TW202042217A - Method for detecting baby cry - Google Patents

Abstract

The disclosure provides a method for detecting baby cry, comprising: extracting at least a feature set of a sound signal to be tested, and further comprising: performing framing on the sound signal to be tested to obtain at least a sound frame; obtaining a fundamental frequency of each sound frame; performing a DC removal on each sound frame; calculate a signal strength and a signal zero-crossing rate of each DC-removed sound frame, the fundamental frequency, signal strength and signal zero-crossing rate constituting a feature set; and determining whether the sound signal to be tested including a baby crying sound to obtain a detection result, further comprising: detecting a sound frame attribute of each feature set, and according to the sound frame attribute, determining whether the sound signal to be tested including a baby crying sound to obtain the detection result.

Description

Baby crying detection method

The invention relates to a method for detecting infant crying.

Due to the increasing popularity of smart electronic products, the common baby monitors with functions similar to walkie-talkies in the past are no longer adequate. More and more functions, such as automatic detection of baby crying, are also favored by many parents. In the current known baby crying detection methods, the common method is to judge whether the received sound source contains a baby by taking the zero-crossing rate in the sound as a characteristic value, and then matching the threshold value and specific judgment rules. Crying. However, since the zero-crossing rate in the sound is greatly affected by the interference of non-baby crying, it is easy to affect the accuracy of judgment. Another common method is to use cepstral coefficients as feature values, and then use machine learning or pattern recognition algorithms for discrimination; the disadvantage of this method is that it requires a large number of labeled samples to be collected for training, and The amount of computation that needs to be performed is high.

An embodiment of the present invention discloses a baby crying detection method, which includes the following steps: a step of extracting characteristic values, which is a step of inputting a sound signal to be measured in time sequence to extract at least one characteristic value group of the sound signal to be measured; and, The value judgment step is to input the characteristic value group in time sequence and determine whether the sound signal to be measured includes a baby cry according to the characteristic value group to obtain a detection result; wherein, the characteristic value extraction step further includes: The sound signal to be tested is sound-framed to generate at least one sound-framed sound signal; the signal fundamental frequency of each sound-framed sound signal is calculated to obtain a sound-frame fundamental frequency; the sound-framed sound signal is DC removed Calculation to generate a direct current to remove the framed audio signal; calculate the signal strength of the DC removed framed audio signal and the signal zero crossing rate to obtain a frame strength and a frame zero crossing rate respectively. The sound frame intensity, the sound frame zero-crossing rate, and the sound frame fundamental frequency constitute a feature value group; and the feature value judgment step further includes: detecting a sound frame attribute of the feature value group, and then determining the sound frame attribute Whether the sound signal to be tested contains baby crying to obtain the detection result.

In a preferred embodiment, calculating the signal strength of the sound signal under test further includes the following steps: performing time-domain energy calculation on the DC-removed sound framed sound signal to generate a sound frame energy; Energy smoothing calculation can get the sound frame intensity.

In a preferred embodiment, calculating the signal zero-crossing rate of the sound signal under test further includes the following steps: calculating the zero-crossing times of the DC-removed sound framed sound signal to generate a sound frame zero-crossing times; The zero-crossing frequency of the sound frame is smoothed to obtain the zero-crossing rate of the sound frame.

In a preferred embodiment, calculating the signal base frequency of the sound signal to be measured further includes the following steps: generating an energy spectrum based on the sound framed sound signal; generating an estimated base frequency based on the energy spectrum; The basic frequency of the sound frame can be obtained by smoothing the estimated value of the fundamental frequency.

In a preferred embodiment, the step of generating an energy spectrum includes: windowing the sound framed sound signal to generate a windowed sound framed sound signal; and performing time-frequency on the windowed sound framed sound signal Converting to generate a frequency spectrum; passing the frequency spectrum through the spectrum energy calculation to generate the energy spectrum.

