KR100322203B1 - Device and method for recognizing sound in car - Google Patents

Abstract

The present invention provides a voice recognition device and method for detecting a starting point and an end point of a voice section by receiving a voice command of a driver in a vehicle, and in particular, a vehicle detecting a voice section by using a zero crossing rate and energy from all input signals. The present invention relates to a speech recognition apparatus and a method thereof. According to the speech recognition apparatus and method of the vehicle according to the present invention, since the speech section is detected by using the zero crossing rate and energy of the input speech data frame, the background noise is severe. Compared to the conventional method using only energy in the environment, it is possible to detect an accurate speech section, thereby improving the speech recognition performance.

Description

Voice recognition device of vehicle and its method {DEVICE AND METHOD FOR RECOGNIZING SOUND IN CAR}

The present invention relates to a vehicle voice recognition device and a method for detecting a start point and an end point of a voice section by receiving a voice command of a driver in a vehicle. Particularly, the present invention relates to a voice section using zero crossing rate and energy in the entire input signal. The present invention relates to a voice recognition device for detecting a vehicle and a method thereof.

In general, the vehicle speech recognizer uses a technology that extracts only the pure speech segment except the noise segment from the entire speech segment, and the conventional speech recognizer uses only energy, so it is accurate in the consonant segment, which is the front part of the speech section, and in the rear part section. There was a problem that the speech recognition rate is considerably lowered due to no detection. More specifically, unvoiced burst sounds such as ㅋ, ッ and ㅌ of Korean have high frequency and low energy, so it is difficult to detect the voice segment using energy alone.

Accordingly, the present invention has been made to solve the above problems, and an object of the present invention is to provide a speech recognition device and a method of a vehicle that can obtain accurate speech recognition results.

In order to achieve the above object, the voice recognition device of a vehicle according to the present invention receives a voice of a vehicle driver, and in the voice recognition device of a vehicle installed in a vehicle to detect a start point and an end point of a voice section, the voice recognition operation A voice recognition start switch for advancing the vehicle, a background noise dedicated microphone for inputting background noise of the vehicle, a voice input dedicated microphone for inputting a driver's voice when the voice recognition start switch is turned on, and an ON of the voice recognition start switch In operation, a voice signal including a background noise and a driver's voice signal is received from the microphone and the voice input microphone, and converted into a digital signal. The voice is divided into frames, and the energy of the divided voice signal Characterized in that it consists of a control means for detecting the start point and the end point of the speech segment using the zero crossing rate do.

In order to achieve the above object, a voice recognition method of a vehicle according to the present invention is a voice recognition method for detecting a start point and an end point of a voice section by receiving a voice of a vehicle driver. The first step of generating a sound to inform the start of the input through the input, and the voice data including the background noise and the driver's voice signal through the background noise dedicated microphone and the voice input dedicated microphone to receive the voice data in a frame unit of a certain size A second step of dividing, a third step of extracting an average energy of a silent section (five frames) after the voice recognition switch is turned on in the divided voice data section and before the driver's voice is input; A fourth step of extracting a reference value of energy by adding a minimum value of energy to an average energy of a section; and using the reference value of energy A fifth step of calculating a zero crossing rate of each frame of the input voice data, a sixth step of calculating an average value of the zero crossing rate of the first five frames of the input voice data, and a seventh step of setting the variable N to 1 And an eighth step of determining whether or not the zero crossing rate of each frame of the input voice data is equal to or greater than the average value of the zero crossing rate; ) Is set to 0, a tenth step of determining whether the zero crossing rate of each frame of the input voice data is equal to or greater than the average value of the zero crossing rate, and the zero crossing rate of each frame of the voice data in the tenth step. When the zero crossing rate average value is equal to or greater than the average value of the zero crossing rate, the variables B and N are each increased by one, and then the eleventh step of determining whether the variable B becomes five and the variable (in the eleventh step) If B) is not 5, the process proceeds to the tenth step, while if the variable B is 5, the starting point of the speech section is determined by subtracting 5 from the variable N to a value obtained by subtracting 5 from the variable N; In step 13, it is determined whether the zero crossing rate of each frame of the data is smaller than the average value of the zero crossing rate; In step 14, if the zero crossing rate of each frame of voice data is smaller than the average value of the zero crossing rate, and in the fifteenth step, the zero crossing rate of each frame of the voice data is smaller than the average value of the zero crossing rate. (S, N) is increased by one each, and the sixteenth step of determining whether the variable (S) is 20, and if the variable (S) is not 20 in the sixteenth step proceeds to the fifteenth step , Prize When the variable S reaches 20, the end point of the voice segment is determined as the 17th step of determining the value obtained by subtracting 20 from the variable N.

1 is a control block diagram of a voice recognition device of a vehicle according to an embodiment of the present invention.

2A to 2C are flowcharts for describing a voice recognition method of a vehicle according to an embodiment of the present invention.

3 is a view for explaining the principle of the zero crossing rate applied to the present invention.

