arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2606.08078cs.SDcs.CL
收藏

说话人验证中的低位量化误差:诊断与缓解

On Low-Bit Quantization Errors in Speaker Verification: Diagnostic and Mitigation

LIA, UPR 4128;Avignon University(阿维尼翁大学);Aday

Hugo Leguillier, Driss Matrouf, Guillaume Lechien, Mickael Rouvier

AI总结:

本文通过逐层和得分级分析,诊断了低比特量化对说话人验证的影响,发现2比特是关键拐点,并提出校准多精度级联方法,在保持低位推理效率的同时接近全精度性能。

摘要:

尽管低比特量化为在资源受限设备上部署说话人验证提供了实用手段,但其对说话人验证性能的影响仍知之甚少。本文通过联合逐层和得分级分析,研究了ResNet-36和ResNet-200的均匀K-means量化感知训练。我们的逐层分析突出了脆弱组件,并表明得分退化不能仅由权重失真完全解释。我们在2比特处识别出一个明显的拐点,较大的得分漂移和有害决策翻转集中在FP32阈值附近。我们的得分级分析揭示了在极端量化下得分误差产生的位置和方式。基于这些发现,我们提出了一种校准的多精度级联方法,该方法在2比特下解决大多数试验,仅升级模糊情况,实现了接近FP32的性能,同时以显著降低的计算和内存成本保留了低位推理的效率优势。

英文摘要:

Although low-bit quantization provides practical means to deploy speaker verification on resource-constrained devices, its effects on speaker verification performance remain poorly understood. In this paper, we study uniform K-means quantization-aware training of ResNet-36 and ResNet-200 through joint layer-wise and score-level analyses. Our layer-wise analysis highlights fragile components and shows that score degradation is not fully explained by weight distortion alone. We identify a clear knee point at 2 bits, with larger score drift and harmful decision flips concentrated near the FP32 threshold. Our score-level analysis reveals where and how score errors emerge under extreme quantization. Building on these findings, we propose a calibrated multi-precision cascade that resolves most trials at 2 bits and escalates only ambiguous cases, achieving performance close to FP32 while preserving the efficiency benefits of low-bit inference with substantially lower compute and memory costs.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