JPS5934598A - Voice input unit for printing - 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 The present invention relates to a printing voice input device for converting voice into kana code, kanji code and symbol code and inputting the converted voice into a computer typesetting system.

2. Description of the Related Art Conventionally, when inputting a printing document (referring to a document referred to by an operator during phototypesetting; the same applies hereinafter), an operator operates a keyboard or the like with his or her hands or fingers. For this reason, there are disadvantages in that inputting data requires a great deal of effort, requires technical skill, and input work is physically tiring.

Printing voice input devices have been proposed to overcome these drawbacks, but with conventional voice input devices, the recognition rate decreases due to changes in the operator's utterances over time or minute differences in the position of the microphone. However, there was a drawback that the speed of document input was reduced. SUMMARY OF THE INVENTION Accordingly, an object of the present invention is to provide a voice input device for printing that does not have the above-mentioned drawbacks.

This invention will be explained below.

The present invention relates to an audio input device for printing, and as shown in FIG.
a parameter extraction device w2 that extracts feature parameters of
A speech recognition device 3 that compares the monosyllabic feature parameters stored in advance in an internal storage device with the above feature parameters, and outputs the one with the highest degree of similarity as the corresponding monosyllabic code.
, a comparison/judgment device 4 that compares the input voice feature parameter with the corresponding monosyllabic feature parameter in an internal storage device and determines whether or not to learn the feature parameter, and a word processor 51 that performs kana-kanji conversion. It also has a storage device 52 for storing symbol codes corresponding to kanji codes and kana codes, inputs output codes from the speech recognition device 3, and outputs each of the above codes from the storage device 52 to the computer typesetting system 6. A kanji processing device 5 is provided.

As shown in FIG.
, a group of band pass filters 22 that divides the preprocessed audio signal SA into bands having different center frequencies, a channel selection circuit that sequentially selects each of the divided band signals using a control signal 'C8I, and a selection circuit. The sampling circuit 24 controls and samples the channel signal CH at a predetermined timing using the X-channel signal C82.

In such a configuration, the audio signal S from the microphone 1 is amplified and preprocessed by the preprocessing circuit 2J in the parameter extraction device 2, and is applied to a group of band pass filters 4 having different center frequencies. Here, the audio signals divided into each band are sequentially selected by the channel selection circuit B, and sent to the subsequent sampling circuit B for sampling.

The timing of each channel selection circuit 23 and sampling circuit is controlled by control signals C8I and C82 from the speech recognition device 3, and the speech signal divided into each band η' is sent to the speech recognition device 3 in time series as a characteristic parameter.
Output to.

On the other hand, the voice recognition device 3
Speech recognition is performed according to the selected mode, and a single 14-clause code is output to the word processor 51. The word processor 51 has a kana-to-kanji conversion function, converts the necessary portions of the printed manuscript into kanji, and stores the input manuscript together with its layout information in the storage device 52. The storage device 52 stores the information of the printing original page on a page-by-page basis.
The stored contents are output from the computer phototypesetting system 6. The speech recognition device 3 also sends control signals C8I and C8I, which give sampling timing to the parameter extraction device 2.
82 is output, but the sampling period is changed to increase the recognition rate according to the recognition mode specification. For example, in monosyllabic recognition mode, sampling is performed at approximately 2 millisecond intervals. In the word recognition mode, characteristic parameters are input at sampling times of about 10 milliseconds.

By the way, when starting voice input, first input the monosyllabic characteristic parameters (standard pattern) registered for each albino from the external storage device into the internal storage device, but due to the conditions at the time of registration or the minute position of the microphone. Due to these differences, some types of pronunciation may not be able to provide sufficient performance. Here, it has been found that updating the standard pattern at the time of starting voice input is an effective means for solving the above problem. However, in order to update the standard pattern, the operator is required to make accurate judgments for each update, which is a heavy burden and is difficult to use. Therefore, in this invention, the standard pattern is in a state where it is necessary to improve it (learning mode), the standard pattern is sufficiently updated, and the characteristic parameters of each internal storage device are appropriate for determining the input voice. It is possible to automatically determine the state (non-learning mode) and notify the operator of the determination result.

