KR100246617B1 - Speech detection method using the continuous pitch information - Google Patents

Abstract

1. TECHNICAL FIELD OF THE INVENTION

The present invention relates to a voice detection method using continuous pitch information.

2. The technical problem to be solved by the invention

According to the present invention, after determining whether a voice is a voice signal with respect to a signal input from the outside using energy and a zero crossing rate, if it is determined to be a voice signal, the frequency change amount and pitch between continuous pitch information and adjacent pitches are further included. An object of the present invention is to provide a voice detection method for detecting voice using a number of occurrences.

3. Summary of Solution to Invention

The present invention provides a method of detecting pitch information of an audio signal input from the outside, checking whether a detected pitch exists in a predetermined reference pitch area, and calculating a frequency change amount after calculating a frequency change amount to a predetermined first reference value. Checking whether it is smaller and detecting a voice according to each verification result.

4. Important uses of the invention

The present invention is used in a speech recognition system.

Description

Speech detection method using continuous pitch information

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech detection method using pitch information, and more particularly, to a speech detection method in which a speech detection apparatus distinguishes a non-voice from a user's speech by using characteristics of a continuous pitch of a speech signal.

The terms used in the present invention are defined as follows.

Pitch is an attribute of the acoustic sensation, which is mainly related to the frequency of the acoustic stimulus, but also somewhat related to sound pressure or waveform.

Non-voice is a sound that is not a voice, such as a ringtone, a door, a knock on a desk, a dry cough and a clap.

Speech recognition technology is a technology that finds out what the contents of speech are by analyzing the speech spoken by the user. The existing speech recognition system includes a speech detector, a speech feature extractor, a reference pattern comparator, a recognition result verification unit, and a recognition result output unit. In the voice recognition system, the voice detector is responsible for detecting a user's voice.

The existing speech recognition system performs a speech feature extraction process, a reference pattern comparison process, a recognition result verification process, and a recognition result output process when a speech is detected by the speech detector. Therefore, when the voice detector incorrectly detects a sound other than the user's voice signal as a voice, the above-described unnecessary processes are performed, and the voice recognition system performs an operation not desired by the user.

Conventional voice detection methods detect the user's voice using energy and zero crossing ratio of the input signal. Conventional voice detection methods using energy and zero crossing rate are used to detect not only the user's voice but also the surrounding noises such as telephone ringing, desk knocking, door closing, dry cough and applause as voice signals. There are many.

This is because the energy and zero crossing rate of the ambient noise exceed the reference values of the energy and zero crossing rate for the voice previously determined by the voice detector.

The conventional speech detection method using pitch information first detects a primary speech region using energy and zero crossing rate information, and assumes that voiced sound exists in the maximum energy section of the detected region, and then surrounds the maximum energy section. Obtain pitch information from. In addition, it is determined whether the pitch estimated in this section is within a specific coefficient range indicating a pitch presence range of a person. If it is determined that the pitch is within a pitch range of a person, it is determined that voice exists in the input signal. In this case, the focus is on whether pitches occur within the analysis interval. In other words, if the pitch detected in the input signal is within a certain range, it is regarded as voice. However, the sources of ambient noise can be present in all frequency bands because of their wide variety. Therefore, a case in which the ambient noise satisfies the above conditions often occurs, so that false voice detection may occur.

In conclusion, the conventional speech recognition method is performed through the speech detection process, the speech feature extraction process, the reference pattern comparison process, the recognition result verification process, and the recognition result output process, in particular, the energy and zero crossing rate and pitch in the speech detection process. There is a problem in that non-speech, not voice command, is incorrectly detected as voice due to the method using information.

Looking at this in detail.

1 is a flowchart of a conventional voice detection method.

First, after receiving the user's voice signal (10) converts the analog voice signal into a digital signal (11), and calculates the energy and zero crossing rate for the voice data as much as the analysis size (12). Subsequently, it is determined whether the calculation result of the energy and the zero crossing rate is larger than the already set reference value (13). If it is larger than the reference value, the process proceeds to the next step, the pitch detection process. In the pitch detection process, it is assumed that voiced sound exists in the maximum energy section of the detected region, and then pitch information is obtained around the maximum energy section (14). After that, if it is determined that the pitch estimated in this section is within a specific coefficient range representing the pitch presence range of the person (15) and is determined to be within the pitch existence range of the person, after the voice is detected (16), the process ends and does not exceed the reference value. Exit immediately.

