JPS6126133A - Voice recognition input device - Google Patents

Abstract

PURPOSE:To reduce the number of rejects and to execute voice input precisely and rapidly by sorting and registering a word to be registered in each group having similarity, matching the word with an input pattern and reading out the registered word including the distance of similarity between both the patterns within a fixed range. CONSTITUTION:A word voice from an input part A is registered in a storage part B as a reference pattern. The reference patterns including the distances of similarity as to the pronunciation of each word are sorted into the same group and registered. In a processing part C, a voice signal from a voice input part A1 is analyzed by an analyzing circuit 1, the analyzed signal is patterned by a patterning circuit 2 and the patterned signal is compared with the reference pattern read out from the storage part B by a comparator 3. When a recognition processing means 4 decides that the voice signal has the similarity in the fixed range as the result of the comparison, a proposed order deciding means 5 reads out the voice signal as a proposed recognition word and displays the word on a display part E. A speaker selects the pronounced word out of the displayed proposed words to output the selected word to a main processor G.

Description

DETAILED DESCRIPTION OF THE INVENTION 3.1 Industrial Application Field The present invention relates to a speech recognition input device for recognizing words uttered by a speaker and inputting the recognized words to a main processing unit.

3.2 Prior Art and Its Disadvantages Conventionally, in voice recognition input devices, the methods for the speaker to confirm the identity of the words the speaker tried to input and the words recognized by the device were as follows: There is a method in which the speaker listens to determine the words output by the device, and a method in which the speaker confirms the words by looking at the words displayed by the device on a display such as a CRT.

And even though the input word is among the registered words,
The conventional processing method when it is determined that there is no such word (when it is determined that there is no word) is that the speaker inputs the word again by voice, or inputs the same code as the input word using the keyboard or numeric keypad. , enter the code of the desired word 111jl
It is output to the location.

Therefore, in the past, each time a rejection was received, it was troublesome to enter the voice again, or it was necessary to operate the keyboard or numeric keypad, making the process troublesome and prone to erroneous operations. There were drawbacks such as complexity.

3.3 Objects of the present invention and means for achieving the same In view of the above points, the present invention provides a simple structure and method to minimize rejections and to perform voice input reliably and quickly. The purpose of the present invention is to provide a reputation recognition input device.

3.4 Examples of the present invention Next, embodiments of the invention will be described based on the drawings.

The speech recognition input device according to the present invention, as schematically shown in FIG. 1, includes an input section A, a storage section B1, a processing section C1, a temporary storage section D1, a display section [, and a control section F].

The input section is equipped with a microphone that converts audio into electrical signals (an audio type consisting of one phone and an amplifier that amplifies the electrical signal).
3 part A1, -do input part △2, and selection input part Δ3
It consists of The code input section Δ2 is for setting the standard pattern storage area of the storage section B using, for example, numerical values, and is composed of a numeric keypad or the like. Selection input section A
Reference numeral 3 is for inputting which of the items displayed on the display section E and required to be selected, and consists of a switch that gives a signal having two meanings, and an optional switch of the numeric keypad. It can be composed of

Storage part B contains a program that defines the operating procedure of this voice recognition input device, a standard pattern that patterns the characteristics of the required words uttered by the speaker, and a display that displays the required characters etc. on the display unit. Patterns are stored.

The standard patterns are classified into groups and registered, with the pronunciation of each word having a similarity distance within a certain range as a - group. That is, as shown in FIG. 2, an example of the standard pattern storage area and its storage contents is replaced with a human organ, each registered word (Fukushima, Tokushima,...)
The code (01, 02...) that matches the address is entered, and the pronunciation pronunciation similarity is within a certain range.
group and by group (a, a'.

a'', b, b', c, c'...
・) are stored separately.

The processing section C analyzes the audio signal input from the audio input section A+ using an analysis circuit 1, compresses its characteristics using a patterning circuit 2, patterns it into a form similar to the standard pattern, and uses a matching circuit 3 to A recognition processing means 4 compares the input pattern with the standard pattern read from the storage unit B, and the recognition processing means determines that the input pattern and the standard pattern have similarity within a certain range. and means 5 for determining the candidate ranking of words in a given case.

