TW201627990A - Time domain based voice event detection method and related device - Google Patents

Abstract

A time domain based voice event detection method is provided. The voice event detection method includes converting an analog audio signal into an audio digital signal according to a sample rate and defining a plurality of sample points as a frame, wherein a sample value of each sample point of the audio digital signal is in decibel, for each sample point of the frame, calculating average difference of a plurality of different sample point groups corresponding to one sample point to generate a plurality of average difference values, determining whether the sample point is a candidate sample point of a voice event according to the plurality of average difference values and determining whether a voice event occurs in the frame to generate a determination result according to the ratio of the number of determined candidate sample points of the frame to the total number of sample points of the frame, and outputting a control signal for controlling an application function according to the determination result.

Description

Sound event detecting method based on time domain operation and related device

The invention relates to a sound event detecting method and related device, in particular to a sound event detecting method and related device based on time domain operation and without complicated complex Fourier transform operation.

Sound detection technology is one of the important processing procedures in speech recognition applications. However, the conventional approach usually has to perform a large and complex conversion calculation process, such as performing a Fast Fourier Transform (FFT) operation to convert the audio signal from the time domain to the frequency domain. Perform related operations, which will consume too much system resources and conversion operation time and is not suitable for immediate applications. Therefore, the prior art is in need of improvement.

In order to solve the above problems, the present invention provides a voice event detecting method and related device based on time domain operation, which can effectively reduce the amount of system computation and is suitable for real-time applications.

The invention discloses a sound event detecting method based on time domain operation, which comprises: converting an audio analog signal into an audio digital signal according to a sampling rate, and sampling values of each sampling point are in decibels (dB); Defining a certain number of sampling points as a sound box, for each sampling point in the sound box, calculating an average difference of a plurality of different sampling groups corresponding to one sampling point, Generating a plurality of average difference values; determining, according to the plurality of average difference values, whether the sampling point is a candidate sampling point of a sound event, and occupying the sound according to the number of sampling points determined as candidate sampling points in the sound box The ratio of the total number of sampling points of the frame determines whether the sound event occurs in the sound box to generate a determination result; and outputs a control signal according to the determination result to control an application function.

The invention further discloses a sound event detecting device based on time domain operation, comprising: a type of comparison digital conversion unit for converting an audio analog signal into an audio digital signal according to a sampling rate, and the audio digital signal The plurality of sampling points are defined as a sound box, and the sampling value of each sampling point is in decibels (dB); a calculating unit is used to calculate a sampling point corresponding to each sampling point in the sound box. The average difference value of the plurality of different sampling groups to generate a plurality of average difference values; a determining unit configured to determine, according to the plurality of average difference values, whether the sampling point is a candidate sampling point of a sound event, and Determining whether the sound event occurs in the sound box to generate a judgment result according to a ratio of the number of sampling points determined to be candidate sampling points in the sound box to the total number of sampling points of the sound box; and a control unit, According to the judgment result, a control signal is output to control an application function.

10‧‧‧Sound event detection device

102‧‧‧ analog-to-digital conversion unit

104‧‧‧Computation unit

106‧‧‧judging unit

108‧‧‧Control unit

20‧‧‧ Process

202, 204, 206, 208‧‧ steps

Average difference between AD1, AD2, AD3‧‧

SP1, SP2, SP3, SP4, SP5, SP6, SP7, SP 8‧‧‧ sampling points

TH1, TH2‧‧‧ threshold

V1, V2, V3, V4, V5, V6, V7, V8‧‧‧ sample values

FIG. 1 is a schematic diagram of a sound event detecting apparatus according to an embodiment of the present invention.

FIG. 2 is a schematic diagram of a method for detecting a sound event according to an embodiment of the present invention.

FIG. 3 is a schematic diagram of a sound box of an audio digital signal according to an embodiment of the present invention.

FIG. 4 is a schematic diagram of corresponding average difference values and statistical results of sampling points according to an embodiment of the present invention.

