Audio signal processing method and apparatus
Disclosed is an operation method of an audio signal processing device configured to process an audio signal including a first audio signal component and a second audio signal component. The operation method includes: receiving the audio signal; normalizing loudness of the audio signal, based on a pre-designated target loudness; acquiring the first audio signal component from the audio signal having the normalized loudness, by using a machine learning model; and de-normalizing loudness of the first audio signal component, based on the pre-designated target loudness.
1. An operation method of an audio signal processing device which operates in at least one process and is configured to process an audio signal comprising a first audio signal component and a second audio signal component, the method comprising:
receiving the audio signal;
normalizing loudness of the audio signal, based on a pre-designated target loudness;
acquiring the first audio signal component from the audio signal having the normalized loudness, by using a machine learning model; and
de-normalizing loudness of the first audio signal component, based on the pre-designated target loudness.
2. The method of claim 1 , wherein at least one of the first audio signal component and the second audio signal component is an audio signal component corresponding to a voice.
3. The method of claim 1 , wherein the normalizing of the loudness of the audio signal, based on the pre-designated target loudness, comprises normalizing loudness in units of contents included in the audio signal.
4. The method of claim 1 , wherein the machine learning model processes the audio signal having the normalized loudness in a frequency area.
5. The method of claim 1 , wherein the normalizing of the loudness of the audio signal, based on the pre-designated target loudness, comprises:
dividing the audio signal into multiple pre-designated time intervals, dividing loudness values in the multiple pre-designated time intervals into multiple levels, and acquiring loudness of the audio signal by using a loudness value distribution for each of the multiple levels; and
normalizing the loudness of the audio signal to target loudness.
6. The method of claim 1 , wherein the machine learning model comprises gate logic.
7. The method of claim 1 , wherein the acquiring of the first audio signal component from the audio signal having the normalized loudness, by using the machine learning model, comprises classifying a frequency bin-specific score acquired from the machine learning model, based on a pre-designated threshold value,
wherein the score indicates a degree of closeness to the first audio signal component.
8. A method for training a machine learning model which operates in at least one process and is configured to classify a first audio signal component from an audio signal comprising the first audio signal component and a second audio signal acquired from different sources, the method comprising:
receiving the audio signal;
normalizing loudness of the audio signal, based on pre-designated target loudness;
acquiring a first audio signal component from the audio signal having the normalized loudness, by using the machine learning model; and
restoring the loudness of the first audio signal component, based on the pre-designated target loudness.
9. The method of claim 8 , wherein at least one of the first audio signal component and the second audio signal component is an audio signal component corresponding to a voice.
10. The method of claim 8 , wherein the normalizing of the loudness of the audio signal, based on the pre-designated target loudness, comprises normalizing loudness in units of contents included in the audio signal.
11. The method of claim 8 , wherein the machine learning model processes the audio signal having the normalized loudness in a frequency area.
12. The method of claim 8 , wherein the normalizing of the loudness of the audio signal, based on the pre-designated target loudness, comprises:
dividing the audio signal into multiple pre-designated time intervals, dividing loudness values in the multiple pre-designated time intervals into multiple levels, and acquiring loudness of the audio signal by using a loudness value distribution for each of the multiple levels; and
normalizing the loudness of the audio signal to target loudness.
13. The method of claim 8 , wherein the machine learning model comprises gate logic.
14. The method of claim 8 , wherein the acquiring of the first audio signal component from the audio signal having the normalized loudness, by using the machine learning model, comprises classifying a frequency bin-specific score acquired from the machine learning model, based on a pre-designated threshold value,
wherein the score indicates a degree of closeness to the first audio signal component.
15. An audio signal processing device configured to process an audio signal comprising a first audio signal component and a second audio signal component, the device comprising at least one processor,
wherein the at least one processor:
receives the audio signal;
normalizes loudness of the audio signal, based on a pre-designated target loudness;
acquires the first audio signal component from the audio signal having the normalized loudness, by using a machine learning model; and
de-normalizes loudness of the first audio signal component, based on the pre-designated target loudness.