In a preferred embodiment, the step of generating a fundamental frequency estimation value further includes: generating a regional peak group according to the energy spectrum, first selecting a frequency point as a candidate peak on the energy spectrum, and then Use the candidate peak as the reference point to compare the regional energy. If the candidate peak is judged to be the winner in the regional energy comparison, the candidate peak is marked as a regional peak, otherwise, it is marked as other, until the energy spectrum Until all the frequency points are marked, the set of all peaks in the region is the peak group of the region. The energy comparison of the region means that if the energy of the candidate peak is greater than all the peaks in the frequency range centered on the candidate peak For the energy of other frequency points, the candidate peak is judged as the winner; then, the peak interval is calculated, including, if the number of regional peaks in the regional peak group is higher than the threshold of a regional peak number, then the adjacent peaks in the regional peak group are calculated In order to generate a peak interval group; otherwise, determine that the fundamental frequency estimation result is unstable; and, calculate the fundamental frequency, calculate the fundamental frequency according to the peak interval group, and generate a fundamental frequency estimation result, including: The abnormal interval is to exclude the abnormal extreme value in the peak interval group to obtain a normal peak interval group; to detect the peak interval variability, to calculate the difference of the extreme value in the normal peak interval group, if the difference is less than a difference threshold, proceed Peak average interval calculation, otherwise it is judged that the fundamental frequency estimation result is unstable; to calculate the peak average interval, calculate the average value of the normal peak interval group to obtain a peak average interval; search for the fundamental frequency peak, based on the energy Search for the fundamental frequency peak at the average interval of the peaks on the spectrum; and fundamental frequency weighted average, which is a weighted average of the fundamental frequency peak and the higher energy of the upper and lower frequency points to obtain the fundamental frequency estimate.

In a preferred embodiment, the step of detecting a sound frame attribute of the feature value group further includes: performing strong sound frame detection on the sound frame, and if the strength of the sound frame is greater than a strength threshold, then determining that the sound frame is strong The sound frame attribute; otherwise, it is determined that the sound frame has the weak sound frame attribute; and, if the sound frame has the strong sound frame attribute, then the crying sound frame detection is performed on the sound frame, and if the zero crossing rate of the sound frame falls within one If it is between the upper and lower bounds of zero rate, or the fundamental frequency of the sound frame falls between the upper and lower bounds of a fundamental frequency, it is determined that the sound frame has the crying sound frame attribute.

In a preferred embodiment, the step of judging whether the sound frame attribute includes baby crying sound further includes: calculating the number of strong sound frame and crying sound frame; if the attributes of two adjacent sound frames are first strong and then weak , The sound length detection is performed; if the sound length detection is passed, the crying degree detection is performed; otherwise, the detection result is determined to be non-crying, and the count of each attribute sound frame is reset to zero.

In a preferred embodiment, the degree of crying detection means that if the crying sound frame count exceeds a crying sound frame count threshold, the detection result is determined to be a crying sound, and the count of each attribute sound frame is reset to zero; If the ratio of the frame count to the strong sound frame count is higher than the crying ratio threshold, the sound signal to be tested is determined to be a cry-like sound, and if the number of occurrences of the cry-like sound exceeds the threshold of a cry-like count, the detection result is determined to be crying The count of each attribute sound frame is reset to zero; if the interval between two adjacent crying sounds is greater than the threshold of the interval of a kind of crying, the crying count is reset to zero.

The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand the other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied by other different specific examples, and various details in the specification of the present invention can also be modified and changed based on different viewpoints and applications without departing from the spirit of the present invention.

Among them, the structure, ratio, size, etc. shown in the drawings in this specification are only used to match the content disclosed in the specification for the understanding and reading of those familiar with the technology, and are not intended to limit the implementation of the present invention. Conditions, so it does not have any technical significance. Any structural modification, ratio change or size adjustment, without affecting the effects and objectives that can be achieved by the invention, should fall within the disclosure of the invention The technical content must be covered.

As shown in FIG. 1, an embodiment of the present invention discloses a baby crying detection method, which includes: Step 100: a step of extracting characteristic values, in which a sound signal to be measured is input in time sequence to extract at least one of the sound signals to be measured Feature value group; and, step 200: feature value judging step, which is to input the feature value group in time sequence and determine whether the sound signal to be tested includes a baby cry according to the feature value group to obtain a detection result; wherein , The step of extracting the characteristic value further includes: step 110 to frame the sound signal to be tested to generate at least one sound framed sound signal; step 120 calculate the signal fundamental frequency of each sound framed sound signal to obtain a sound frame Fundamental frequency; step 130 performs a DC removal operation on the audio framed audio signal to generate a direct current removed audio framed audio signal; step 140 calculates the signal strength of the DC audio framed audio signal and the signal zero crossing rate , To obtain a sound frame intensity and a sound frame zero-crossing rate respectively, the sound frame strength, the sound frame zero-crossing rate, and the sound frame fundamental frequency constitute a feature value group; and the feature value judging step further includes: Step 210: Detect a sound frame attribute of the feature value group, and Step 220: Determine whether the sound signal to be tested includes baby crying according to the sound frame attribute, so as to obtain the detection result.