FIG. 5 is a schematic diagram of candidate sampling points in a sound box according to an embodiment of the present invention.

Please refer to FIG. 1 , which is a sound event detecting apparatus 10 according to an embodiment of the present invention. Schematic diagram. The sound event detecting device 10 includes a type of comparison digital bit converting unit 102, a calculating unit 104, a determining unit 106, and a control unit 108. The analog-to-digital conversion unit 102 is configured to convert an audio analog signal into an audio digital signal according to a sampling rate, and define a plurality of sampling points of the audio digital signal as a sound box, wherein each sampling point of the audio digital signal The value of the sampled value is in decibel (dB). The calculating unit 104 is configured to calculate an average difference of a plurality of different sampling groups corresponding to a sampling point for each sampling point in each of the audio digital signals to generate a plurality of average difference values. The determining unit 106 is configured to determine, according to the plurality of average difference values, whether the sampling point is a candidate sampling point of the sound event, and the number of sampling points determined as the candidate sampling point in the sound box accounts for the total number of sampling points of the sound box. The ratio determines whether a sound event has occurred in the sound box. The control unit 108 is configured to output a control signal according to the determination result to control a specific application function.

In short, the sound event detecting device 10 can perform correlation processing on the sampling value of the sampling point of the audio digital signal in the time domain to determine whether a sound event occurs, and output the relevant control signal accordingly. Controls the execution of related application functions. In this way, the present invention can determine the sound event based on the time domain operation without complicated FFT operation, and can effectively reduce the amount of system computation and is suitable for real-time applications, such as the application of an instant embedded system.

For the operation mode of the sound event detecting device 10, please refer to FIG. 2, which is a schematic diagram of a sound event detecting process 20 according to an embodiment of the present invention. The sound event detection process 20 includes the following steps: Step 202: Convert the audio analog signal into an audio digital signal according to the sampling rate.

Step 204: Calculate, for each sampling point in the sound box, an average difference of a plurality of different sampling groups corresponding to one sampling point to generate a plurality of average difference values.

Step 206: Determine, according to the plurality of average difference values, whether the sampling point is a candidate sampling point of the sound event, and according to the number of sampling points that are determined as candidate sampling points in the sound box. The ratio of the total number of sampling points of the sound box determines whether a sound event occurs in the sound box.

Step 208: Output a control signal according to the judgment result to control the application function.

The above steps will be described in detail. First, in step 202, after receiving the audio analog signal, the analog-to-digital conversion unit 102 converts the audio analog signal into an audio digital signal according to a sampling rate. The audio digital signal can be divided into a sound box form for subsequent processing, that is, a plurality of sampling points of the audio digital signal can be defined as a sound box, and the sample value of each sampling point of the audio digital signal is It is in decibels (dB), and the sampling point length of each frame can be set according to requirements.

In step 204, for each sampling point in each of the audio digital signals, the calculating unit 104 calculates an average difference of a plurality of different sampling groups corresponding to each sampling point of each of the sound frames to generate a plurality of samples. Average difference value.

Further, in an embodiment, for each sampling point in each of the sound frames, the calculating unit 104 calculates an average difference of the plurality of different sampling groups corresponding to a sampling point according to the plurality of reference sampling rates, The plurality of average difference values are generated. Preferably, the reference sampling rate is less than the sampling rate of the audio digital signal.

More specifically, for a plurality of different sampling groups corresponding to a sampling point, the calculating unit 104 calculates the sampling between the sampling points and the adjacent sampling points separated by a corresponding sampling point according to each reference sampling rate. Absolute difference in value and absolute difference between sample points between at least one sample point after the sample point and adjacent sample points spaced apart from each other, and averaged the absolute difference results to produce a corresponding The average difference value. The sampling point interval of the corresponding sampling point interval is equal to the ratio of the sampling rate of the audio digital signal to the corresponding reference sampling rate. For example, please refer to FIG. 3, which is an audio digital signal according to an embodiment of the present invention. A schematic diagram of a sound box F. It should be noted that, for convenience of explanation, as shown in FIG. 3, the sampling point length of the sound box F is 8 sampling points, but not limited thereto, the sampling point length of the sound box F may be 256 sampling points, 512 sampling points...etc., depending on system requirements. Assume that the sampling rate of the audio digital signal is 8 kHz (ie, 8000 sampling points per second). The reference sampling rates are 8 kHz, 4 kHz, and 2.67 kHz, respectively.

