상세정보
학술대회
Spatial-Cue Based Audio Channel Extension Using Convolutional Neural Networks
- 발행일
- 201906
- 출처
- International Symposium on Broadband Multimedia Systems and Broadcasting (BMSB) 2019, pp.1-4
- 협약과제
- 18HS1700, 다중소스 데이터 지능형 분석기반 고수준 정보추출 원천기술 연구, 유장희
- 초록
- In this paper, we introduce an audio channel extension tool using a spatial audio cue predicted using convolutional neural networks. The channel extension tool is applied into a common stereo signal to produce a high-dimensional audio signal, such as a 5.1 layout. To extend the channels from a stereo signal, we predict the spatial cues from the stereo signal, and synthesize the multichannel signals by allocating the spectral components according to the direction of the predicted spatial cues. Our subjective evaluation shows that a synthesized multichannel signal guarantees a high quality when compared with the original stereo signal.
- KSP 제안 키워드
- Audio signal, Convolution neural network(CNN), High-dimensional, Multichannel signals, Spatial audio, Spatial cue, spectral components, subjective evaluation