The so-called frame is to gather N sampling points into an observation unit, called frame. Usually, the value of N is 256 or 512, and the time covered is about 20-30 ms. In order to avoid excessive changes between two adjacent sound frames, there will be an overlap area between two adjacent sound frames. This overlap area contains M sampling points. Usually the value of M is about half or 1/3 of N. Among them, when the sound frame is performed (step 110), the generated sound frames will partially overlap. It is worth noting that the above-mentioned N value, M value, the length of time covered, and whether the sound frames overlap or not are all conventionally used to illustrate the embodiments of the present invention, but are not limited to these in practical applications.

It is worth noting that the main purpose of step 130 is to extract a set of feature values with baby cry recognition for the sound signal to be tested; in the setting of the present invention, the feature value set with baby cry recognition is at least It includes a signal strength, a signal zero-crossing rate, and a signal base frequency. Among them, the sound frame intensity, the sound frame zero-crossing rate, and the sound frame fundamental frequency can be calculated separately, and the calculation sequence is not prioritized.

2 is a schematic diagram of the flow chart of calculating the signal intensity of the sound signal to be measured in a method for detecting infant crying of the present invention; FIG. 3 is a flow chart of calculating the signal zero-crossing rate of the sound signal to be measured in a method for detecting infant crying of the present invention Figure 4 is a schematic flow chart of calculating the signal base frequency of the sound signal to be tested in a method for detecting baby crying according to the present invention.

Specifically, as shown in FIG. 2, calculating the signal intensity of the sound signal to be measured further includes: step 1301, performing time-domain energy calculation on the DC removed sound framed sound signal to generate a sound frame energy; step 1302 Perform energy smoothing calculation on the sound frame energy to obtain the sound frame intensity. Among them, in a preferred embodiment, the method used in the time-domain energy calculation is to take the average of the absolute values of all sampling points in the sound frame, but it is not limited to this. Similarly, in a preferred embodiment, the method used when performing the energy smoothing calculation is: weighted average the energy of the current sound frame and the energy of the previous sound frame, but it is not limited to this.

On the other hand, as shown in FIG. 3, calculating the signal zero-crossing rate of the sound signal under test further includes: step 1303, calculating the number of zero-crossings of the DC-removed sound framed sound signal to generate a zero-crossing frequency of a sound frame Step 1304: Perform a zero-crossing smoothing operation on the zero-crossing frequency of the sound frame to obtain the zero-crossing rate of the sound frame. Among them, in a preferred embodiment, the method used when performing the zero-crossing times smoothing operation is: weighted average the current frame zero-crossing times and the previous frame zero-crossing times, but it is not limited to this.

Similarly, as shown in FIG. 4, calculating the signal base frequency of the sound signal under test further includes: step 1305, generating an energy spectrum based on the sound framed sound signal; step 1306, generating a base frequency estimate based on the energy spectrum Value; Step 1307: Perform the basic frequency estimated value smoothing operation on the estimated fundamental frequency to obtain the fundamental frequency of the sound frame.

FIG. 5 is a schematic diagram of the process of generating energy spectrum in a method for detecting infant crying of the present invention; FIG. 6 is a schematic diagram of the process of generating an estimated fundamental frequency in a method of detecting infant crying of the present invention.