This prior art is not only the user's voice but also the surrounding noise, such as the ringing of the phone, the knocking of the desk, the closing door, and the dry cough, etc., exceed the reference values of the energy and zero crossing rate for the predetermined voice and the input signal. In the case where the detected pitch is within a specific range, there is a problem that an unnecessary operation is often performed by recognizing non-voice as a voice signal.

In order to solve the problems described above, the present invention, after determining whether the voice is a voice signal with respect to the signal input from the outside using energy and zero crossing rate, and furthermore, if it is determined that the voice signal, further pitch using continuous pitch information Is a range of a reference pitch, whether the frequency variation of the pitch is within a range of the reference value, and whether the number of pitches has exceeded the reference value. The purpose is to provide.

1 is a flowchart of a conventional voice detection method.

2 is a block diagram of an embodiment of a speech recognition system to which the present invention is applied.

3 is a flowchart illustrating an embodiment of a voice detection method using continuous pitch information according to the present invention;

* Explanation of symbols for the main parts of the drawings

21: voice input device 22: analog / digital converter

23: memory 24: central processing unit

25: recognition result output device

According to another aspect of the present invention, there is provided a voice detection method applied to a voice detection device, comprising: a first step of detecting pitch information of a voice signal input from an external device; A second step of confirming whether the detected pitch exists in a predetermined reference pitch area; A third step of checking whether the calculated frequency change amount is smaller than a predetermined first reference value after calculating the frequency change amount; And a fourth step of determining and detecting a signal satisfying a given condition in each step as a voice signal.

The present invention also provides a voice detection method applied to a voice detection device, comprising: a first step of detecting pitch information of a voice signal input from an external device; A second step of confirming whether the detected pitch exists in a predetermined reference pitch area; A third step of detecting the number of pitch occurrences from the pitch information to determine whether a predetermined reference value is exceeded; And a fourth step of determining and detecting a signal satisfying a given condition in each step as a voice signal.

Hereinafter, an embodiment according to the present invention will be described in detail with reference to FIGS. 2 and 3.

2 is a block diagram of an embodiment of a speech recognition system to which the present invention is applied.

The voice input device 21 receives a user's voice signal from the outside, and the analog / digital converter 22 receives an analog signal output from the voice input device 21, converts the digital signal into a digital signal, and outputs the central signal. The device 23 receives digital voice data from the analog-to-digital converter 22, detects a voice section, extracts a voice feature vector for this region, and stores each vocabulary stored in the feature vector and the storage device 24. After finding the most similar vocabulary by comparing with the reference pattern of these, the verification result is verified and the speech recognition result is output through the recognition result output device 25.

The speech signal used in the speech recognition system consists of a combination of consonants and vowels. Therefore, pitch exists in the audio signal. In general, the pitch of men is lower and that of women is relatively higher than that of men. The voice signal uttered by the user for speech recognition has a unique pitch presence region, although the region varies depending on the person. This can be expressed as (Equation 1).

Pitch Min≤Pitch (t) ≤ Pitch Max

In Equation 1, Pitch Min means a minimum value of pitch for a user voice signal, and Pitch Max means a maximum value of a pitch. On the other hand, non-voice is present in almost all frequency bands. Therefore, it is premature to determine that the user's voice exists in the input signal simply by detecting the pitch in a specific frequency band.

The pitch is generated by the vibration of the vocal cords. When a person using a speech recognition system speaks a normal speech signal, the pitch change is not severe within the speech. In particular, the region of existence of the pitches adjacent to each other, that is, the frequency region, has very similar characteristics. This can be expressed as (Equation 2).

| Pitch (t) -Pitch (t + 1) ｜ ≤ Pitch TH

In Equation 2, Pitch TH is the amount of change between neighboring pitches. On the other hand, most non-negative values are very large. This property helps to distinguish between voice signal and non-voice.

The user's voice signal, unlike non-voice, always satisfies the above two conditions simultaneously. In addition, the number of occurrences of the pitch that satisfies the above two conditions becomes N or more. Since N varies depending on the type of speech signal and the user's speech method, the value of N used in the speech detection apparatus may change according to the use condition of the speech recognition system.

3 is a flowchart illustrating an embodiment of a voice detection method using continuous pitch information according to the present invention.

