllabic standard patterns stored in the monosyllabic standard pattern storage unit 6, and recognizes the input speech as a specific monosyllable. The syllable recognition processing unit 12 is an output unit that outputs a signal corresponding to the word or monosyllable recognized by both recognition processing units 1, 11.The speech recognition device configured as described above will be described in detail below. 0 First, in the word registration mode, press the mode changeover switch TtA.
Words are registered by uttering words corresponding to various output signals outputted from the output section 12 and inputting them to the input section 1. That is, inputted speech is converted into an electrical signal by the input section 1, features of the speech are extracted from this electrical signal by the feature extraction section 2, and the features of the speech are digitized by the A/D conversion section 3 to generate words. The characteristics of this digitized voice are registered in the standard pattern storage section 5.

In addition, in the monosyllable registration mode, by connecting the switch 7 to the C side and uttering monosyllables corresponding to various output signals output from the output section 12 and inputting them to the input section 1,
Monosyllables are registered in the same way as when registering words.

Next, in the recognition mode, connect the mode changeover switch 7 to the B side, and input a desired word or monosyllable from among the plural words or monosyllables registered in the both registration modes to the input section 1.
, the output unit 12 outputs a signal corresponding to the standard pattern determined by one of the recognition processing units 10 and 11 to be most similar to the input cover pattern. That is, the voice input to the input section 1 is converted into a digitalized feature pattern by the feature extraction section 2 and the D/D conversion section 3, and once stored in the input pattern storage section 4, it is stored in the input pattern storage section 4. The speech length detector 8 checks whether the input speech length is longer than a certain threshold. First, if the input speech length is longer than a certain value, the input pattern is sent to the word recognition processing section 10 via the gate 9 controlled by the speech detection section 8, where it is stored in the Hogo standard pattern storage section 5vc. It is compared with a plurality of standard patterns, and a signal corresponding to the most similar word is output. On the other hand, the input audio length is the threshold, 1
If the length is also shorter, the input pattern is sent to the monosyllabic recognition processing unit 11 via the gate 9, where it is compared with a plurality of standard patterns stored in the monosyllabic standard pattern storage unit 6, and the most similar pattern is selected. A signal corresponding to the first section of the table is output.

However, in the speech recognition device configured as described above, a word recognition processing section for recognizing words and a monosyllable recognition processing section for recognizing monosyllables are provided separately, and words and monosyllables are processed separately. However, since it is necessary to increase the size of the device and complicate the processing, it is necessary to increase the size of the device and complicate the processing.In addition, since it is determined whether the sound is a single syllable or a word based on the length of the sound, words consisting of one syllable, For example, it has the disadvantage that it is prone to errors, such as determining the numbers 2 (``ni'') and 5 (``do'') to be monosyllables, or determining a monosyllable uttered for a long time to be a word.

Purpose of the Invention In view of the above-mentioned drawbacks, the present invention recognizes words with characteristics at the beginning using the same processing as word clauses, thereby reducing the size of the device.
It is an object of the present invention to provide a speech recognition device that can simplify processing and prevent erroneous recognition.

Structure of the Invention The present invention includes a word standard pattern storage unit that stores the characteristics of the beginnings of words having different syllables at the beginning as a standard pattern, and a word standard pattern storage unit that stores a standard pattern to be compared with input and < turn. a switch means for selecting whether to receive from the part or the monosyllabic standard pattern storage part, and a recognition processing means for comparing and recognizing each standard nok turn and the input kaku turn sent through the switch means. By selecting whether the recognition target is a word or a monosyllable using a switch when inputting speech, it is possible to recognize words and monosyllables in the same process. Confusion between words and monosyllables can also be avoided.

Description of examples Embodiments of the present invention will be described below with reference to the drawings.

FIG. 2 is a block diagram showing a speech recognition device according to an embodiment of the present invention, which is capable of recognizing 10 numbers and 68 syllables.

In the figure, 1 is an input section consisting of a microphone or the like that converts voice into an electrical signal, 2 is a feature extraction section that extracts features of the voice from the electrical signal from the input section 1, and 3 is a section that extracts the features of the voice from the electrical signal from the input section 1. Audio 11! 4 is an input cover pattern storage unit that temporarily stores the audio features digitized by the A/D converter 3 as input cover patterns; 5 is 1Q numbers; 1,
2,3゜4.6, 6, 7, 8, 9.0 is a word standard noku turn storage unit that stores the voice characteristics of 0 as standard patterns, 6 is the voice characteristic of 68 monosyllables. 1 is a monosyllabic standard pattern storage section in which the standard pattern is stored as a standard pattern, Ryo is a mode changeover switch that connects both standard pattern storage sections 6, 6 or the input cover pattern storage section 4 and the A/D conversion section 3, and 11 is an input cutout pattern storage section. 12 is a signal corresponding to the monosyllable or number recognized by the monosyllabic recognition processing unit 11. An output unit 13 is a switch for selecting whether to compare the input kabata turn with a word standard noku turn or a monosyllabic standard pattern.