As shown in FIG. 3, for the sampling point SP1, the sampling points SP2 to SP4 corresponding to the sampling point SP1 and the sampling point SP are first selected in the frame F. Further, if the reference sampling rate is 8 kHz, the length of the corresponding sampling point interval is equal to 1 sampling point interval (ie, 8 kHz/8 kHz = 1). Next, the calculating unit 104 calculates the difference between the sampling values of the sampling points SP1 to SP4 and the sampling values of the adjacent sampling points separated by one sampling point interval to generate an average difference value AD1. For example, calculating the absolute difference between the sampling point SP1 and the sampling value between the adjacent sampling points SP2 separated by one sampling point interval, and calculating the sampling between the sampling point SP2 and the adjacent sampling point SP3 separated by one sampling point interval. The absolute difference of the values is calculated, and the absolute difference between the sampling point SP3 and the sampling point SP4 separated by one sampling point interval is calculated, and the sampling point SP4 is calculated between the sampling point SP4 and the sampling point SP5 spaced apart from one sampling point. The absolute difference of the sampled values. Then, the calculated absolute differences are averaged to generate an average difference value AD1, wherein the average difference value AD1 is as shown in the formula (1).

For the sampling point SP1, if the reference sampling rate is 4 kHz, the length of the corresponding sampling point interval is equal to 2 sampling point intervals (ie, 8 kHz / 4 kHz = 2). The calculating unit 104 calculates the difference between the sampling values of the sampling points SP1 to SP4 and the sampling values of the adjacent sampling points separated by two sampling point intervals, respectively, to generate an average difference value AD2. For example, calculating the absolute difference between the sampling point SP1 and the sampling value between the adjacent sampling points SP3 separated by two sampling point intervals, and calculating the sampling between the sampling point SP2 and the adjacent sampling point SP4 separated by two sampling point intervals. The absolute difference of the values, the absolute difference between the sampled point SP3 and the sampled value between the adjacent sample points SP5 of the interval between the two sample points is calculated, and the sample point is calculated. The absolute difference between the sampled value between SP4 and the adjacent sampling point SP6 spaced apart by 2 sampling points. Then, the calculated absolute differences are averaged to generate an average difference value AD2. The average difference value AD2 is as shown in the formula (2).

Similarly, for the sampling point SP1, if the reference sampling rate is 2.67 kHz, the length of the corresponding sampling point interval is equal to 3 sampling point intervals (ie, 8 kHz / 2.67 kHz = 3). The calculating unit 104 calculates the difference between the sampling values of the sampling points SP1 to SP4 and the sampling values of the adjacent sampling points separated by three sampling point intervals, respectively, to generate an average difference value AD3. For example, the absolute difference between the sampling point SP1 and the sampling value between the adjacent sampling points SP4 separated by three sampling point intervals is calculated, and the sampling between the sampling point SP2 and the adjacent sampling point SP5 separated by three sampling point intervals is calculated. The absolute difference of the values is calculated, and the absolute difference between the sampled point SP3 and the sampled value between the adjacent sample points SP6 separated by 3 sample points is calculated, and the adjacent sample of the sample point SP4 and the interval of 3 sample points is calculated. The absolute difference between the sample values between points SP7. Then, the calculated absolute differences are averaged to generate an average difference value AD3. The average difference value AD3 is as shown in the formula (3).