The display section is a CRT (cathode ray tube), liquid crystal display, plasma display, LED display, etc. that displays text and images, and is used to request selection of actions, operations, etc.
Displays a message asking for confirmation or indicating the end of an operation.

In a preferred embodiment, the selection input section A3 includes:
A transparent touch switch (pressure-sensitive or temperature-sensitive switch) A30. placed on the display surface of this display unit provides signals with two contradictory meanings. It is composed of A3□.

The control section F sends and receives signals and data between the above-mentioned sections A to E, and controls operations, and finally outputs recognized words to the main processing device G. Such a control section F can be composed of one microphone and one viewer, together with a storage section B1, a processing section C1, and a temporary storage section.

Next, the operation of the voice recognition input device having the above configuration and other A' layer configurations will be explained.

For example, when the voice input switch (not shown) provided on the voice type horn section A1 is turned on, the control section ``starts executing the program'' as shown in flowcharts 1 to 4 in FIG. The display unit F displays that voice input is possible (pl). When the speaker utters the desired word (p2), the recognition processing means 4 converts the voice signal into an input pattern, and The matching circuit 3 of the recognition processing means 4 compares it with the standard pattern registered in the storage part B. When the distance of similarity between the incoming horn pattern and the standard pattern is within a preset range, If it is determined, it outputs a signal of “1” which means recognition, otherwise it outputs a signal of “0” which means no recognition.
) is output (p3).

When the recognition processing means 4 outputs "'1", all words in the group to which the standard pattern within the predetermined range belongs are stored in the -C-115 storage unit as recognition word candidates.

Then, the candidate rank determining means 5 uses the candidate ranking determining means 5 to determine the magnitude of the similarity distance between the memorized Noboru 8 () word and the input 111 words when comparing each standard pattern and the Itsuno pattern.
-( is determined based on the degree of r4, and candidate class 1c7 is determined.

Next, based on the output of the candidate ranking determining means 5, the word closest to the input word is displayed on the display section F as the first candidate from among the groups having similarity in the temporary storage section (p
4).

At the same time, behind the displayed A32, characters that prompt recognition and a frame that clearly indicates the range to be touched are displayed on the display section [. By using such touch switches, we can expect the effect of promoting confirmation and preventing false confirmation1.In this way, depending on which touch switch is touched, it is possible to check whether the displayed 141 words are the same as the input word. The signal is input to the control section F (p6). Regarding the displayed first candidate word.

When the touch switch △3° is operated to indicate that it matches the input single item, the control unit sends the code of the first candidate AA 5R to the main processing unit G via the interface based on that signal.
Output to (06).

However, if the touch switch A32 is operated to indicate that the displayed first candidate word does not match the input word, based on that signal and the ranking signal of the candidate ranking determining means 5, one is selected from among the groups in the temporary storage section. The word with the second highest similarity is displayed as the second candidate word (p7). While a match is not confirmed, the display of the next candidate word and the operation of the touch switch are repeated until there are no more words in the group (until the end of the set) (+) 9 to p11) - Similar to the display section F above All words (recognized word candidates) within a group may be displayed at once. In this case, a touch switch is placed at the beginning of each word, and by touching the part where the word is displayed with your fingertip, a signal is sent that means that the speaker has IR'd that word. is input to the control section F from the selection input section. As a result, the control section outputs the "-" of the selected word to the main processing device G.

Above) If the speaker accurately utters a word that matches the registered middle part, as in E, then the registered middle word B within a certain similarity range with the input word. Since i is always displayed on the display in the order of closest similarity, it usually matches the input word by the end of the set! 1 oil can be output to the main treatment equipment.

However, if the speaker is not feeling well, for example,
:”; Because of 'cK, it is possible that the ``ili;ji'' uttered may be rejected without being recognized.In such a case, it is inconvenient if voice input is not possible forever. In the preferred embodiment, a function that can deal with this is added, and this will be explained below.

