Cheng Song

Cheng Song

Auto-generated profile, based on publicly available information

Cheng Song

School of Artificial Intelligence, Anhui University, Anhui, China

Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

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

Representative Publications

View All Publications
19 May 2026 in IEEE Transactions on Computational Social Systems
Verified
Journal Article
Closed access
BLBo Li
Bo Li
KZKun Zhang
Kun Zhang
CSCheng Song
Cheng Song
XCXi Chen
Xi Chen
HZHailin Zhou
Hailin Zhou
RHRichang Hong
Richang Hong
Text watermarking focuses on embedding identifiable information into the generated content, which has become increasingly important with the rapid development of large language models (LLMs). Existing watermarking works either divide the vocabulary of LLMs into “green” and “red” tokens for the watermark generation (i.e., token-level watermark), or use the distance of generated sentence embeddings to distinguish the “green” and “red” partitions (i.e., sentence-level watermark). Despite the achieved progress, existing methods remain vulnerable to various attacks or exhibit limited generalization to out-of-distribution (OOD) scenarios. To this end, we focus on sentence-level watermarking and propose a novel semantic-oriented robust text watermark for LLMs (SoTW). Specifically, we first employ a pre-trained embedding model to extract representations of generated sentences. Then, different from existing sentence-level works, we design a novel semantic quantization autoencoder (SQAE) to generate discrete sentence partition representations. Moreover, a semantic loss and a consistency loss are designed to ensure the generalization and robustness of generated watermarks. Furthermore, we develop an efficient detection method for our proposed SoTW. Extensive experiments on three LLMs across three publicly available datasets demonstrate the robustness of SoTW against various attacks in OOD settings. We also release the code to facilitate the community1https://github.com/libolb/SoTW.

27 April 2026 in IEEE Sensors Journal
Verified
Journal Article
Closed access
YWYetong Wang
Yetong Wang
XWXuqiang Wu
Xuqiang Wu
CSCheng Song
Cheng Song
BJBowen Ji
Bowen Ji
SYShuaikang Yin
Shuaikang Yin
BYBenli Yu
Benli Yu
This paper addresses the signal distortion issues in interferometric fiber-optic sensors caused by factors such as non-flat probe frequency response, insufficient detection circuit bandwidth, and imperfect demodulation algorithms. To mitigate these distortions, we introduce a deep neural network based on the Dual-Path Convolution Recurrent Network (DPCRN). The Scale-Invariant Signal-to-Distortion Ratio (SI-SDR) is employed as the loss function, replacing the conventional Mean Squared Error (MSE) in the frequency domain. A dataset consisting of 8,000 10-second audio segments was constructed for training and validation, resulting in a well-trained DPCRN model. Experimental results demonstrate improvements, with the Short-Time Objective Intelligibility (STOI) score increasing by 0.58, the Perceptual Evaluation of Speech Quality (PESQ) score improving by 0.059, and the Signal-to-Distortion Ratio (SDR) increasing by 8.13 dB. These results highlight the effectiveness of the proposed method in compensating for signal distortion in fiber-optic microphones and show promising potential in enhancing the consistency of fiber-optic microphone arrays.