Figures 3 (A) and (B) show how the standard pattern in the internal memory is adapted by updating in the learning mode.The vertical axis shows the normalized evaluation value, and the horizontal The axis shows the number of updates.

Note that No in the same figure indicates the stable level of the evaluation value, and (5) in the same figure shows how the standard pattern of pronunciation 'ka' has been updated, and - in the figure indicates % a.
This figure shows how the standard pattern for pronunciation f is updated. As will be clear from this, there are cases where the effect of updating is large and cases where it is small, and this tendency varies from person to person, but generally the consonants are %p#, %1#, %ni#, %m" and 'n' have the effect of updating the standard pattern.

Here, the learning mode is based on the stability of the evaluation value→
Automatic conversion is performed in the non-learning mode, but since the stability level of the evaluation value differs depending on each sound, the stability level of each evaluation value is stored in the storage device in the speech recognition device 3, and will be described later. Judgments are made using software such as Further, the comparison/judgment device 4 calculates an evaluation value by comparing the characteristic parameter of the input speech with the corresponding monosyllabic characteristic parameter, and makes a determination using each evaluation value in the storage device within the speech recognition device 3.

Next, a flowchart of the software used for determining the evaluation value will be described with reference to FIG. 4.

First, as shown in FIG. 4(A), initial values are set at the time of initialization. A mask value for determining a learning state or a non-learning state is set based on the number of times a stable level is reached for all monosyllables (step 511), and a stable level of a preset evaluation value is set (step 512). . After that, the learning mode is set for all monosyllables (step 513).

FIG. 4(B) shows a learning/non-learning judgment and evaluation value judgment routine during execution, in which it is first judged whether the input voice is in learning mode or non-learning mode (
Step 521). If the learning mode is selected, the speech recognition device 3 calculates the evaluation value of the characteristic parameter of the input speech and compares it with the evaluation value stability level of the roaring monosyllable (steps 822, 523). Here, if the stable level has been reached (1), the corresponding monosyllabic feature parameter is updated using the feature parameter of the input speech (steps 824, 82.5). updates the mask value and determines whether the mask value has become MI OI (steps 826, 827).Here, if the mask value is 10
'', it is determined that the corresponding monosyllabic feature parameter has reached a sufficiently stable level, and the non-learning mode is set (step 528).

As described above, according to the voice input device of the present invention, it is not necessary to judge whether the voice input is at a stable level at the start of voice input, and if it is not at a stable level, the pre-registered premature syllable characteristic parameter is updated. As a result, the speech recognition rate can be significantly improved.

[Brief explanation of drawings]

Fig. 1 is a block diagram showing one embodiment of the present invention, Fig. 2 is a block diagram showing a part of it in detail, and the third section and (B) show the adaptation by updating the Sobare standard pattern. The diagram illustrating the situation, FIG. 4 (5), 03) is a flowchart of the software used for evaluating the evaluation value of the present invention. 1...Microphone, 2...Parameter extraction device,
3... Voice recognition device, 4... Comparison/judgment device, 5...
・Kanji processing device, 6...Computer typesetting system, 21...
・Pre-processing circuit, U... sampling circuit, 51...
Word processor, 52...Storage device. Applicant's agent Yu Yasugata Three-turning bee

Claims (1)

[Claims] a) A parameter extraction device which extracts feature parameters of input speech; b) Compares monosyllabic feature parameters stored in advance in an internal storage device with the feature parameters. C) A speech recognition device that outputs the one with the highest degree of similarity as the corresponding monosyllabic code; and C) Compares the input speech feature parameter with the corresponding monosyllabic feature parameter in the internal storage device, and d) a word processor for converting the monosyllabic code from kana to kanji, and a storage device for storing symbol codes corresponding to the kanji code and the kana code; a kanji processing device that inputs output codes from the voice recognition device and outputs each of the codes from the storage device to the computer phototypesetting system, learns the feature parameters when voice input starts, and stores them in the internal storage device. 1. A speech input device for printing, characterized in that a monosyllable characteristic parameter is improved.