JPS63159900A - Voice information input 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 [Field of Industrial Application] The present invention relates to a voice information input method suitable for use in a character reading device to improve the efficiency of input work.

[Conventional technology]

In conventional devices, as in Japanese Patent Laid-Open No. 57-200100, it was considered to combine a character reading device and voice recognition hardware by sharing part of the hardware. However, no consideration was given to the recognition range during voice input.

[Problem that the invention seeks to solve]

The above conventional technology does not take into consideration the recognition range, and in particular, character reading in OCR is performed in units of words, so voice input is inevitably L&M in units of syllables.
As is well known, this poses a problem in terms of recognition accuracy (recognition rate).

An object of the present invention is to stably and quickly input characters by voice that cannot be input by OCR.

[Means for solving problems]

The above objective is achieved by reflecting the information captured by OCR in the recognition range.

[Effect]

The character input data read by OCR is subjected to similarity calculation (matching) with a standard character pattern, and then the result is determined by a determination unit. At this time, even if the external display such as a CRT is only for the first place, internally the nth place (
The results up to (n is a parameter given separately) are held in memory, and these candidate words are used as the recognition range when correcting or re-inputting by voice input. In this way,
Even if there are tens to hundreds of types of input characters, only n words are input by voice, so accurate recognition can be expected. Furthermore, since the amount of calculations is reduced, the processing time can also be shortened.

〔Example〕

An embodiment of the present invention will be described below with reference to FIG. In the figure, numeral 1 is a photoelectric conversion unit that converts character data into electrical signals, and the character information read here has its features extracted by a feature extraction unit 2, and then sent to a similarity calculation unit 3. The similarity calculation section performs expansion/contraction matching between the input data and the standard character data in the character standard pattern memory 4 to determine the similarity. If the degree of similarity determined by the similarity degree determination section is sufficiently high and there is a sufficient difference from the degree of similarity of the second or lower candidates, the determination section 5 transmits the character code of the first place to the host device or C.
It is output to RT, etc., but if the degree of similarity is low or there is not much difference in degree of similarity from the second or lower candidate, re-input by voice input is prompted. After the voice generated by the operator is amplified and AD converted in the voice input section 6, features and characteristic parameters are obtained in the feature extraction section 7, and these feature parameters are calculated in the similarity calculation section 9. Matching with the standard voice data in the voice standard pattern memory of No. 8 is performed.

At this time, without performing matching with all standard patterns, the recognition range control section 10 limits the words to words that have been selected as candidates in advance by the OCR section.

The characters read by OCR are usually numbers, alphabets,
There are 10 to 100 katakana characters, and some of these include words with similar pronunciations, such as ``2'', ``niJ'', and rEJ, so let's re-enter or correct them aloud. However, according to this embodiment, the number of characters to be recognized is considerably limited, so the recognition accuracy is increased and the amount of processing in the similarity calculation section 9 is reduced. This has the effect of shortening processing time.

〔Effect of the invention〕

According to the present invention, even if there are a large number of characters to be input, the voice recognition range is limited to candidates selected in advance by OCR, so that the recognition accuracy is increased and the time required for arithmetic processing is reduced.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing one embodiment of the present invention. 1... Photoelectric conversion device, 3... Similarity calculation unit, 4...
・Character standard pattern memory, 5... Judgment section, 6...
- Voice input section, 9... Similarity calculation section, 10... Recognition range control section. Xi diagram

Claims (1)

[Claims] 1. In an OCR (optical character reader) equipped with a voice recognition device, when correcting characters misread or rejected by the OCR using voice input, limit the voice recognition range to the top candidates determined by the OCR. A voice information input method characterized by improving recognition accuracy and speed.