Next, in step 206, the determining unit 106 evaluates whether the calculated sampling point is a candidate sampling point of a sound event according to the average difference result calculated by the calculating unit 104. That is, after calculating the average difference value corresponding to each reference sampling rate for each sampling point in each of the sound frames, the determining unit 106 respectively compares the calculated average difference value with a threshold value TH1. And count the number of the average difference values that is greater than the threshold TH1, as shown in FIG. Since the calculated average difference value is greater than the threshold value TH1, it indicates that there is a possibility of a sound event in the signal. Therefore, it is judged based on the threshold value TH1 that there may be a sound event. Moreover, for the same sampling point, when the average difference value calculated corresponding to the different reference sampling rates is greater than the number of the threshold TH1, the probability that there is a sound event in the signal is higher. In this case, when the number of the average difference values greater than the threshold value TH1 is greater than the threshold value TH2, the determining unit 106 determines that the sampling point is a candidate sampling point. For example, if the threshold TH2 is one, as shown in FIG. 4, for the sampling point SP1, if the average difference value AD1 to AD3 calculated based on the three different reference sampling rates, the average difference value AD1 , AD, AD3 are all greater than the threshold TH1. That is to say, for the sampling point SP1, the number of the calculated average difference values that is greater than the threshold value TH1 is three, that is, the calculated average difference values are greater than the threshold. The number of values TH1 is greater than the threshold TH2 (ie, 3>TH2=1). Therefore, the sampling point SP1 is judged and classified as a candidate sampling point, as shown in FIG. 5 (solid is a candidate point, hollow is not Candidate point).

Next, referring to FIG. 5, in step 206, the determining unit 106 determines whether a sound event occurs in the sound box according to the ratio of the number of sampling points determined to be candidate sampling points in the sound box to the total number of sampling points of the sound box. To produce a judgment result. The more sampling points that are judged to be candidate sampling points in a sound box, the higher the probability of occurrence of a sound event. When the ratio of the number of sampling points determined to be the candidate sampling point in the sound box to the total number of sampling points of the sound box is greater than a third threshold value TH3, the determining unit 106 determines that a sound event occurs in the sound box, and According to the results of the judgment. In other words, through the above steps, the present invention can accurately determine whether a sound event occurs by performing correlation processing on the sampled value of the sampling point of the audio digital signal in the time domain.

For example, referring to FIG. 5, if the threshold TH3 is 0.6, for the sampling points SP1 to SP8 in the frame F, in step 206, the sampling points SP1, SP2, SP3, SP5, SP6, SP7, SP8 are Determined and classified as candidate sampling points. In this case, the number of sampling points determined to be candidate sampling points in the sound box F is seven, and the total number of sampling points in the sound box F is eight, and thus, the number of sampling points that are determined as candidate sampling points is determined. The ratio of (7) to the total number of sampling points (8) of the frame F is 7/8=0.875, that is, the ratio of the number of sampling points judged to be candidate sampling points in the sound box F to the total number of sampling points is greater than the threshold TH3 (0.875>0.6). Therefore, the judging unit 106 judges that a sound event has occurred in the sound box F.

In short, through the arithmetic processing in step 206, it can be determined which sampling points in a certain sound box of the audio signal of the audio digital signal can be regarded as candidate sampling points, and according to the sampling determined as candidate sampling points. The proportion of the number of points is used to evaluate the sound events in the sound box.

In step 208, the control unit 108 outputs a control signal according to the determination result to control the related application function. That is to say, through steps 202 to 206, the sound event detecting device 10 continuously performs a sound event detecting process on the input audio signal. When the judgment result indicates that a sound event occurs in a certain sound box of the audio digital signal, the control unit 108 outputs a control signal to control the related application function. For example, the sound event detecting device 10 is applied to a baby crying detection system (or an anti-theft system). If there is a baby crying (or other sound) in the environment, the judgment output by the determining unit 106 is used. The result indicates that a sound event occurs in the sound box of the audio digital signal, and the control unit 108 outputs a control signal to trigger the baby crying detection system (or the anti-theft system) to transmit the audio digital signal to the remote communication device. Alternatively, the control unit 108 outputs a control signal to trigger the baby crying detection system (or the anti-theft system) to transmit the short message to the remote communication device, so that the user holding the communication device can know The occurrence of a sound event.

