BROWSE
Detail
Conference Paper
Normalizing Flow based Audio Coding
Cited - time in 
Share 
Share
- Authors
- Inseon Jang, Wootaek Lim, Seungkwon Beack, Taejin Lee
- Issue Date
- 2022-10
- Citation
- International Congress on Acoustics (ICA) 2022, pp.1-5
- Language
- English
- Type
- Conference Paper
- Abstract
- Recently, deep neural network-based audio compression has been widely studied. Some of them include attempts at generative models, but most of them are based on autoencoder networks and focus on improving the decoded sound quality through adversarial training of generative adversarial network (GAN). In this paper, we present an audio compression using a flow based-generative model. With the benefit of the normalizing flow factor-out structure, the proposed method improves the compression efficiency while maintaining the encoding quality. To verify the performance of the proposed scheme, the objective evaluation using the signal-to-distortion ratio (SDR) and perceptual evaluation of speech quality (PESQ) as metrics are performed and compared with the autoencoder-based approach. The result of objective assessment shows the outperformance of the proposed method.
- KSP Keywords
- Adversarial Training, Audio coding, Audio compression, Based Approach, Deep neural network(DNN), Distortion ratio, Objective Evaluation, Objective assessment, compression efficiency, flow factor, generative adversarial network