Based on the foregoing, as shown in Figure 5, the step of generating an energy spectrum (step 1305) includes: step 1305a, windowing the sound framed sound signal to generate a windowed sound framed sound signal; step 1305b, Perform time-frequency conversion on the windowed sound framed sound signal to generate a frequency spectrum; step 1305c, calculate the energy spectrum through the frequency spectrum to generate the energy spectrum. Among them, the so-called windowing refers to multiplying each sound frame by a window function, for example, a Hamming window, to increase the continuity between the left and right ends of the sound frame, but it is not limited to this. On the other hand, in a preferred embodiment, the conversion method used in the time-frequency conversion is fast Fourier transform, but it is not limited to this. Similarly, in a preferred embodiment, the calculation function used in the calculation of spectral energy is an absolute value function, but it is not limited to this.

As shown in FIG. 6, the step of generating a fundamental frequency estimate further includes: step 1306a, generating a regional peak group based on the energy spectrum; step 1306b, calculating the peak interval; step 1306c, calculating the fundamental frequency. The details are as follows:

Wherein, in step 1306a, a regional peak group is generated based on the energy spectrum, and a frequency point is first selected as a candidate peak on the energy spectrum; then the candidate peak is used as a reference point to compare regional energy. If the candidate peak is judged as the winner in the regional energy comparison, the candidate peak is marked as a regional peak, otherwise, it is marked as other, until all the frequency points on the energy spectrum are marked. The set of all peaks in the region is the peak group of the region. The energy comparison of the region means that if the energy of the candidate peak is greater than the energy of all other frequency points in a frequency range centered on the candidate peak, then the candidate The peak value is judged as the winner.

Step 1306b. Calculate the peak interval, which further includes: if the number of regional peaks in the regional peak group is higher than the threshold of the number of regional peaks, then calculate the interval between adjacent peaks in the regional peak group to generate a peak interval group; otherwise, determine The fundamental frequency estimation result is unstable.

Step 1306c: Calculate the fundamental frequency, calculate the fundamental frequency based on the peak interval group, and generate a fundamental frequency estimation result, including: excluding abnormal intervals, excluding abnormal extreme values in the peak interval group to obtain a normal peak interval group ; Detection of peak interval variability is to calculate the difference between the extreme values in the normal peak interval group. If the difference is less than a difference threshold, the peak average interval calculation is performed; otherwise, the fundamental frequency estimation result is judged to be unstable; the peak average is calculated The interval is to calculate the average value of the normal peak interval group to obtain a peak average interval; to search for the fundamental frequency peak, to search for the fundamental frequency peak at the peak average interval on the energy spectrum; and the fundamental frequency weighted average to obtain the The base frequency peak value and the higher energy of the upper and lower frequency points are weighted and averaged to obtain the estimated value of the base frequency. Among them, in a preferred embodiment, the method used in the smoothing operation of the fundamental frequency estimation value is: if the current fundamental frequency estimation value is stable, the sound frame fundamental frequency is the current fundamental frequency estimation Value; otherwise, the fundamental frequency of the sound frame is the fundamental frequency of the previous sound frame.

FIG. 7 is a schematic diagram of the process of detecting the frame attributes of the feature value group in a method for detecting infant crying according to the present invention. As shown in FIG. 7, the step of detecting a sound frame attribute of the feature value group further includes: step 2101, performing strong sound frame detection on the sound frame, and if the sound frame intensity is greater than an intensity threshold, it is determined that the sound frame has If the tone frame has the attribute of strong tone frame, if the tone frame has the attribute of strong tone frame, then perform crying sound frame detection on the tone frame, and if the tone frame has zero crossing rate If it falls between the upper and lower bounds of a zero-crossing rate, or the fundamental frequency of the sound frame falls between the upper and lower bounds of a fundamental frequency, it is determined that the sound frame has a crying sound frame attribute.

Fig. 8 is a schematic diagram of the process of judging whether the sound signal to be measured includes baby crying according to the properties of the sound frame in a method for detecting infant crying; Schematic diagram of the process. Wherein, as shown in FIG. 8, the step of judging whether the sound signal to be tested contains baby crying for the sound frame attribute further includes: step 2201, calculating the number of strong sound boxes and crying sound boxes; step 2202, if the two are adjacent The attributes of the sound frame are strong first and then weak, then the sound length detection is performed; step 2203, if the sound length detection is passed, then the crying degree detection is performed; otherwise, the detection result is judged to be non-crying, and each attribute sound frame is counted Zero. Wherein, the sound length detection in step 2202 means that if the strong tone frame count is lower than a strong tone frame count threshold, it is deemed to pass the detection.

