Yunpeng Liu

Yunpeng Liu

Auto-generated profile, based on publicly available information

Yunpeng Liu

School of Information Systems Engineering, University of information Engineering, Zhengzhou, China

Research Focus & Metrics

Academic Career Length

6Years (2021-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 June 2024 in EURASIP Journal on Audio, Speech, and Music Processing
Verified
Journal Article
Open access
PDF file
YLYunpeng Liu
Yunpeng Liu
XYXukui Yang
Xukui Yang
DQDan Qu
Dan Qu
Limited data availability remains a significant challenge for Whisper’s low-resource speech recognition performance, falling short of practical application requirements. While previous studies have successfully reduced the recognition error rates of target language speech through fine-tuning, a comprehensive exploration and analysis of Whisper’s fine-tuning capabilities and the advantages and disadvantages of various fine-tuning strategies are still lacking. This paper aims to fill this gap by conducting comprehensive experimental exploration for Whisper’s low-resource speech recognition performance using five fine-tuning strategies with limited supervised data from seven low-resource languages. The results and analysis demonstrate that all fine-tuning strategies explored in this paper significantly enhance Whisper’s performance. However, different strategies vary in their suitability and practical effectiveness, highlighting the need for careful selection based on specific use cases and resources available.

3 January 2025 in IEEE Signal Processing Letters
Verified
Journal Article
Open access
YLYunpeng Liu
Yunpeng Liu
XYXukui Yang
Xukui Yang
JZJiayi Zhang
Jiayi Zhang
YXYangli Xi
Yangli Xi
DQDan Qu
Dan Qu
Parameter-efficient fine-tuning of pre-trained multilingual speech models can significantly enhance the speech recognition performance of target languages. However, traditional parameter-efficient fine-tuning methods, such as adapter tuning, often face challenges related to random initialization. This can lead to suboptimal performance when adapting to languages with limited resources. To address this issue, this paper introduces TAML-Adapter, which utilizes the Task-Agnostic Meta-Learning algorithm to initialize the parameters of the adapters before fine-tuning in target low-resource languages. Comprehensive experiments conducted on the Common Voice and Fleurs datasets highlight the superior performance of TAML-Adapter in five languages with limited resources. In addition, the TAML-Adapter demonstrates superior generalizability and extensibility compared to similar competing methods.

5 May 2025 in Neural Processing Letters
Verified
Journal Article
Open access
PDF file
YLYunpeng Liu
Yunpeng Liu
XYXukui Yang
Xukui Yang
DQDan Qu
Dan Qu
Multilingual automatic speech recognition represents a crucial research direction in tackling the challenges associated with low-resource scenarios. To effectively incorporate language-specific information in the joint training of a model across multiple languages and leverage linguistic similarities to enhance performance on the target low-resource language, this paper introduces a language similarity evaluation approach based on forced decoding. Specifically, when the target language is specified, the speeches of the source language are decoded into transcription in the target language, and the normalized posterior is utilized as the foundation for evaluating language similarity. Comprehensive experiments and analyses conducted on six low-resource languages reveal that the proposed approach achieves an average word error rate relative reduction of 21.74, 7.68, and 3.45% compared to three widely used benchmark methods, respectively, thereby validating the effectiveness of our approach.

9 October 2021
Verified
Conference Paper
Closed access
PDF file
YLYunpeng Liu
Yunpeng Liu
CWChenqing Wang
Chenqing Wang
TJTao Jing
Tao Jing
XYXingpeng Yan
Xingpeng Yan
Holographic stereogram opens up a new way for holographic 3D display of objects, and has high research value and significance in commercial, military and other aspects. In order to write real and virtual three-dimensional (3D) scenes in holographic stereogram at the same time, so as to achieve the display effect of augmented reality, a light field fusion method is proposed. The basic principle of light field fusion is introduced. The light field sampling is introduced and completed, and depth image based rendering (DIBR) algorithm is applied for regularization and densification of sampled images. The projection relationship of the light rays are analyzed, and the pixel level light field fusion is accomplished. Effective perspectives segmentation and mosaicking (EPISM) holographic stereogram printing method is used to write and reproduce the fused light field which further verifies the effectiveness of the proposed method.

21 October 2025 in Signal, Image and Video Processing
Verified
Journal Article
Closed access
PDF file
YLYunpeng Liu
Yunpeng Liu
XYXukui Yang
Xukui Yang
DQDan Qu
Dan Qu

13 September 2024
Verified
Conference Paper
Open access
QMQuanGong Ma
QuanGong Ma
CHChaoLong Hao
ChaoLong Hao
YLYunpeng Liu
Yunpeng Liu
NSNianwen Si
Nianwen Si
JZJiale Zhang
Jiale Zhang
DQDan Qu
Dan Qu
Parameterized quantum circuits play an essential role in the performance of many variational hybrid quantum-classical (HQC) algorithms. One challenge in implementing such algorithms is to choose an effective circuit that well represents the solution space while maintaining a low circuit depth and number of parameters. In this paper, we delve into the influence of utilizing quantum circuits with different types, depths, and widths in the Quantum Generative Adversarial Network (QGAN) generator on its overall performance. Our findings reveal that employing parametric two-qubit gates in the QGAN generator slightly outperforms the use of fixed-parameter two-qubit gates. Additionally, as the number of quantum circuit layers increases (i.e., the circuit depth), the performance of the QGAN generator improves, though there exists ’saturation’ where further layer addition leads to a decline in performance. Furthermore, considering the number of qubits available for the QGAN generator to accomplish the generation task, we observe a gradual decline in QGAN's performance as the number of ancillary qubits (i.e., circuit width) increases, though it still outperforms the scenario without auxiliary qubits. Additionally, we introduce the Unrolled QGAN model, which incorporates the Unrolled training technique to improve the performance of QGAN. Our experimental results demonstrate that this technique effectively enhances the performance of the QGAN.