JPH03135600A - Voice recognition device - Google Patents

Abstract

PURPOSE:To accurately discriminate plural specific voiced words in Japanese with simple, low-cost constitution by detecting one voice section, then counting how many times following voice sections are detected within a specific time from the end of the voice section, and outputting the counting result as a recognition result. CONSTITUTION:A power detecting means 3 detects the power of an input voice signal and a voice section detecting means 5 detects the voice section according to the detected power. A counting means 6 detects one voice section, then counts how many following voice sections are detected in the specific time from the end of said voice section, and outputs the result as the recognition result. For example, when 'maru' is voiced, the period wherein this word is voiced is one voice section T0. When 'batu' is voiced, on the other hand, the period wherein 'ba' is voiced and the period wherein 'tu' is voiced are detected as two voice sections T1 and T2. Consequently, the specific voiced Japanese words can accurately be discriminated by the practical, simple, and low-cost voice recognition device.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a speech recognition device that can identify a plurality of specific words when they are input as speech.

[Conventional technology]

In recent years, in the field of speech recognition, research has been conducted particularly toward recognition of large words or continuously uttered speech. On the other hand, considering practical issues, current speech recognition devices require restrictions on utterances in order to recognize speech, so even if the number of words that can be recognized is increased, it is not practical enough. It is difficult to say that it is.

From a practical point of view, what is desired is something that can recognize only specific words and, for example, be able to determine whether or not a certain word is being uttered, or be able to distinguish between "hi" and "yay". Of course, if the device is capable of recognizing many words, it will be able to identify two words like the one above, but only two words can be recognized.

For purposes of determination alone, such devices have generally been complex and expensive.

Incidentally, US Pat. No. 3,688,126 discloses a practical, simple and low-cost device intended to identify only two words as described above. This device is designed to identify input voices rYEsJ and "NO" based on the difference in their frequency distributions. In other words, focusing on the fact that the "S" sound in "YES" is a sibilant sound and has a high frequency component, it is possible to determine whether there is a high frequency component at the end of the voice section or there is no high frequency component in the entire voice section. Depending on the situation, the former is rYES, the latter is rNo,
It is designed to be recognized and identified as such.

[Problem to be solved by the invention]

However, when trying to apply the above-mentioned speech recognition device capable of distinguishing rYES and rNo to the recognition of Japanese speech,
None of these have high frequency components, so they cannot be effectively identified.Therefore, although the above speech recognition device is suitable for identifying English speech, it is not suitable for Japanese speech. There was a problem that it could not be used.

An object of the present invention is to provide a practical, simple, and low-cost speech recognition device suitable for identifying specific Japanese speech.

[Means to solve the problem]

In order to achieve the above object, the present invention includes a power detecting means for detecting the power of an input sound signal, a speech section detecting means for detecting a speech section based on the detected power, and a speech section detecting means for detecting one speech. The present invention is characterized by comprising a counting means for counting the number of times the next voice section is detected within a predetermined time from the end of the voice section after detecting the section, and outputting the counting result as a recognition result.

[Effect]

In the above configuration, the power detection means detects the power of the input sound signal, and the speech interval detection means detects the speech interval based on the detected power. If the number of plosives included in these sounds is different from each other, the number of voice segments detected until the utterance of the entire word is finished is different from each other.

That is, if there is a plosive, a silence state occurs within a predetermined period of time, and a speech interval is separated from the utterance of the entire word. Focusing on this, the counting means, after detecting one speech section, counts the number of times the next speech section is detected within a predetermined time from the end of the speech section, and outputs this result as a recognition result. . This allows the sounds of multiple words to be identified.

〔Example〕

Hereinafter, one embodiment of the present invention will be described based on the drawings.

FIG. 1 is a block diagram of an embodiment of a speech recognition device according to the present invention.

The present invention can be expressed as "hai", "iiie" and "yoi" in Japanese.
"Maru" is used as a word to express a pair such as "warui" and "warui".
(“O”) and “bara” or “beke” (“x”)
Referring to FIG. 1, the speech recognition device of this embodiment includes a converter 1 such as a microphone that converts input sound into an electric signal, and a converter 1 that converts input sound into an electric signal. an amplification section 2 that amplifies the sound signal, a power detection section 3 that detects the power or energy of the sound signal, and a reference value storage section 4 that stores the power of ambient noise before inputting the sound as a reference value. , a voice section detecting section 5 which compares the power detected by the power detecting section 3 with a reference value stored in the reference value storage section 4 and detecting a voice section of the input voice; After one voice section is detected, the counting section 6 counts the number of times the next voice section is detected within a predetermined time from the end of the voice section, and outputs a recognition result according to the counting result.

For example, as shown in FIG. 2(a), the power detection unit 3
This is realized by a rectifier circuit 10 and a low-pass filter 1, and a specific example of the circuit is shown in FIG. 2(b). In other words, the sound signal from the amplifier section 2 is transferred to the diodes D and D2.
Rectified by a rectifier circuit 10 consisting of a resistor R and a capacitor C.
The power of the sound signal is detected by filtering it with a low-pass filter 11 consisting of the following.