As shown in Figure 9, the crying level detection in step 2203 further includes: step 2203a. If the crying sound frame count exceeds a crying sound frame count threshold, then the detection result is determined to be crying, and the attribute sound frame count is reset to zero Step 2203b, if the ratio of the count of the crying sound frame to the count of the strong sound frame is higher than a crying sound ratio threshold, determine that the sound signal to be measured is a cry-like sound; Step 2203c, if the number of occurrences of the crying sound exceeds a kind of crying If the sound count threshold is reached, it is determined that the detection result is a cry, and the count of each attribute sound frame is reset to zero; in step 2203d, if the interval between two adjacent crying sounds is greater than the first kind of crying interval threshold, the crying sound count is reset to zero.

Although the implementation has been described with reference to many illustrative embodiments of the present application, it should be understood that those skilled in the art can think of many other changes and embodiments, and these changes and embodiments will fall within the spirit and scope of the principles of the present disclosure. Inside. In particular, within the scope of the present disclosure, the drawings and the attached patent application, various changes and modifications can be made to the components and/or arrangements of the subject combination arrangement. In addition to changes and modifications to the components and/or settings, alternative uses will be obvious to those skilled in the art.

100: Extract characteristic value 110: Frame the sound signal to be tested 120: Calculate the fundamental frequency of each framed sound signal 130: Perform DC removal operation on the framed sound signal 140: Calculate the frame strength and the zero-crossing rate of the sound frame of the DC removed framed sound signal. The frame strength, the zero-crossing rate of the sound frame, and the fundamental frequency of the sound frame constitute a characteristic value group 200: eigenvalue judgment 210: Detect the frame attributes of the feature value group 220: Determine whether the sound frame contains baby crying 1301: Perform time domain energy calculation 1302: Perform energy smoothing operations 1303: Calculate the number of zero crossings 1304: Perform zero-crossing smoothing operation 1305: Generate energy spectrum 1306: Generate estimated fundamental frequency 1307: Perform a smoothing operation on the estimated fundamental frequency 1305a: Perform sound frame and window 1305b: Perform time-frequency conversion 1305c: Perform spectrum energy calculation 1306a: Generate a regional peak group based on the energy spectrum 1306b: Calculate the peak interval 1306c: Calculate the fundamental frequency 2101: Perform strong sound frame detection to determine whether it has strong sound frame attributes 2102: If there is a strong sound frame attribute, perform crying sound frame detection 2201: Calculate the number of strong sound boxes and crying sound boxes 2202: If the attributes of two adjacent sound frames are strong first and then weak, the sound length detection is performed 2203: If the sound length detection is passed, the crying level detection is performed 2203a: If the crying sound frame count exceeds the one crying sound frame count threshold, the detection result is determined to be crying, and the sound frame count of each attribute is reset to zero 2203b: If the ratio of the crying sound frame count to the strong sound frame count is higher than a crying sound ratio threshold, the sound signal to be measured is determined to be a cry-like sound 2203c: If the number of occurrences of crying sounds exceeds the count threshold of crying sounds, the detection result is determined to be crying, and the count of each attribute sound frame is reset to zero 2203d: If the interval between two adjacent crying sounds is greater than the threshold of the first crying sound, the crying sound count will be reset to zero

Fig. 1 is a schematic flow diagram of a method for detecting infant crying according to the present invention; 2 is a schematic diagram of the process of calculating the signal intensity of the sound signal to be measured in a method for detecting infant crying according to the present invention; 3 is a flow chart of calculating the zero-crossing rate of a sound signal to be measured in a method for detecting baby crying according to the present invention; 4 is a schematic flow chart of calculating the signal base frequency of the sound signal to be measured in a method for detecting baby crying according to the present invention; FIG. 5 is a schematic diagram of the process of generating energy spectrum in a method for detecting infant crying according to the present invention; FIG. 6 is a schematic diagram of a flow chart of generating a fundamental frequency estimation value in a method for detecting infant crying according to the present invention; FIG. 7 is a schematic diagram of the process of detecting the sound frame attributes of the feature value group in a method for detecting baby cry of the present invention; FIG. 8 is a schematic diagram of the process of judging whether the sound signal to be tested contains a baby cry according to the sound frame attribute in a method for detecting baby cry of the present invention; FIG. 9 is a schematic diagram of the flow of detecting the degree of crying in a method for detecting infant crying according to the present invention.

