Stacked U-Net with High-level Feature Transfer for Parameter Efficient Speech Enhancement

International Conference
2021-09-01 14:48
Authors : Jinyoung Lee and Hong-Goo Kang

Year : 2021

Publisher / Conference : APSIPA ASC

Research area : Speech Signal Processing, Speech Enhancement

In this paper, we present a stacked U-Net structure-based speech enhancement algorithm with parameter reduction and real-time processing. To significantly reduce the number of network parameters, we propose a stacked structure in which several shallow U-Nets with fewer convolutional layer channels are cascaded. However, simply stacking the small-scale U-Nets cannot sufficiently compensate for the performance loss caused by the lack of parameters. To overcome this problem, we propose a high-level feature transfer method that passes all the multi-channel output features, which are obtained before passing through the intermediate output layer, to the next stage.Furthermore, our proposed model can process analysis frames with short lengths because its down-sampling and up-sampling blocks are much smaller than the conventional Wave U-Net method; theses smaller layers make our proposed model suitable for low-delay online processing. Experiments show that our proposed method outperforms the conventional Wave U-Net method on almost all objective measures and requires only 7.21%of the network parameters when compared to the conventional method. In addition, our model can be successfully implemented in real time on both GPU and CPU environments.
전체 356
346 International Conference WooSeok Ko, Seyun Um, Zhenyu Piao, Hong-goo Kang "Consideration of Varying Training Lengths for Short-Duration Speaker Verification" in APSIPA ASC, 2023
345 International Journal Hyungchan Yoon, Changhwan Kim, Seyun Um, Hyun-Wook Yoon, Hong-Goo Kang "SC-CNN: Effective Speaker Conditioning Method for Zero-Shot Multi-Speaker Text-to-Speech Systems" in IEEE Signal Processing Letters, vol.30, pp.593-597, 2023
344 International Conference Miseul Kim, Zhenyu Piao, Jihyun Lee, Hong-Goo Kang "BrainTalker: Low-Resource Brain-to-Speech Synthesis with Transfer Learning using Wav2Vec 2.0" in The IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI), 2023
343 International Conference Seyun Um, Jihyun Kim, Jihyun Lee, Hong-Goo Kang "Facetron: A Multi-speaker Face-to-Speech Model based on Cross-Modal Latent Representations" in EUSIPCO, 2023
342 International Conference Hejung Yang, Hong-Goo Kang "Feature Normalization for Fine-tuning Self-Supervised Models in Speech Enhancement" in INTERSPEECH, 2023
341 International Conference Jihyun Kim, Hong-Goo Kang "Contrastive Learning based Deep Latent Masking for Music Source Seperation" in INTERSPEECH, 2023
340 International Conference Woo-Jin Chung, Doyeon Kim, Soo-Whan Chung, Hong-Goo Kang "MF-PAM: Accurate Pitch Estimation through Periodicity Analysis and Multi-level Feature Fusion" in INTERSPEECH, 2023
339 International Conference Hyungchan Yoon, Seyun Um, Changhwan Kim, Hong-Goo Kang "Adversarial Learning of Intermediate Acoustic Feature for End-to-End Lightweight Text-to-Speech" in INTERSPEECH, 2023
338 International Conference Hyungchan Yoon, Changhwan Kim, Eunwoo Song, Hyun-Wook Yoon, Hong-Goo Kang "Pruning Self-Attention for Zero-Shot Multi-Speaker Text-to-Speech" in INTERSPEECH, 2023
337 International Conference Doyeon Kim, Soo-Whan Chung, Hyewon Han, Youna Ji, Hong-Goo Kang "HD-DEMUCS: General Speech Restoration with Heterogeneous Decoders" in INTERSPEECH, 2023