Further, the counting section 6 is constituted by, for example, a counter.

Next, an example of the operation of the speech recognition device having such a configuration will be explained.

First, let us consider the case where the sounds of the opposing words ``maru'' and ``bara'' or ``beke'' are recognized and distinguished from each other. When any of the words mentioned above is uttered, the voice section detecting section 5 detects the voice section based on the power of the input sound detected by the power detecting section 3. That is, when the power of the sound signal detected by the power detection section 3 becomes larger than the reference value stored in the reference value storage section 4, that is, the power of the surrounding noise, the speech section detecting section 5 detects the power of the sound signal in the speech section. Detected as the start end. Note that the reference value stored in the reference value storage unit 4 is preferably one that is the average of the power data of the surrounding noise for several milliseconds before the actual speech is input. Can be detected accurately. After the start of the voice section is detected, the reference value is not updated, and after the start of the voice section is detected, the voice section detection unit 5 determines whether the power of the sound signal has become smaller than the reference value. is detected, and when it becomes smaller than a reference value, this is detected as the end of the voice section.

By the way, when "maru" is uttered, the power of this voice becomes as shown in FIG. 3(a), and the period during which this voice is uttered is one voice section T.

If you say ``bara'' in response to this, this sound will include a plosive ``t'', and because you close your lips once to produce the plosive, before you pronounce the later sound ``tsu'' There is a short period of silence, that is, a state where the power is close to 0'', and as a result, as shown in FIG. Two voice sections T1 and T2 are detected.

Similarly, when ``beke'' is uttered, this voice includes the plosive sound ``k'', so as shown in Figure 3 (C), the period of uttering ``beke'' and the period when ``beke'' is uttered are different. ” is uttered and two voice sections T3. T4 is detected. The period of silence between these two speech intervals can occur when the following sound is a consonant, or when the plosive sounds “t”, ′, ρ”, etc.
However, in the case of a consonant, the period of silence is approximately 350 milliseconds, so in the case of a plosive, the period of silence is shorter than this.

That is, in the above example, after the end of the first voice section T1°T3 is detected, the subsequent voice section T2. The starting edge of T4 will be detected.

In the counting unit 6, after the end of one voice section is detected,
It monitors whether the start of the next voice section is detected within the above time, and when the start of the next voice section is detected within the above time, the count value is incremented by "1", and within the above time When the start of the next voice section is no longer detected, it is determined that the entire word has been uttered, the counting operation is stopped, and the counting results up to that point are output as recognition results. Therefore, when "maru" is uttered, one vocal section T
. Since it is a bamboo, the counting result is “O” and “rose”.
When ``beke'' or ``beke'' is uttered, the counting result will be ``1'' depending on the starting point of the subsequent vocal section T2 or T4.This makes it possible to easily determine the opposition between ``maru'' and ``bara'' or ``bege.'' It becomes possible to accurately recognize and distinguish between two Japanese sounds.

In the above explanation, "maru" and "bara" or "pege" were taken as examples, but the speech recognition device of the present invention is not limited to this.
For example, if only one of the opposing words contains a plosive, such as "zero"("0") and "ichi"("1"), it can be applied to the identification of all opposing words.

It is also possible to identify not only two Japanese words but also three or more Japanese words.
For example, consider the case of identifying three words: ``maru''(``○''),``bara'' or ``bege''(``X''), and ``sankafu''(``△''), which represents the middle word. . When uttering ``sankafu'', this voice contains two plosive sounds ``k'', so as shown in Figure 4, the period during which ``san'' is uttered and the period during which ``power'' is uttered are divided. Three voice sections Ts: the period when the voice is uttered, and the period when the ``ri'' is uttered.
, Ts, and T7 are detected by the voice section detection section 5.

As a result, the counting result of the counting unit 6 when uttering "sankafu" is "2", and the counting result is "0" for "maru", and the counting result is "1" for "bara" or "peke".
It becomes possible to identify the

〔Effect of the invention〕

As explained above, according to the present invention, after detecting one speech section, the number of times the next speech section is detected within a predetermined time from the end of the speech section is counted, and this counting result is used as the recognition result. Since it is output, it is possible to accurately identify a plurality of specific Japanese sounds with a simple and low-cost configuration.

FIG. 4 is a diagram showing the time course of the power of the voice when the word "R Sancuff" is uttered.

DESCRIPTION OF SYMBOLS 1... Conversion section, 2... Amplification section, 3... Power detection section, 4... Reference value storage section, 5... Voice section detection section,
6...Counting section

Claims (1)

[Claims] power detection means for detecting the power of the input sound signal;
A voice section detecting means detects a voice section based on the detected power, and after the voice section detecting means detects one voice section, the number of times the next voice section is detected within a predetermined time from the end of the voice section is determined. A speech recognition device comprising a counting means for counting and outputting the counting result as a recognition result.