100: Extract characteristic value

110: Frame the sound signal to be tested

120: Calculate the fundamental frequency of each framed sound signal

130: Perform DC removal operation on the framed sound signal

140: Calculate the frame strength and the zero-crossing rate of the sound frame of the DC removed framed sound signal. The frame strength, the zero-crossing rate of the sound frame, and the fundamental frequency of the sound frame constitute a characteristic value group

200: eigenvalue judgment

210: Detect the frame attributes of the feature value group

220: Determine whether the sound frame contains baby crying

Claims (18)

A method for detecting baby crying includes the following steps: The step of extracting characteristic values is to input a sound signal to be measured in time sequence to extract at least one characteristic value group of the sound signal to be measured; and, The feature value judging step is to input the feature value group in time sequence and determine whether the sound signal to be tested includes a baby crying according to the feature value group, so as to obtain a detection result; Wherein, the step of extracting characteristic values further includes: sound frame the sound signal to be tested to generate at least one sound frame sound signal; calculate the signal fundamental frequency of each sound frame sound signal to obtain a sound frame fundamental frequency; Perform a DC removal operation on the audio framed audio signal to generate a DC audio framed audio signal; calculate the signal strength of the DC audio framed audio signal and the signal zero crossing rate to obtain a audio frame respectively Intensity and the zero-crossing rate of a sound frame, the sound-frame intensity, the zero-crossing rate of the sound frame, and the fundamental frequency of the sound frame constitute a characteristic value group; The feature value judging step further includes: detecting a sound frame attribute of the feature value group, and then judging whether the sound signal to be tested includes baby crying according to the sound frame attribute, so as to obtain the detection result. For example, the baby crying detection method described in the first item of the scope of patent application, wherein when the sound frame is performed, the generated sound frames will partially overlap. For the infant cry detection method described in claim 1, wherein calculating the signal strength of the sound signal to be measured further includes the following steps: performing time-domain energy calculation on the DC-removed sound framed sound signal to generate A sound frame energy; the sound frame energy is subjected to an energy smoothing operation to obtain the sound frame intensity. For the infant cry detection method described in item 3 of the scope of patent application, the method used in the time domain energy calculation is to take the average of the absolute values of all sampling points in the sound frame. The method for detecting infant crying as described in item 3 of the scope of patent application, wherein the method used in the energy smoothing operation is: weighted average of the energy of the current sound frame and the energy of the previous sound frame. As described in the first item of the patent application, the method for detecting baby crying, wherein calculating the signal zero-crossing rate of the sound signal to be tested further includes the following steps: calculating the number of zero-crossings of the DC removed sound framed sound signal , Generate the zero-crossing frequency of a sound frame; perform the zero-crossing smoothing operation on the zero-crossing frequency of the sound frame to obtain the zero-crossing rate of the sound frame. For example, the baby crying detection method described in item 6 of the scope of patent application, wherein the method used for smoothing the number of zero crossings is: weighting the number of zero crossings of the current frame with the number of zero crossings of the previous frame average. According to the infant crying detection method described in claim 1, wherein, calculating the signal base frequency of the sound signal to be measured further includes the following steps: generating an energy spectrum according to the sound framed sound signal; according to the energy spectrum Generate a fundamental frequency estimated value; perform the fundamental frequency estimated value smoothing operation on the fundamental frequency estimated value to obtain the fundamental frequency of the sound frame. For example, the baby crying detection method described in claim 8, wherein the step of generating an energy spectrum includes: windowing the sound framed sound signal to generate a windowed sound framed sound signal; The windowed sound framed sound signal is time-frequency converted to generate a frequency spectrum; the frequency spectrum is calculated through spectrum energy to generate the energy spectrum. The method for detecting baby crying as described in item 8 of the scope of patent application, wherein the function used in the windowing function is the Hanning window. The method for detecting infant crying as described in item 8 of the scope of patent application, wherein the conversion method used in the time-frequency conversion is fast Fourier transform. The method for detecting baby crying as described in item 8 of the scope of patent application, wherein the calculation function used in the calculation of spectral energy is an absolute value function. According to the infant crying detection method described in claim 