<Explanation of symbols for the main parts of the drawings>

100: microphone for background noise 110: microphone for voice input

120: voice recognition start switch 200: control means

300: driving device of the vehicle 310: speaker

Hereinafter, a voice recognition apparatus and a method thereof for a vehicle according to an embodiment of the present invention will be described in detail with reference to the accompanying drawings.

1 is a control block diagram of a voice recognition device for a vehicle according to an embodiment of the present invention, the voice recognition device for a vehicle according to an embodiment of the present invention is a voice recognition start switch 120, a background noise-only microphone 100 ), A voice input dedicated microphone 110, the control means 200 and the speaker 310.

The voice recognition start switch 120 serves to advance the voice recognition operation, the background noise-only microphone 100 is a device for inputting the background noise of the vehicle, the voice input only microphone 110 is When the voice recognition start switch 120 is turned on, the driver's voice can be input.

The control means 200 receives the background noise and the driver's voice signal input from the background noise-only microphone 100 and the voice input-only microphone 110, and amplifies them to a predetermined level and converts them into digital signals. It consists of an amplifier, a DSP chip with a built-in control program, and memory elements for storing data.

In particular, the control means 200 is a voice signal including the background noise and the driver's voice signal from the background noise dedicated microphone 100 and the voice input microphone 110 during the on operation of the voice recognition start switch 120 After converting the signal into a digital signal, the voice is divided into frames, and the start point and end point of the voice section are detected by using the energy and the zero crossing rate of the divided voice signal.

In addition, the control means 200 outputs a driving control signal for controlling the driving device 300 and the speaker 310 of the vehicle after the voice recognition is performed.

The speaker 310 receives a control signal output from the control means 200 during the on operation of the voice recognition start switch 120 and generates a sound indicating an input start.

A speech recognition apparatus and a method of a vehicle according to an embodiment of the present invention having the above configuration will be described.

2A to 2C are flowcharts illustrating a voice recognition method of a vehicle according to an embodiment of the present invention, where S represents a step.

First, when the voice recognition switch 120 is turned on (S1), the control means 200 generates a sound indicating the start of the input through the speaker 310 (S2).

Subsequently, the control unit 200 receives the voice data including the background noise and the driver's voice signal through the background noise-only microphone 100 and the voice input-only microphone 110 (S3), and provides a frame unit having a predetermined size ( Example: 10ms) to divide the voice data (S4).

Subsequently, the control unit 200 extracts the average energy of the silent section (five frames) after the voice recognition switch 120 is turned on among the sections of the divided voice data and before the driver's voice is input (S5). .

The average energy of the silent period is equal to Equation 1, and the short term average energy E (N) is equal to Equation 2 when A / D is converted to a sampling frequency of 10 kHz.

(Equation 1)

(Equation 2)

N = 0, 1, 2, ...

Subsequently, the control means 200 extracts a reference value of energy by adding a minimum value of energy to the average energy of the silent section (S6), and calculates a zero crossing rate of each frame of the input voice data using the reference value of the energy. (S7).

In step S6, the reference values ZCR _UP and ZCR _DOWN of energy are expressed by Equation 3 below.

(Equation 3)

ZCR _UP , ZCR _DOWN = ± (mean energy +20)

As shown in FIG. 3, the zero crossing rate of each frame of the voice data input in the step S7 is such that the change in the signal value of each frame crosses the zero axis (0) to the upper reference value (ZCR _UP ) or the lower reference value (ZCR). _DOWN ) is set to 1 when including one or more, and is set to 0 when the change in the signal value does not include any upper or lower reference value across the zero axis.

Subsequently, the control unit 200 calculates a zero crossing rate average value ZCR _TH of the initial five frames of the input voice data (S8), sets an arbitrary variable N to 1 (S9), and It is determined whether or not the zero crossing rate {ZCR (N)} of each frame of the voice data is equal to or greater than the zero crossing rate average value ZCR _TH (S10).

When the zero crossing rate {ZCR (N)} of each frame is equal to or greater than the zero crossing rate average value ZCR _TH in step S10 (YES), the control means 200 sets an arbitrary variable B to zero. Subsequently (S11), it is determined whether or not the zero crossing rate {ZCR (N)} of each frame of the input voice data is equal to or greater than the zero crossing rate average value ZCR _TH (S12).

When the zero crossing rate {ZCR (N)} of each frame of the voice data becomes equal to or greater than the zero crossing rate average value ZCR _TH in step S12 (YES), the control means 200 sets the variables B and N, respectively, one by one. After increasing (S13), it is determined whether the variable (B) becomes 5 (S14).

If the variable B does not become 5 in step S14 (NO), the process proceeds to the step S12, while when the variable B becomes 5 (YES), the control means 200 generates a voice section. After determining the starting point of the value as a value obtained by subtracting 5 from the variable N (S15), it is determined whether the zero crossing rate {ZCR (N)} of each frame of the voice data is smaller than the zero crossing rate average value (ZCR _TH ). (S16).

