JPH0415960B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an information input device, and particularly to an information input device for inputting information generated by a human, such as speech uttered in phoneme units or handwritten characters, into a machine.

The information input device according to the present invention can be applied whether the input information is voice or handwritten characters. However, in the following explanation, only the case of voice input will be described for the sake of clarity. Furthermore, in the following explanation, the units of input information referred to here are Japanese monosyllables such as "a", "i", "u", "e", and "o".

2. Description of the Related Art Conventionally, there are already speech input devices that recognize speech uttered in units of monosyllables. As an example of this, a method has been proposed in which uttered monosyllables are registered in advance as standard patterns, and recognition is performed by pattern matching between the uttered monosyllables and the input monosyllables. In such a registration type voice input device, selection of a standard pattern is important.
There are cases where the utterance is inappropriately uttered at the time of registration, or where the actual utterance differs from the standard pattern as time passes from the time of registration. To eliminate this effect,
Automatic learning of standard patterns based on the latest input audio is effective. If a recognition error occurs,
If the user always makes corrections, the standard pattern can be updated based on the input voice pattern. However, especially in the case of monosyllables, there is a risk that the user may overlook the error, so updating in this way may result in the incorrect speech pattern becoming the standard pattern.

The purpose of the present invention is to
To provide an information input device that can correctly update a standard pattern based on a new speech pattern without making corrections on a monosyllable basis.

The present invention includes a recognition unit that recognizes input information such as voice and handwritten characters based on a standard pattern registered in advance for each predetermined unit, and outputs a recognition result, and a word dictionary section in which words are stored; a dictionary matching section that matches the contents of the dictionary section with the recognition results to obtain word recognition results; and a dictionary matching section that matches the contents of the dictionary section with the recognition results to obtain word recognition results; and an update instruction section that makes a judgment and instructs updating of the standard pattern based on the judgment result.

The present invention will be described in detail below based on a specific example. FIG. 1 is a block diagram of an embodiment of the present invention. In the figure, 1 is a microphone, 2 is an analysis section, 3 is a voice matching section as a recognition section, 4 is a standard pattern memory section, 5 is a dictionary matching section, 6 is a word dictionary section, and 7 is an update instruction section. . A monosyllable inputted from the microphone 1 is analyzed by the analysis section 2 and outputted as a speech pattern P. At the same time, the voice pattern P is held in the analysis section 2. The audio pattern P is matched with the standard pattern R stored in the standard pattern memory section 4 in the audio matching section 3. As for the recognition results M in units of monosyllables, the results of the above-mentioned numbers are outputted in order of likelihood together with the likelihood value, that is, the degree of similarity. The recognition results for monosyllables are usually in the top 3.
Since more than 99% of the results are correct, this number of results is sufficient.

The dictionary matching section 5 matches the recognition result M with the contents of the word dictionary section 6 to obtain a word recognition result W.
Output. A word here means not only a normal word but also a series of several monosyllables.
It also includes phrases and the like. When the recognition result of one word of monosyllables is input to the dictionary matching section 5, first, among the words in the word dictionary section 6, the word with the matching number of characters is selected and read out. Compare the monosyllabic string of recognition results with the monosyllabic string of words in the word dictionary,
The sum of the degrees of similarity among the matched single syllables is taken as the degree of similarity of that word. The word with the highest degree of similarity is taken as the word recognition result.

For example, FIG. 2 shows an example of the relationship between the recognition results of each monosyllable and the degree of similarity when "ka-na-ga-wa" is input by voice. It is assumed that the larger the degree of similarity, the closer. In this case, the word "Kanagawa"
and "Kanazawa", the former has a similarity of 80 + 50 + 80 + 80
= 290 Since the latter is 80 + 50 + 10 + 80 = 220, the recognition result is "Kanagawa", which has a higher degree of similarity.

According to this example, it is possible to detect that the second "na" is incorrectly translated into "ma". Based on the result signal DR from dictionary matching 5,
The update instruction section 7 outputs an instruction signal C for updating the standard pattern in the standard pattern memory section 4. That is, instead of the standard pattern "na" which is already in the standard pattern memory section 4, the pattern input this time is
Among the voice patterns P held in the analysis section 2, the pattern of "na" is held as a standard pattern in the standard pattern memory section 4.

In addition to the above methods, several other methods can be considered for updating the standard pattern. For example, a counter is provided for each standard pattern of each monosyllable. When a recognition error of a single syllable is detected by matching with a dictionary, the counter of the erroneously recognized standard pattern is counted up. There is a method of updating this standard pattern with a new pattern when the number of misrecognitions for each standard pattern exceeds a predetermined number of times. In addition to the method of updating the standard pattern by replacing it with a new pattern, it is also possible to create a new standard pattern by taking the average of the input pattern and the standard pattern.

As described above, the embodiments described above are only examples selected for convenience of explanation, and the present invention is not limited to these embodiments. Various other methods can be considered for matching the input monosyllable string with the dictionary.

As mentioned at the beginning, the present invention can also be applied to handwritten character input. In this case, one character may be used as a unit instead of a single syllable. The present invention can be applied to online handwritten character recognition, etc., since a pattern matching method with a standard pattern is considered to be effective.

According to the present invention, in speech and character recognition,
An information input device capable of updating a standard pattern of a predetermined unit to a correct pattern is obtained.

[Brief explanation of drawings]

FIG. 1 is a block diagram of an embodiment of the present invention.
FIG. 2 is a diagram showing an example of the relationship between monosyllable recognition results and similarity. In the figure, 1... Microphone, 2... Analysis department,
3... Voice matching section, 4... Standard pattern memory section, 5... Dictionary matching section, 6... Word dictionary section, 7... Update instruction section.

Claims (1)

[Claims] 1. A recognition unit that recognizes input information such as voice and handwritten characters based on standard patterns registered in advance for each predetermined unit and outputs the recognition result, and a stored word dictionary section; a dictionary matching section that matches the contents of the dictionary section with the recognition result to obtain a word recognition result; An information input device comprising: an update instruction section that instructs updating of the standard pattern based on the determination result.