The operation of the speech recognition apparatus configured as described above will be described below with reference to FIG. 2.

First, when in word registration mode, press mode selector switch 7 to A.
Registering the numbers is done by connecting to the side and manually inputting the numbers corresponding to the signals outputted from the output section 12 into the input section 1 in sequence. That is, input audio is input to the input unit 1.
Convert it into an electrical signal by , and from this electrical signal, you can calculate the number - the characteristic of the sound chamber at the beginning of the prison, for example, 1 for "i" and 2 for 2.
The feature of this syllable is extracted by the feature extraction unit 2, and the feature of this syllable is defined as A.
The /D converter 3 converts the number into a digit, and registers all the features of the first syllable of this number in the word standard pattern storage 5.

In addition, in the monosyllabic registration mode, by connecting the mode selector switch 7 to the C side and uttering monosyllables corresponding to the signal output from the output section 12 and sequentially inputting them to the input section 1, monosyllables can be registered. will be done.

That is, input speech is converted into an electrical signal by the input section 1, and features of this monosyllable are extracted by the feature extraction section 2,
The features of this monosyllable are digitized by the A/D converter 3 and are registered in the monosyllabic standard pattern storage 7'111e.

Next, in the recognition mode, connect the mode changeover switch 7 to the D side, connect the switch 13 to the D side, and input a desired number from among the 10 numbers registered in the word registration mode into the input section 1. A signal corresponding to the number determined to be most similar to the input speech by the monosyllable recognition processing section 11 is output from the output section 12, and the 68 syllables registered at the time of monosyllable registration are output by connecting the switch 13 to the E side. When a desired monosyllable among the above is input to the input section 1, a signal corresponding to the monosyllable determined by the monosyllable recognition processing section 11 to be most similar to the input speech is outputted from the output section 12. That is, when first inputting numbers, the switch 13 is connected to the D side, and the voice input to the input section 1 is converted into a digital feature pattern by the feature extraction section 2 and the A/D conversion section 3. The monosyllable recognition processing unit 11 compares the standard pattern extracted from the initial syllable of the number stored in the word standard pattern storage unit 5 with the above input cover pattern, which is temporarily stored in the input cover pattern storage unit 4. . The first syllable of a number is
``I''(1)l''ni''(2).

'sa' (3) r 'yo' (4), 'go' (5
), +1ro” (6) l “na” (7), “ha”
(s > + "ki" (9), "ze" (0) can be sufficiently recognized by the same recognition process as a monosyllable, and the number corresponding to the number recognized by the monosyllable recognition processing unit 11 A signal is output from the output unit 12. On the other hand, when a monosyllable is input 2)
At this time, the switch 13 is connected to the E side, and each standard pattern stored in the monosyllabic standard pattern storage section 6 is compared with the input pattern pattern described above by the monosyllabic recognition processing section '11. A signal corresponding to the single syllable is output.

As described above, according to this embodiment, the switch 13 is provided to separately send the standard pattern extracted from the initial syllables of 10 numbers and the standard pattern of 68 syllables to the monosyllable recognition processing unit 11. 10 with different syllables at the beginning of the word
numbers and 68 i'? clauses can be recognized by the same monosyllable recognition processing unit, and the apparatus can be simplified.

A second embodiment of the present invention will be described below with reference to the drawings.

FIG. 3 is a block diagram of a speaker-independent speech recognition device showing a second embodiment of the present invention.

In the same figure, 1 is an input section, 2 is a feature extraction and output section, and 3 is an A
/D conversion section, 4 is an input cover pattern storage section, 11 is a monosyllable recognition processing section, and 12 is an output section, which is the same as the configuration shown in FIG. 2. The difference from the configuration shown in FIG. 2 is that a wavy standard pattern storage unit 14 is provided in which a standard pattern for unspecified speakers extracted from the first syllable of the 1o number is stored in advance.
``Tensou'' in advance.

``Henkan II, ++Sakushiyon, ++Matsusyo''.