It is to be noted that the sound event detecting device 10 of the first embodiment is merely illustrative of one of the present invention, and those skilled in the art can modify or change the present invention without departing from the scope of the present invention. For example, the sampling rate may be 8 kHz, 11.025 kHz, 22.05 kHz, 16 kHz, 37.8 kHz, 44.1 kHz, 48 kHz, but not limited thereto. The sampling point length of the frame can be set according to requirements. The detection sensitivity of the sound event detecting device 10 can also be adjusted by changing the size of the threshold values TH1, TH2, and TH3. In addition, in step 204, adjacent audio frames in the audio digital signal overlap each other, As a result, the association of the data between the frames will be preserved.

In summary, the sound event detecting apparatus 10 of the present invention can perform correlation processing on the sampling value of the sampling point of the audio digital signal in the time domain to determine whether a sound event occurs, and output the relevant control signal accordingly. Controls the execution of related application functions. In this way, the present invention can determine the sound event based on the time domain operation without complicated FFT operation, and can effectively reduce the system calculation amount and is suitable for the instant application.

The above are only the preferred embodiments of the present invention, and all changes and modifications made to the scope of the present invention should be within the scope of the present invention.

20‧‧‧ Process

202, 204, 206, 208‧‧ steps

Claims (15)

A method for detecting a sound event based on a time domain operation includes: converting an audio analog signal into an audio digital signal according to a sampling rate, and defining a plurality of sampling points of the audio digital signal as a sound box, each The sampling value of the sampling point is in decibels (dB); for each sampling point in the sound box, an average difference of a plurality of different sampling groups corresponding to one sampling point is calculated to generate a plurality of average difference values; Determining, according to the plurality of average difference values, whether the sampling point is a candidate sampling point of a sound event, and according to the ratio of the number of sampling points determined as candidate sampling points in the sound box to the total number of sampling points of the sound box Determining whether the sound event occurs in the sound box to generate a determination result; and outputting a control signal according to the determination result to control an application function. The sound event detecting method of claim 1, wherein calculating the average difference of the plurality of different sampling groups corresponding to the sampling point to generate the plurality of average difference values comprises: according to the plurality of reference sampling rates And calculating an average difference of the plurality of different sampling groups corresponding to the sampling point to generate the plurality of average difference values. The method for detecting a sound event according to claim 2, wherein the step of calculating the average difference of the plurality of different sampling groups corresponding to the sampling point according to the plurality of reference sampling rates to generate the plurality of average difference values The method includes: for a first sampling group of the plurality of different sampling groups, calculating, according to one of the plurality of reference sampling rates, a first reference sampling rate, respectively, the sampling point is adjacent to a first sampling point interval The absolute difference between the sampled values between the sampling points and the absolute difference between the sampled points at least one of the sampling points after the sampling point and the adjacent sampling points spaced apart from the first sampling point And generating a plurality of first absolute differences, wherein a length of the first sampling point interval is equal to a ratio of the sampling rate of the audio digital signal to the first reference sampling rate; and calculating the plurality of first absolute differences An average value as a first average difference value; for the second sampling group of the plurality of different sampling groups, the sampling point is calculated according to the second reference sampling rate of the plurality of reference sampling rates Absolute difference between sample values between adjacent sample points separated by a second sample point interval and absolute difference between sample points between at least one sample point located after the sample point and adjacent sample points spaced apart from the second sample point interval a plurality of second absolute difference values, wherein a length of the second sampling point interval is equal to a ratio of the sampling rate of the audio digital signal to the second reference sampling rate; and calculating the plurality of second absolute differences The average value is taken as a second average difference value. The sound event detecting method of claim 2, wherein the plurality of reference sampling rates are smaller than the sampling rate of the audio digital signal. The sound event detecting method of claim 1, wherein the step of determining whether the sampling point is a candidate sampling point of the sound event according to the plurality of average difference values comprises: comparing the plurality of average difference values with a first a threshold value, and counting the number of the plurality of average difference values that is greater than the first threshold value; and the number of the plurality of average difference values that is greater than