JPS59184398A - Voice recognition equipment - 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 speech recognition device.

In recent years, voice recognition devices have come into full-scale practical use as input devices for converters and various control devices. It is already well known that voice recognition devices that can recognize the words spoken by humans have many merits, such as no special 8+1 training is required to use them, and the eyes and hands and feet are not restricted. While word speech recognition devices for so-called specific speakers, which are now in practical use, have such advantages, recognition is performed in units of words or word strings, so they are not suitable for recognition of fixed descriptive sentences, etc. The disadvantage is that it cannot be applied.

In contrast, in Japan, a monosyllable recognition device has appeared that takes advantage of the fact that the language system is composed of phonetic characters. According to this, it is possible to construct any descriptive sentence by uttering multiple monosyllables, but at present the recognition rate of such monosyllable recognition devices, that is, how accurately the uttered sounds can be recognized. This ratio is slightly lower than that of word speech recognition devices. Furthermore, this method has the disadvantage that it can only be applied to languages such as Japanese, which have a multilingual system composed of phonetic characters. On the other hand, from a practical point of view, frequently occurring descriptive sentences are limited to short sentences, and most of them can be included by assuming a few typical fixed descriptive sentence forms. This is a logic that can be applied not only to Japanese but also to other languages, and it means that most fixed-form descriptive sentences can be composed of a limited number of combinations of words without using a monosyllable recognition device. do.

An object of the present invention is to provide a speech recognition device that recognizes fixed predicate sentences according to the grammar of any language.

According to the present invention, it is possible to obtain a speech recognition device that can predict and select words to be recognized sequentially according to grammatical rules so that grammatically impossible large words do not occur, and can recognize fixed-form descriptive sentences with high accuracy.

The present invention will be described in detail below using the drawings of one embodiment of the present invention. The drawing shows one embodiment of the device of the present invention. In the drawing, 100 is a pickup section that converts human voice into an electrical signal, and 200 is a frequency analyzer that analyzes the frequency of the electrical signal obtained by the pickup section lOO. Department, 300
400 is a sampling/quantization unit that samples and quantizes the output of the frequency analysis unit 200;
500 is a judgment unit that compares the sampled and quantized sound □ voice with a pre-stored speech pattern, that is, a standard pattern, and judges the result; The regular large storage section 600 is a standard pattern storage section.

To explain the operation, the voice of the speaker is picked up by the pickup section 10.
0, it is converted into an electrical signal and input to the frequency analysis section 200, where it is divided into power spectra for each frequency band. At this stage, the electrical signals are still analog quantities, but normally in the subsequent sampling/quantization section 300, these power spectra are subjected to sampling and quantization to become digital quantities. The determination unit 400 stores in advance a fixed large storage unit 500.
, the standard pattern to be compared with the output of the sampling/quantization unit 300 is stored in the standard pattern storage unit 60.
Read from 0 and perform computation for comparison and determination. By limiting the standard patterns to be recognized at one time in this way,
It is possible to obtain the advantages of both improved recognition speed and improved recognition rate. The determination unit 400 also has a function of outputting the recognition results obtained in this manner to an arbitrary device. Examples of grammatical rules that should be stored in advance in the fixed form large storage unit 500 include those developed at the word level, such as the basic sentence pattern S (King language) + ■ (Predicate) + 0 (Object) in English. Good morning.

In this explanation, it has been explained that the frequency analysis section 200 and the sampling/quantization section 300 digitize the audio and perform the recognition operation, but depending on the matching algorithm, this method may not necessarily be used. Even in such cases, the intent of the present invention is not impaired in the slightest. Furthermore, in this description, it has been explained that the pickup unit 100 is used to record audio, but it is also possible to use a device such as a tape recorder in addition to a microphone.

As explained above, the present invention is extremely effective in identifying fixed-form descriptive sentences without grammatical errors with high accuracy by using grammatical rules specific to any language.

[Brief explanation of drawings]

The drawing is a configuration diagram showing an embodiment of the present invention. In the figure, 100. Old pickup section, 200.
...Frequency analysis section, 30o...Sampling...
Quantization section, 400... Judgment section, 500...
. . . Regular large storage section, 600 . . . Standard pattern storage section.

Claims (1)

[Claims] 1. A speech recognition device that recognizes a string of uttered input words, wherein the speech recognition device recognizes fixed-form descriptive sentences by sequentially predicting and selecting words to be recognized according to a predetermined grammar.