``Torikeshi u, 'a Kaigyo 11, u space''.

A command standard pattern list 1, part C 15, is provided, which stores standard patterns for unspecified speakers extracted from the first syllable of commands used in text input devices such as A monosyllabic standard pattern storage unit 16 is provided to store standard patterns for unspecified speakers, and a switch 17 is provided to change the input cover turns to numbers, numbers, etc.
The point is that we have eight choices to choose from compared to the standard patterns of commands and single syllables.

The operation of the speech recognition device of the second embodiment configured as described above will be explained below.

When inputting numbers without selecting them, connect the switch 17 to the D side, speak the desired number out of the 10 numbers, and input it into the human power section 1. This voice is converted into a digitized feature pattern by the feature extraction section 2 and the A/D conversion section 3, and once stored in the input pattern storage section 4, the monosyllable recognition processing section 11 stores the numerical standard pattern. Each standard pattern in part 14 (pai, "ni II, ++sa II, 11yo, 'Go n, II Ro".

“'na u,,,++ is II, IIsaII, It
The symbol corresponding to the most similar number is output to the output section 12.

Similarly, if you want to output a command to be used in the text input device, connect the switch 17 to the E side, utter the desired command and input it into the input unit 1, and the monosyllable recognition processing unit 11 Each standard pattern in the command standard pattern storage section 15 is compared with the human cover pattern, and the recognition result is outputted by the output section 12. That is, the features of the first monosyllable from the beginning of the input command are extracted and compared with the standard pattern extracted from the beginning of each command, and the command G′ whose initial syllable is most similar to the input is selected. A signal is output by the output section 12.

Next, when inputting a monosyllable, connect it to the switch 17iF side, utter a desired monosyllable among the monosyllables stored in the monosyllable standard pattern storage section 16 in advance, and input it to the input section 1. In exactly the same manner as in the embodiment described above, the output section 12 outputs a signal corresponding to the monosyllable most similar to the input cover turn.

As described above, according to the present embodiment, the switch 17 is provided, and the standard pattern extracted from the initial syllables of the words of 10 numbers and the standard pattern extracted from the initial syllables of the commands used in the text input device. , and 68 syllable patterns to the single syllable recognition processing unit 11, 10 numbers with different syllables at the beginning of words, commands with different syllables at the beginning of words, and 68 syllables can be combined into the same word. It can be recognized by a syllable recognition processing section, and the structure of the three storage sections can also be made the same, making it possible to simplify the device.

In the first and second embodiments, numbers and commands are used as objects to be recognized, but any group of words having different syllables at the beginning may be used.

Although the small A/D converter 3 is installed after the feature extractor 2, it is actually installed before the feature extractor 2, digitizes the electrical signal first, and extracts the features from the digitized signal. Needless to say, it's a good thing.

Effects of the Invention The speech recognition device of the present invention is equipped with a switch for selecting a standard pattern storage section, and divides and narrows down recognition candidates into groups of words that have different syllables at the beginning of each word. It is possible to recognize single passages and words using only recognition processing, and the structure of the word and monosyllable dictionaries can be standardized, simplifying the device and processing.
(You can proceed.

[Brief explanation of drawings]

FIG. 1 is a block diagram of a conventional speech recognition device, FIG. 2 is a block diagram of a speech recognition device according to a first embodiment of the present invention, and FIG. 3 is a block diagram of a speech recognition device according to a second embodiment of the present invention. FIG. 2 is a block diagram of the device. 13...Switch, 14...Numeric standard pattern storage section, 15...Command standard pattern storage section, 16...Monosyllabic standard pattern storage section, 17.
switch.

Claims (1)

[Claims] an input means for converting input speech into an electrical signal; a feature extraction means for extracting features of a monosyllable at the beginning of a word from the electrical signal;
input cover pattern storage means for temporarily storing the voice features extracted by the feature extraction means as input cover patterns; and input cover pattern storage means for temporarily storing the voice features extracted by the feature extraction means as input cover patterns; at least one word standard pattern storage means for storing, a monosyllabic standard pattern storage means for storing as a standard pattern the features extracted by the feature extraction means from the monosyllabic speech input in advance, and the standard pattern. A switch means for selecting one of the storage means compares the standard pattern stored in the standard pattern storage means selected by the switch means with the input cover pattern, and selects the input cover pattern as a specific standard pattern. What is claimed is a speech recognition device comprising a recognition processing means for recognizing the specified standard pattern, and an output means for outputting a signal corresponding to the specific standard pattern recognized by the recognition processing means.