Papers

On Fine-Tuning Pre-Trained Speech Models With EMA-Target Self-Supervised Loss

International Conference
2021~
작성자
김병현
작성일
2023-12-14 16:35
조회
1986
Authors : Hejung Yang, Hong-Goo Kang

Year : 2024

Publisher / Conference : ICASSP

Research area : Speech Signal Processing

Presentation/Publication date : 2024.04.19

Presentation : Poster

Representation models pre-trained on self-supervised objectives are often fine-tuned for solving downstream tasks.
However, fine-tuning can degrade the general knowledge that was originally built up by the pre-training, which could help prevent the model from overfitting given sparse fine-tuning data or bridge gaps between different domains.
We hypothesize that preserving this general knowledge in pre-trained models is crucial for improving performance on downstream tasks.
Based on this idea, we propose a novel method for fine-tuning self-supervised speech models that utilizes a self-supervised loss over the course of fine-tuning.
Then, an Exponential Moving Average (EMA) technique is applied to smoothly transition the domain of the model from the generalized to the task-oriented one.
We perform various downstream tasks using the proposed method, finding that our method improves performance on most of the tasks. Results show that our method induces the generalization ability of the model to be retained without overshadowing the downstream task performance.
전체 370
162 International Conference Yeona Hong, Hyewon Han, Woo-jin Chung, Hong-Goo Kang "StableQuant: Layer Adaptive Post-Training Quantization for Speech Foundation Models" in IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2025
161 International Conference Sangmin Lee, Woojin Chung, Hong-Goo Kang "LAMA-UT: Language Agnostic Multilingual ASR through Orthography Unification and Language-Specific Transliteration" in Association for the Advancement of Artificial Intelligence (AAAI), 2025
160 International Conference Juhwan Yoon, Hyungseob Lim, Hyeonjin Cha, Hong-Goo Kang "StylebookTTS: Zero-Shot Text-to-Speech Leveraging Unsupervised Style Representation" in APSIPA ASC, 2024
159 International Conference Doyeon Kim, Yanjue Song, Nilesh Madhu, Hong-Goo Kang "Enhancing Neural Speech Embeddings for Generative Speech Models" in APSIPA ASC, 2024
158 International Conference Miseul Kim, Soo-Whan Chung, Youna Ji, Hong-Goo Kang, Min-Seok Choi "Speak in the Scene: Diffusion-based Acoustic Scene Transfer toward Immersive Speech Generation" in INTERSPEECH, 2024
157 International Conference Seyun Um, Doyeon Kim, Hong-Goo Kang "PARAN: Variational Autoencoder-based End-to-End Articulation-to-Speech System for Speech Intelligibility" in INTERSPEECH, 2024
156 International Conference Jihyun Kim, Stijn Kindt, Nilesh Madhu, Hong-Goo Kang "Enhanced Deep Speech Separation in Clustered Ad Hoc Distributed Microphone Environments" in INTERSPEECH, 2024
155 International Conference Woo-Jin Chung, Hong-Goo Kang "Speaker-Independent Acoustic-to-Articulatory Inversion through Multi-Channel Attention Discriminator" in INTERSPEECH, 2024
154 International Conference Juhwan Yoon, Woo Seok Ko, Seyun Um, Sungwoong Hwang, Soojoong Hwang, Changhwan Kim, Hong-Goo Kang "UNIQUE : Unsupervised Network for Integrated Speech Quality Evaluation" in INTERSPEECH, 2024
153 International Conference Yanjue Song, Doyeon Kim, Hong-Goo Kang, Nilesh Madhu "Spectrum-aware neural vocoder based on self-supervised learning for speech enhancement" in EUSIPCO, 2024