In step S16, if the zero crossing rate {ZCR (N)} of each frame is smaller than the average zero crossing rate value (ZCR _TH ) (YES), the control means 200 sets an arbitrary variable S to zero. Subsequently (S17), it is determined whether or not the zero crossing rate {ZCR (N)} of each frame of voice data is smaller than the zero crossing rate average value ZCR _TH (S18).

In step S18, if the zero crossing rate {ZCR (N)} of each frame of the voice data is smaller than the zero crossing rate average value ZCR _TH (YES), the variables S and N are each increased by one (S19), It is determined whether the variable S is 20 (S20).

If the variable S does not reach 20 in step S20 (NO), the process proceeds to step S18, while when the variable S reaches 20 (YES), the control means 200 generates a voice section. The end point of is determined as a value obtained by subtracting 20 from the variable N (S21), and then ends.

On the other hand, if the zero crossing rate {ZCR (N)} of each frame is smaller than the zero crossing rate average value ZCR _{TH in} the steps S10 and S12 (N0), the control means 200 increases the variable N by one. After step S22, the process proceeds to step S10.

On the other hand, if the zero crossing rate {ZCR (N)} of each frame in the steps (S6, S8) is greater than or equal to the average value of the zero crossing rate (ZCR _TH ) (NO), the variable N is increased by one (S23). The flow proceeds to step S16.

As described above, according to the speech recognition apparatus and method of the vehicle according to the present invention, since the speech section is detected using the zero crossing rate and the energy of the input speech data frame, only the conventional energy is used in an environment with high background noise. Compared to the method, it is possible to detect an accurate speech section, thereby improving the speech recognition performance.

Claims (7)

Claim 1 has been deleted. Claim 2 has been deleted. In the voice recognition method of the vehicle for receiving the voice of the vehicle driver to detect the start point and the end point of the voice section, A first step of generating a sound for notifying input start through a speaker when the voice recognition switch is turned on; A second step of receiving the voice data including the background noise and the driver's voice signal through the microphone for background noise and the microphone for voice input and dividing the voice data into frame units of a predetermined size; A third step of extracting an average energy of a silent section (five frames) after the voice recognition switch is turned on among the divided voice data sections and before the driver's voice is input; A fourth step of extracting a reference value of energy by adding a minimum value of energy to the average energy of the silent section; A fifth step of calculating a zero crossing rate of each frame of voice data input using the energy reference value; A sixth step of calculating an average value of zero crossing rates of the first five frames of the input voice data; The seventh step of setting the variable N to 1, An eighth step of determining whether the zero crossing rate of each frame of the input voice data is equal to or greater than the average value of the zero crossing rate; A ninth step of setting the variable B to 0 when the zero crossing rate of each frame is equal to or greater than the average value of the zero crossing rate in the eighth step; A tenth step of determining whether the zero crossing rate of each frame of the input voice data is equal to or greater than the average value of the zero crossing rate; An eleventh step of determining whether or not the variable B becomes 5 when the zero crossing rate of each frame of the voice data is equal to or greater than the average value of the zero crossing rate in the tenth step, and then increase the variables B and N one by one; , If the variable B does not become 5 in the eleventh step, the process proceeds to the tenth step. On the other hand, when the variable B becomes five, the start point of the voice interval is determined as a value obtained by subtracting 5 from the variable N. 12th step, A thirteenth step of determining whether the zero crossing rate of each frame of voice data is smaller than the average value of the zero crossing rate; A fourteenth step of setting the variable S to 0 when the zero crossing rate of each frame is smaller than the average value of the zero crossing rate in the thirteenth step; A fifteenth step of determining whether the zero crossing rate of each frame of the voice data is smaller than the average value of the zero crossing rate; In the fifteenth step, if the zero crossing rate of each frame of the voice data is less than the average value of the zero crossing rate, each of the variables S and N is increased by one, and step 16 of determining whether the variable S is 20; If the variable S is not 20 in the sixteenth step, the process proceeds to the fifteenth step. If the variable S is 20, the end point of the speech section is determined as a value obtained by subtracting 20 from the variable N. Voice recognition method of a vehicle, characterized in that consisting of a seventeenth step. The method of claim 3, wherein In the fourth step, the reference value of energy is ± (average energy of silence period + 20), characterized in that the vehicle voice recognition method. The method according to claim 3 or 4, The zero crossing rate of each frame of the voice data input in the fifth step is 1 when the change in the signal value of each frame includes one or more upper or lower reference values across the zero axis, and the change in the signal value crosses the zero axis. The voice recognition method of a vehicle, characterized in that it is determined as 0 if none of the upper or lower reference values are included. The method of claim 3, wherein In the eighth and tenth step, if the zero crossing rate of each frame is smaller than the average value of the zero crossing rate, the vehicle is characterized in that the eighteenth step is performed by increasing the variable N by one and then proceeding to the eighth step. Voice recognition method. The method of claim 3, wherein When the zero crossing rate of each frame is equal to or greater than the average value of the zero crossing rate in the thirteenth and fifteenth steps, a nineteenth step in which the variable N is increased and then proceeds to the thirteenth step is added. Voice recognition method.