Headset and operating method thereof
Disclosed are a headset and an operating method thereof, which execute a mounted artificial intelligence (AI) algorithm and/or machine learning algorithm and communicate with different electronic apparatuses and external servers in a 5G communication environment. The headset includes a plurality of microphones, an AEC, a VAD, and a BSS, which process sound signals received from the microphones, resulting in improved convenience for a user.
1. A headset, comprising:
one or more speakers;
a first microphone;
a second microphone;
an acoustic echo canceller (AEC) which attenuates or eliminates an output sound signal of the speaker which is input through the second microphone;
a voice activity detector (VAD) which determines a voice section based on a ratio of power of a first sound signal input from the first microphone to a second sound signal output from the AEC;
a blind source separator (BSS) which applies a source separation algorithm to the first sound signal and the second sound signal in the determined voice section so as to extract one or more sound source signals; and
a sound source selector which selects, among the extracted sound source signals, a sound source signal having a dominant power in the first sound signal,
wherein the VAD determines a sound signal section of the first sound signal in which a voice presence probability exceeds a first threshold.
2. The headset according to claim 1 , wherein the AEC updates a coefficient value of a first adaptive filter in real time through a least mean square (LMS) algorithm-based machine learning so as to attenuate or eliminate the output sound signal of the speaker.
3. The headset according to claim 1 , wherein the VAD determines, as the voice section, a sound signal section of the determined sound signal section in which the ratio of power of the first sound signal with respect to the second sound signal exceeds a second threshold.
4. The headset according to claim 3 ; wherein the BSS sets a cost function such that respective sound signals input from different sound sources become independent signal components, and updates a coefficient value of a second adaptive filter through machine learning in real time based on the set cost function, so as to extract one or more sound source signals in the determined voice section.
5. The headset according to claim 1 , wherein the first microphone is disposed to be closer to an utterance point of an utterer wearing the headset than the second microphone, and the second microphone is disposed to be closer to the speaker outputting a sound signal than the first microphone.
6. The headset according to claim 1 , wherein the second microphone is disposed in a predetermined area of a band which connects the speakers.
7. The headset according to claim 1 , further comprising a post-processor which eliminates residual noise included in the sound source signal extracted by the BSS, and equalizes the sound source signal.
8. An operating method of a headset, the method comprising:
obtaining a first sound signal through a first microphone;
obtaining a second sound signal obtained by attenuating or eliminating an output sound signal of a speaker from a sound signal input through a second microphone;
determining a voice section based on a ratio of power of the first sound signal to the second sound signal;
extracting one or more sound source signals by applying a source separation algorithm to the first sound signal and the second sound signal in the determined voice section; and
selecting, among the extracted sound source signals, a sound source signal having a dominant power in the first sound signal,
wherein the determining the voice section includes determining a sound signal section of the first sound signal in which a voice presence probability exceeds a first threshold.
9. The operating method according to claim 8 , wherein the obtaining the second sound signal includes updating a coefficient value of a first adaptive filter in real time through a least mean square (LMS) algorithm-based machine learning so as to attenuate or eliminate the output sound signal of the speaker.
10. The operating method according to claim 8 , wherein the determining the voice section includes determining, as the voice section, a sound signal section of the determined sound signal section in which the ratio of power of the first sound signal with respect to the second sound signal exceeds a second threshold.
11. The operating method according to claim 10 , wherein the extracting the sound source signal includes setting a cost function such that respective sound signals input from different sound sources become independent signal components, and updating a coefficient value of a second adaptive filter through machine learning in real time based on the set cost function, so as to extract one or more sound source signals in the determined voice section.
12. The operating method according to claim 11 , further comprising eliminating a residual noise included in the extracted sound source signal and equalizing the sound source signal.