the first threshold value is greater than a second threshold When the value is determined, the sampling point is judged to be the candidate sampling point. The method for detecting a sound event according to claim 1, wherein the sound box determines whether the number of sampling points of the candidate sampling point is the ratio of the total number of sampling points of the sound box, and whether the sound box occurs in the sound box. The step of generating the judgment result includes: determining the sound box when the ratio of the number of sampling points determined to be candidate sampling points in the sound box to the total number of sampling points of the sound box is greater than a third threshold value The sound event occurs in the middle to generate the critical result. The sound event detecting method of claim 1, wherein the step of outputting a control signal to control an application function according to the determining result comprises: outputting the sound event when the determining result indicates that the sound event occurs in the sound box Control signals to control the application's functionality. The method for detecting a sound event according to claim 1, wherein adjacent sound frames in the audio digital signal overlap each other. A sound event detecting device based on time domain operation includes: a type of comparison digital conversion unit for converting an audio analog signal into an audio digital signal according to a sampling rate, and sampling the plurality of audio digital signals The point is defined as a sound box, and the sampling value of each sampling point is in decibels (dB); a calculating unit is used to calculate a plurality of different ones corresponding to one sampling point for each sampling point in the sound box. The average difference of the sampling groups is generated to generate a plurality of average difference values; a determining unit is configured to determine, according to the plurality of average difference values, whether the sampling point is a candidate sampling point of a sound event, and according to the sound box The ratio of the number of sampling points determined as candidate sampling points to the total number of sampling points of the sound box, determining whether the sound event occurs in the sound box to generate a determination result; and a control unit for determining the result according to the determination , output a control signal to control an application function. The sound event detecting device of claim 9, wherein the calculating unit calculates an average difference of the plurality of different sampling groups corresponding to the sampling point according to the plurality of reference sampling rates to generate the plurality of average difference values, The determining unit compares the plurality of average difference values with a first threshold value and counts the number of the plurality of average difference values that is greater than the first threshold value, and the judgment sheet And determining, according to the plurality of average difference values, that the number of the first threshold is greater than a second threshold, determining each of the sampling points as the candidate sampling point. The sound event detecting device of claim 10, wherein the calculating unit calculates, according to one of the plurality of different sampling groups, the first sampling group, according to the first reference sampling rate of the plurality of reference sampling rates An absolute difference between the sample point and a sample value between adjacent sample points spaced apart from a first sample point and at least one sample point located after the sample point and a sample between adjacent sample points spaced apart from the first sample point An absolute difference of values to generate a plurality of first absolute difference values, the calculation unit calculates an average of the plurality of first absolute difference values as a first average difference value, the calculation unit is for the plurality of different sampling groups a second sampling group, and calculating, according to one of the plurality of reference sampling rates, a second reference sampling rate, an absolute difference between the sampling points of the sampling points and adjacent sampling points spaced apart from the second sampling point, and at least one An absolute difference between sample points located after the sampling point and adjacent sample points spaced apart from the second sampling point to generate a plurality of second absolute differences to The calculating unit calculates an average of the plurality of second absolute differences as a second average difference value, wherein the length of the first sampling point interval is equal to the sampling rate of the audio digital signal and the first reference sampling rate The ratio of the second sampling point interval is equal to the ratio of the sampling rate of the audio digital signal to the second reference sampling rate. The sound event detecting device of claim 10, wherein the plurality of reference sampling rates are less than the sampling rate of the audio digital signals. The sound event detecting device of claim 9, wherein the determining unit determines that the number of sampling points of the candidate sampling points in the sound box is greater than a third threshold of the total number of sampling points of the sound box. When it is determined, the sound event occurs in the sound box. The sound event detecting device of claim 9, wherein the control unit outputs the control signal to control the application function when the determining result indicates that the sound event occurs in the sound box. The sound event detecting device of claim 9, wherein the adjacent sound frames in the audio digital signal overlap each other.