Method for detecting speech in audio data

1. A method for detecting a voice from audio data, performed by a computing device including at least one processor, the method comprising:

obtaining audio data;

generating image data based on a spectrum of the obtained audio data;

extracting a feature on the image data by utilizing a first neural network model; and

extracting a time-series feature of the image data by utilizing a second neural network model sequentially connected with the first neural network model;

determining whether an automated response system (ARS) voice is included in each of a plurality of sections of the audio data, based on the feature on the image data and the time-series feature of the image data;

eliminating a section including the ARS voice, among the plurality of sections of the audio data; and

generating input data or training data for analysis of spoken voice based on remaining sections excluding the eliminated section, among the plurality of sections of the audio data.

2. The method according to claim 1 , wherein the generating of the image data includes:

generating a mel-spectrogram based on the spectrum of the obtained audio data.

3. The method according to claim 2 , wherein the generating of the image data further includes:

generating a plurality of frames by dividing the generated mel-spectrogram in a predetermined time unit.

4. The method according to claim 3 ,

wherein the extracting the feature on the image data includes:

extracting a feature on the image of the mel-spectrogram by utilizing the first neural network model; and

wherein the extracting the time-series feature of the image data includes:

further extracting a time-series feature of the mel-spectrogram by utilizing the second neural network model.

5. The method according to claim 4 , wherein the first neural network model includes a convolutional neural network (CNN) model and the second neural network model includes a long short-term memory (LSTM) model.

6. The method according to claim 4 , wherein the mel-spectrogram is divided into the plurality of frames,

the first neural network model includes a plurality of first sub modules, and

the extracting of the feature on the image of the mel-spectrogram includes:

extracting the feature on the image for each of the plurality of frames by utilizing the plurality of first sub modules.

7. The method according to claim 2 , wherein the determining of whether an ARS voice is included in the audio data includes:

determining a first probability representing whether a spoken voice is included in the audio data and a second probability representing whether the ARS voice is included in the audio data;

determining whether the spoken voice is included in the audio data based on the first probability; and

determining whether the ARS voice is included in the audio data based on the second probability.

8. The method according to claim 7 , wherein the mel-spectrogram is divided into a plurality of frames,

the determining of whether the ARS voice is included in the audio data further includes:

determining the first probability and the second probability with respect to each of the plurality of frames;

determining whether the spoken voice is included in each of the plurality of frames based on the first probability for each of the plurality of frames; and

determining whether the ARS voice is included in each of the plurality of frames based on the second probability for each of the plurality of frames.

9. A computer program stored in a non-transitory computer readable storage medium, wherein when the computer program is executed by one or more processors to perform the following operations to detect a voice from audio data, the operations including:

obtaining audio data;

generating image data based on a spectrum of the obtained audio data;

extracting a feature on the image data by utilizing a first neural network model; and

extracting a time-series feature of the image data by utilizing a second neural network model sequentially connected with the first neural network model;

determining whether an automated response system (ARS) voice is included in each of a plurality of sections of the audio data, based on the feature on the image data and the time-series feature of the image data;

eliminating a section including the ARS voice, among the plurality of sections of the audio data; and

generating input data or training data for analysis of spoken voice based on remaining sections excluding the eliminated section, among the plurality of sections of the audio data.

10. A computing device, comprising:

a processor including at least one core;

a memory including executable program codes in the processor; and

a network unit for obtaining an audio file,

wherein the processor is configured to:

generate image data based on a spectrum of the obtained audio data;

extract time-series feature of the image data by utilizing a first neural network model; and

extract time-series feature of the image data by utilizing a second neural network model sequentially connected with the first neural network model;

determine whether an automated response system (ARS) voice is included in each of a plurality of sections of the audio data, based on the feature on the image data and the time-series feature of the image data;

eliminate a section including the ARS voice, among the plurality of sections of the audio data; and

generate input data or training data for analysis of spoken voice based on the remaining sections excluding the eliminated section, among the plurality of sections of the audio data.