Selected work

Representative Papers

Effects of automotive microphone frequency response characteristics and noise conditions on speech and ASR quality -- an experimental evaluation

Oct 10, 2025

This study addresses the critical microphone selection problem for in-vehicle hands-free communication and automatic speech recognition (ASR) systems, focusing on the practical impact of frequency response characteristics—specifically bandwidth and amplitude-frequency curve—under realistic in-cabin noise conditions. Leveraging the ETSI TS 103 281 standard (S-MOS/N-MOS/G-MOS), signal-to-noise ratio (SNR), and word error rate (WER), we conduct a systematic evaluation using multi-source noise data collected from real vehicles. Our analysis quantifies, for the first time, the nonlinear effects of low-frequency cutoff, high-frequency upper limit, and mid-band gain profile on noise robustness and ASR accuracy. Based on these findings, we propose microphone frequency response optimization guidelines tailored to automotive environments. These guidelines provide empirically validated, reproducible evidence and actionable engineering insights for defining automotive-grade microphone specifications and guiding system-level acoustic design.

0 citationsRead paper

Fine-Grained Erasure in Text-to-Image Diffusion-based Foundation Models

Mar 25, 2025

Existing concept forgetting methods for text-to-image diffusion models often overlook semantic proximity, leading to collateral damage to semantically related concepts during target concept erasure—the adjacency challenge. This paper introduces FADE, the first adjacency-aware fine-grained forgetting framework. FADE constructs concept neighborhoods via a semantic graph and jointly optimizes three objectives—Expungement, Adjacency preservation, and Guidance fidelity—using its novel Mesh module. It employs gradient-constrained optimization over diffusion model parameters and a multi-objective weighted loss design. Evaluated on six benchmarks including Stanford Dogs, FADE significantly reduces adjacent-concept degradation, improves knowledge retention by ≥12% over state-of-the-art methods, and maintains both thorough concept erasure and model generalization stability.

0 citationsRead paper

Recent publications

Latest Papers

Effects of automotive microphone frequency response characteristics and noise conditions on speech and ASR quality -- an experimental evaluation

Oct 10, 2025

This study addresses the critical microphone selection problem for in-vehicle hands-free communication and automatic speech recognition (ASR) systems, focusing on the practical impact of frequency response characteristics—specifically bandwidth and amplitude-frequency curve—under realistic in-cabin noise conditions. Leveraging the ETSI TS 103 281 standard (S-MOS/N-MOS/G-MOS), signal-to-noise ratio (SNR), and word error rate (WER), we conduct a systematic evaluation using multi-source noise data collected from real vehicles. Our analysis quantifies, for the first time, the nonlinear effects of low-frequency cutoff, high-frequency upper limit, and mid-band gain profile on noise robustness and ASR accuracy. Based on these findings, we propose microphone frequency response optimization guidelines tailored to automotive environments. These guidelines provide empirically validated, reproducible evidence and actionable engineering insights for defining automotive-grade microphone specifications and guiding system-level acoustic design.

0 citationsRead paper

Fine-Grained Erasure in Text-to-Image Diffusion-based Foundation Models

Mar 25, 2025

Existing concept forgetting methods for text-to-image diffusion models often overlook semantic proximity, leading to collateral damage to semantically related concepts during target concept erasure—the adjacency challenge. This paper introduces FADE, the first adjacency-aware fine-grained forgetting framework. FADE constructs concept neighborhoods via a semantic graph and jointly optimizes three objectives—Expungement, Adjacency preservation, and Guidance fidelity—using its novel Mesh module. It employs gradient-constrained optimization over diffusion model parameters and a multi-objective weighted loss design. Evaluated on six benchmarks including Stanford Dogs, FADE significantly reduces adjacent-concept degradation, improves knowledge retention by ≥12% over state-of-the-art methods, and maintains both thorough concept erasure and model generalization stability.

0 citationsRead paper