상세정보
학술대회
Development of a Psychoacoustic Loss Function for the Deep Neural Network (DNN)-Based Speech Coder
- 발행일
- 202109
- 출처
- International Speech Communication Association (INTERSPEECH) 2021, pp.1694-1698
- 협약과제
- 21ZH1200, 초실감 입체공간 미디어·콘텐츠 원천기술연구, 이태진
- 초록
- This paper presents a loss function to compensate for the perceptual loss of the deep neural network (DNN)-based speech coder. By utilizing the psychoacoustic model (PAM), we design a loss function to maximize the mask-to-noise ratio (MNR) in multi-resolution Mel-frequency scales. Also, a perceptual entropy (PE)-based weighting scheme is incorporated onto the MNR loss so that the DNN model focuses more on perceptually important Mel-frequency bands. The proposed loss function was tested on a CNN-based autoencoder implementing the softmax quantization and entropy-based bitrate control. Objective and subjective tests conducted with speech signals showed that the proposed loss function produced higher perceptual quality than the previous perceptual loss functions.
- KSP 제안 키워드
- Deep neural network(DNN), Mel-frequency, Multi-resolution, Perceptual Quality, Psychoacoustic Model, Speech Signals, Speech coder, Subjective test, entropy-based, frequency band, loss function