The word uttered by the speaker was judged to be unrecognized and the word was judged to be unrecognized.
(+) 3), 9191 times are counted, and the number of rejections n is, for example, 112'
It is determined whether or not it is II (Ill 2). If it is not 0-2, the display section E (displays that voice input is possible again and prompts the speaker to speak again (Dt)
.

If the result is Rejiek 1~@number n-2, the speaker is instructed to enter the code of the word that was determined to be absent from the code input area A2 in the display area E, and to repeat the code. Displays what should be registered (+)'13).

When the speaker follows this and inputs the code for the word from the =1-do input section Δ2, the program jumps to p6, the code for A is output to the main processing unit G, and the desired word is input. Further, following the input of the code, the display section [2] displays characters or figures to the same effect requesting a selection as to whether or not to re-register the word just entered in the storage section B (D15). At this time, on the same display screen, characters are displayed on the touch switch A to request the designation of whether to select re-registration or not.
31. It is displayed at the position corresponding to A32.

When the speaker operates the switch A3□ indicating that re-registration is not to be performed, the execution of the recognition/registration program currently being executed is terminated.

In contrast, switch A indicates that the speaker selects re-registration.
If you operate 31, the registration mode is set, and characters indicating that you should enter the code of the word you want to register from the code input section A2 are displayed on the display section E (+)I.
T).

Based on this, when the speaker enters the code (pt8), the code just entered is displayed in the code display area of the same screen, and the position corresponding to the touch switch on the lower side of is displayed. Displays characters requesting confirmation of the code (pt,).

If you operate the No touch switch and it is IC, you will be prompted to enter the code again. Further, when the OK touch switch is operated, a message indicating that voice input is possible is displayed on the display section (+120). When the speaker utters the desired word (+)21), at the stage where the recognition processing section C has finished the analysis and patterning, the control section F stores the input pattern in the storage section specified by the previous input code. Store in storage area B. This completes the re-registration of the word. At 5 o'clock when the input of the word is completed, characters indicating completion are displayed on the display section E (+) 23), and the execution of this recognition/registration program is completed.

Therefore, if this speaker inputs the same word from now on,
Speech recognition is performed using this re-registered standard pattern.

3.5 Effects of the present invention As explained above, according to the present invention, words uttered by a speaker are registered in the storage unit as a standard pattern, and words uttered by the same speaker (f 5'6. F'l L/ 7-1, with Kaura as the input pattern [, this input pattern 4 the mark i
% pattern and output it to the same platform on the same platform as the voice recognition input. (The similarity distance between the two is constant in comparison with Based on the speaker's operation of the selection section, the selected t11 word is output to the main processing unit, in case it is misrecognized due to the similarity of the speaker's R name. , words within a certain range are displayed sequentially by the selection input operation, and the words within a certain range are displayed sequentially by the selection input operation.
Since all or part of 11 can be selected and output, the number of times of rejigging can be minimized, and the tone insertion saw can be performed reliably and quickly.

[Brief explanation of drawings]

The drawings show one embodiment of the present invention, and FIG. 1 is a block diagram showing a schematic configuration, FIG. 2 is an explanatory diagram showing an example of a standard pattern storage mode, and FIG. 3 is an illustration of a display section. A front view showing an example, and FIG. 4 is a flowchart showing the operation flow. A...Input section C...Processing section

Claims (1)

[Claims] Words uttered by a speaker are registered in the storage unit as standard patterns, words uttered arbitrarily by the same speaker are used as input patterns, and this input pattern is compared with the standard pattern, and if they are the same, In the speech recognition input device that outputs to the main processing unit, (a) the registered words are classified and registered into groups with similarity, and (b) the distance of similarity between the two is determined in comparison with the input pattern. All or part of the registered words of a group to which words with registered patterns within a certain range belong are read out as recognition word candidates, displayed on the display section, and (c) selected based on the speaker's operation of the selection section. A speech recognition input device, characterized in that it outputs words to the main processing device.