用知识蒸馏让语音增强模型变小，性能不降反升。
Comparison of Knowledge Distillation Methods for Low-complexity Multi-microphone Speech Enhancement using the FT-JNF Architecture
- 用教师-学生框架，通过多层特征匹配压缩模型。
- 参数减至25%时，0dB信噪比下语音质量接近原模型。
- 模型大小最多缩小96%，适合嵌入式设备部署。
近年来，基于深度神经网络（DNN）的多麦克风语音增强技术取得了显著进展。然而，许多现有算法因计算资源受限而难以在低复杂度设备上部署。单纯减少参数量常导致性能下降。知识蒸馏（KD）是一种有望在保持性能的同时降低模型复杂度的有效方法。本文以近期提出的频时联合非线性滤波器（FT-JNF）架构为基础，评估了五种不同的知识蒸馏方法，包括直接输出匹配、中间层自相似性建模以及融合多层损失。在包含五个麦克风的紧凑阵列模拟数据集上的实验表明，三种KD方法显著提升了学生模型的性能。当学生模型仅保留教师模型25%的参数时，在0 dB信噪比下仍能达到相近的语音质量（PESQ得分）。此外，模型规模最多可缩减96%，且PESQ得分下降极小。
原文摘要 · Abstract (English)
Multi-microphone speech enhancement using deep neural networks (DNNs) has significantly progressed in recent years. However, many proposed DNN-based speech enhancement algorithms cannot be implemented on devices with limited hardware resources. Only lowering the complexity of such systems by reducing the number of parameters often results in worse performance. Knowledge Distillation (KD) is a promising approach for reducing DNN model size while preserving performance. In this paper, we consider the recently proposed Frequency-Time Joint Non-linear Filter (FT-JNF) architecture and investigate several KD methods to train smaller (student) models from a large pre-trained (teacher) model. Five KD methods are evaluated using direct output matching, the self-similarity of intermediate layers, and fused multi-layer losses. Experimental results on a simulated dataset using a compact array with five microphones show that three KD methods substantially improve the performance of student models compared to training without KD. A student model with only 25% of the teacher model's parameters achieves comparable PESQ scores at 0 dB SNR. Furthermore, a reduction of up to 96% in model size can be achieved with only a minimal decrease in PESQ scores.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。