Data collecting method and system

1. A method of determining a highlight section of a sound source, the method comprising:

obtaining, by a processor, a sound source and classification information of the sound source, wherein the classification information is in a form of a vector indicating a probability of corresponding to each classification;

learning, by the processor, a neural network by using the sound source and the classification information, the neural network comprising an input layer including a node corresponding to a feature value of each of a plurality of sections obtained by splitting the sound source into said plurality of sections according to a time axis, an output layer including a node corresponding to the classification information, a hidden layer defined between the input layer and the output layer, a first function between the input layer and the hidden layer, and a second function between the hidden layer and the output layer, wherein the first function comprises an attention model for calculating a weighted sum of the feature value of each section; and

determining, by the processor, the highlight section of the sound source, based on weight information of a feature value node of each section included in the first function, wherein the weight information of the feature value node of each section indicates a degree of contribution of each section to an estimation of the classification information of the sound source.

2. The method of claim 1 , wherein the hidden layer comprises a node corresponding to an integrated feature value of the sound source obtained from the feature value of each section, according to the first function.

3. The method of claim 1 , wherein the first function further comprises a 1-1st function calculating a similarity of an output value of the attention model and an output value of a recurrent neural network (RRN) model, wherein the hidden layer comprises a node of the similarity.

4. The method of claim 1 , further comprising, after the obtaining of the sound source and the classification information,

converting the sound source comprising sound data regarding the time axis to include energy data regarding the time axis,

wherein the plurality of sections are obtained by splitting the converted sound source according to the time axis.

5. The method of claim 4 , wherein the determining of the highlight section comprises determining the highlight section based on the weight information of the feature value node of each section and the energy data of each section.

6. The method of claim 1 , wherein the determining of the highlight section comprises determining an important section based on the weight information of the feature value node of each section and determining the highlight section among a plurality of sections of the sound source by referring to energy data within a section of a pre-set range before and after the important section.

7. The method of claim 6 , wherein the determining of the highlight section comprises determining the highlight section in response to a point of time when momentum of the energy data is greatest within the section of the pre-set range.

8. A non-transitory tangible computer readable recording medium storing a computer program for determining a highlight section of a sound source, the program when executed by a computer performing the method of claim 1 .

9. An apparatus for determining a highlight section of a sound source, the apparatus comprising:

a processor configured to include a plurality of functional units for performing a plurality predetermined functions, the functional units including,

a sound source obtainer configured to obtain a sound source and classification information of the sound source, wherein the classification information is in a form of a vector indicating a probability of corresponding to each classification;

a neural network processor configured to learn a neural network by using the sound source and the classification information, the neural network comprising an input layer including a node corresponding to a feature value of each of a plurality of sections obtained by splitting the sound source into said plurality of sections according to a time axis, an output layer including a node corresponding to the classification information, a hidden layer defined between the input layer and the output layer, a first function between the input layer and the hidden layer, and a second function between the hidden layer and the output layer, wherein the first function comprises an attention model for calculating a weighted sum of the feature value of each section; and

a highlight determiner configured to determine a highlight section of the sound source, based on weight information of a feature value node of each section included in the first function, wherein the weight information of the feature value node of each section indicates a degree of contribution of each section to an estimation of the classification information of the sound source.

10. The method of claim 1 , wherein the classification information includes at least one of a genre, a mood, a preferred age group, a theme and an atmosphere of the sound source.

11. The apparatus of claim 9 , wherein the classification information includes at least one of a genre, a mood, a preferred age group, a theme and an atmosphere of the sound source.