First, after receiving the user's voice signal (10) converts the analog voice signal into a digital signal (11), and calculates the energy and zero crossing rate for the voice data as much as the analysis size (12). Subsequently, it is determined whether the result of calculating the energy and the zero crossing rate is greater than the reference value that is already set (13). If the reference value is larger than the reference value, the process proceeds to the next step, the pitch detection process. In the pitch detection process, it is assumed that voiced sound exists in the maximum energy section of the detected region, and then pitch information is obtained around the maximum energy section (14). Subsequently, it is determined whether the estimated pitch in this section is within a specific coefficient range representing a human pitch existence range (15). If it is determined to be within a specific coefficient range, the process proceeds to a comparison process of difference in frequency variation, and determines that it is not within a specific coefficient range. If so, terminate the procedure. In the frequency difference difference comparison process, the frequency change amount between neighboring pitch values is calculated (30) to compare whether or not to exceed the set reference value (31). Otherwise exit. In the pitch generation number determination process, after detecting the number of pitches generated by the voice signal (32), it is determined whether the detected pitch number exceeds the reference value (33). Exit otherwise exit immediately.

In the present invention described above, an effect that satisfies the above object can be obtained without performing one of the frequency variation difference comparison process 31 and the pitch occurrence number determination process 33.

In addition, in the present invention, the pitch presence range determination process 15, the frequency variation difference comparison process 31, and the pitch occurrence number determination process 33 are not necessarily performed sequentially, and the same effect may be obtained even if performed in an arbitrary order. Can be.

In addition, the present invention can be variously substituted, modified and changed within the scope without departing from the spirit of the present invention for those skilled in the art to which the present invention pertains to the foregoing embodiments and the accompanying drawings. It is not limited.

The present invention as described above, it is possible to prevent the input of the non-voice signal in addition to the voice command in advance to improve the performance of the speech recognition system, to recognize the non-voice as a voice to perform unnecessary operations that the user does not want This prevents false reactions to non-voice signals.

Claims (6)

In the voice detection method applied to the voice detection device, A first step of detecting pitch information of an audio signal input from the outside; A second step of confirming whether the detected pitch exists in a predetermined reference pitch area; A third step of checking whether the calculated frequency change amount is smaller than a predetermined first reference value after calculating the frequency change amount; And A fourth step of detecting and detecting a signal satisfying a given condition in each step as a voice signal Speech detection method using the continuous pitch information made, including. The method of claim 1, Before performing the fourth step, A fifth step of detecting whether the number of pitch occurrences exceeds the predetermined second reference value by detecting the number of pitch occurrences in the pitch information; Speech detection method using the continuous pitch information further comprises. The method of claim 2, The fifth step, A sixth step of detecting a number of pitch occurrences in the pitch information; A seventh step of comparing whether the detected number of pitch occurrences exceeds the predetermined second reference value; And As a result of the comparison of the seventh step, when the number of pitch occurrences exceeds the predetermined second reference value, the next step is performed; and if the number of pitch occurrences does not exceed the second reference value, an eighth step of ending Speech detection method using the continuous pitch information made, including. The method according to any one of claims 1 to 3, The third step, A ninth step of calculating a frequency change amount from the pitch information; A tenth step of comparing whether the calculated frequency change amount is smaller than the predetermined first reference value; And As a result of the comparison of the tenth step, if the frequency change amount is smaller than the predetermined first reference value, the next step is performed; and if the frequency change amount is not smaller than the first reference value, the eleventh step ends Speech detection method using the continuous pitch information made, including. In the voice detection method applied to the voice detection device, A first step of detecting pitch information of an audio signal input from the outside; A second step of confirming whether the detected pitch exists in a predetermined reference pitch area; A third step of detecting the number of pitch occurrences from the pitch information to determine whether a predetermined reference value is exceeded; And A fourth step of detecting and detecting a signal satisfying a given condition in each step as a voice signal Speech detection method using the continuous pitch information made, including. The method of claim 5, The third step, A fifth step of detecting a number of pitch occurrences in the pitch information; A sixth step of comparing whether the detected number of pitch occurrences exceeds the predetermined reference value; And As a result of the comparison of the sixth step, if the number of pitches exceeds the predetermined reference value, the next step is performed, and if the number of pitches does not exceed the reference value, the seventh step ends. Speech detection method using the continuous pitch information made, including.

Images