JPS6235120Y2 - - Google Patents

Description

[Detailed explanation of the idea] The present invention relates to a speech recognition device.

In recent years, interest in voice recognition devices has been increasing, and for example, in the computer field, they are expected to replace conventional keyboard input devices. It goes without saying that if the data input by the operator's voice were directly processed by the computer, the utilization efficiency of the computer would be dramatically increased. This speech recognition device is composed of many functional parts, and the present invention will describe the speech input functional part among them. Examples of the large number of functional parts include a text analysis functional part and the like. Normally, analog audio signals input via a microphone etc.
After undergoing processing such as A/D (analog/digital) conversion and sampling, the data is temporarily stored in a data memory, and then guided to a speech recognition function section. Once each voice is recognized, the structure of the sentence input by voice is analyzed. This is the text analysis function section mentioned above. Analyzing sentences is an extremely difficult operation for machines such as computers. In other words, it is extremely difficult to recognize sentence breaks, such as clause breaks, clause breaks, and sentence ends. For this reason, the speaker (operator) is generally forced to insert a pause at each such break, thereby making the presence of these breaks clear to the machine. However, such speech recognition devices are unnatural to speakers and should be improved.

SUMMARY OF THE PRESENT EMBODIMENT An object of the present invention is to provide a speech recognition device which is less unnatural for the speaker and has high recognition accuracy.

In accordance with the above object, the present invention is characterized in that a break signal generated by the speaker is input to the sentence analysis section for each break in the speech uttered by the speaker.

The present invention will be described below with reference to the drawings.

In general, Japanese is composed of keywords such as particles ``ga'', ``ha'', ``no'', etc. Therefore, the sentence analysis unit often detects these keywords and recognizes the structure of the sentence.Of course, there are equivalent keywords in other languages, and it detects these and recognizes the structure of the sentence. can. Therefore, if so-called break signals including these keywords are given to sentence breaks, clause breaks, clause breaks, sentence ends, and if necessary, phrase or word breaks, it will be useful for the text analysis unit. Needless to say, this will be a big clue in elucidating the structure of sentences. In addition, it is clear that the accuracy of text analysis is dramatically improved. Therefore, in the present invention, the sentence analysis section receives a break signal generated by the speaker at each break in the voice uttered by the speaker. This is illustrated in the attached figure. In the figure, 11 is a microphone that receives analog audio signals;
The output signal from the microphone 11 is sent to the amplifier 12.
The signal is digitized by the A/D converter 13 via the A/D converter 13. This digital data is temporarily stored in the memory 14.
After being stored in the voice recognition unit 15, the voice recognition unit 15 receives the voice recognition unit 15. Here, the digital data is recognized as a syllable whose constituent units are consonants and vowels. Next, each syllable is input to the sentence analysis section 16 to elucidate the structure of the sentence. For example, for the sentence "Today's weather is clear,""KYOUNO
TENKIWA KAISEIDESU".However, in reality, there are no sharp cuts like the one shown above, so in this invention, this is broken down into "KYOUN〓OTENKIW〓AKAISEIDE〓S.
U'', etc., gives information to the text analysis unit 16. These marks mean signals indicating a break, and specifically, the break signal is generated by the switch 17, for example. Each time the speaker's voice reaches a break, the speaker presses the button and, for example, the voice recognition unit 15 is notified.Among the recognized syllables, the syllable corresponding to the press of the switch 17 is A break signal is added.
The sentence analysis unit 16 detects a syllable in the break signal, and when the syllable is detected, it recognizes this as a break in the sentence. As described above, by adding a break signal, the efficiency of text analysis increases dramatically. In this case, the speaker is forced to press the switch, but this is a much more natural and easy operation than the unnatural operation of inserting a pause at each break as described above. be. In the case of the present invention, it is possible that the user forgot to press the switch or pressed the switch incorrectly, but this break signal is only an auxiliary means for text analysis.
This takes over the hardware operations in the text analysis section and is expected to provide faster and more accurate analysis, so even if such a situation occurs, it is fine to rely on the original functions of the text analysis section. There will be no inconvenience or trouble.

As explained above, according to the present invention, rapid and highly accurate sentence analysis can be realized in a speech recognition device using an extremely simple method.

[Brief explanation of the drawing]

The accompanying drawing is a block diagram illustrating the apparatus of the present invention. In this figure, 15 is a speech recognition section, 16 is a text analysis section, and 17 is a switch.

Claims (1)

[Scope of claim for utility model registration] 1. A voice input section for inputting the voice uttered by the speaker;
a speech recognition section that receives the output from the speech input section, decomposes the speech into syllables, and recognizes the speech; and a sentence of the speech uttered by the speaker using each of the syllables outputted from the speech recognition section as input. A speech recognition device comprising: a sentence analysis unit that analyzes a sentence; and a switch that is operated by the speaker to output a break signal for each break in the sentence; In addition to each of the above syllables,
A speech recognition device characterized in that the break signal is also input. 2 The break signal indicates the break of a bunsetsu, the break of a clause,
The speech recognition device according to claim 1, wherein the speech recognition device is generated at either end of a sentence.