1, wherein the step of generating a fundamental frequency estimation value further includes: According to the energy spectrum, a regional peak group is generated. First, a frequency point is selected as a candidate peak on the energy spectrum, and then the candidate peak is used as a reference point to compare the regional energy. If the candidate peak is in the regional energy If it is judged as a winner in the comparison, the candidate peak is marked as a regional peak, otherwise, it is marked as other. Until all the frequency points on the energy spectrum are marked, the set of all peaks in the region is the region Peak group, where the regional energy comparison means that if the energy of the candidate peak is greater than the energy of all other frequency points in a frequency range centered on the candidate peak, the candidate peak is judged as the winner; Then, calculate the peak interval, including, if the number of regional peaks in the regional peak group is higher than the threshold of a regional peak number, then calculate the interval between adjacent peaks in the regional peak group to generate a peak interval group; otherwise, determine the fundamental frequency The estimated result is unstable; and, Calculate the fundamental frequency, calculate the fundamental frequency based on the peak interval group, and generate a fundamental frequency estimation result, including: excluding abnormal intervals, excluding abnormal extreme values in the peak interval group to obtain a normal peak interval group; detecting the peak interval The degree of variability is to calculate the difference between the extreme values in the normal peak interval group. If the difference is less than a difference threshold, the peak average interval calculation is performed. Otherwise, the fundamental frequency estimation result is judged to be unstable; to calculate the peak average interval, calculate The average value of the normal peak interval group to obtain a peak average interval; the search for the fundamental frequency peak is to search for the fundamental frequency peak at the peak average interval on the energy spectrum; and the fundamental frequency weighted average is to obtain the fundamental frequency peak and the peak average interval. The weighted average of the energy of the upper and lower frequency points is performed to obtain the estimated value of the fundamental frequency. For example, the baby crying detection method described in item 13 of the scope of patent application, wherein the method used in the smoothing calculation of the fundamental frequency estimation value is: if the current fundamental frequency estimation value is stable, the sound frame The fundamental frequency is the estimated value of the current fundamental frequency; otherwise, the fundamental frequency of the sound frame is the fundamental frequency of the previous sound frame. According to the infant crying detection method described in claim 1, wherein the step of detecting a frame attribute of the characteristic value group further includes: Perform strong frame detection on the sound frame, and if the strength of the sound frame is greater than an intensity threshold, it is determined that the sound frame has the attribute of strong sound frame; otherwise, it is determined that the sound frame has the attribute of weak sound frame; and, If the sound frame has the attribute of strong sound frame, then the sound frame is detected again. If the zero crossing rate of the sound frame falls between the upper and lower bounds of the zero crossing rate, or the fundamental frequency of the sound frame falls on the first base Between the upper and lower bounds of the frequency, it is determined that the sound frame has the crying sound frame attribute. According to the infant crying detection method described in the first item of the scope of patent application, the step of judging whether the sound frame attribute contains the infant crying further includes: Calculate the number of strong sound boxes and crying sound boxes; If the attributes of two adjacent sound frames are strong first and then weak, the sound length detection is performed; If the sound length is detected, the degree of crying is detected; otherwise, the detection result is determined to be non-crying, and the count of each attribute sound frame is returned to zero. For example, the baby crying detection method described in item 16 of the scope of patent application, wherein the sound length detection means that if the high-tone frame count is lower than a high-tone frame count threshold, the detection is deemed to be passed. Such as the infant crying detection method described in item 16 of the scope of patent application, wherein the crying level detection refers to: If the crying sound frame count exceeds a crying sound frame count threshold, it is determined that the detection result is a crying sound, and the count of each attribute sound frame is reset to zero; If the ratio of the count of the crying sound frame to the count of the strong sound frame is higher than a crying ratio threshold, it is determined that the sound signal to be measured is a cry-like sound; If the number of occurrences of similar crying sounds exceeds the threshold for counting the number of crying sounds, the detection result is determined to be crying, and the count of each attribute sound frame is reset to zero; If the interval between two adjacent crying sounds is greater than the threshold of the interval between crying sounds, the count of crying sounds will be reset to zero.

Images