Topological Deep Learning for Speech Data
This work addresses speech recognition challenges by integrating topological data analysis into deep learning, offering an incremental advancement with specific gains in noise robustness.
The study tackled the problem of improving speech recognition networks by designing topology-aware convolutional kernels, resulting in superior performance in phoneme recognition, especially in low-noise scenarios, with demonstrated cross-domain adaptability.
Topological data analysis (TDA) offers novel mathematical tools for deep learning. Inspired by Carlsson et al., this study designs topology-aware convolutional kernels that significantly improve speech recognition networks. Theoretically, by investigating orthogonal group actions on kernels, we establish a fiber-bundle decomposition of matrix spaces, enabling new filter generation methods. Practically, our proposed Orthogonal Feature (OF) layer achieves superior performance in phoneme recognition, particularly in low-noise scenarios, while demonstrating cross-domain adaptability. This work reveals TDA's potential in neural network optimization, opening new avenues for mathematics-deep learning interdisciplinary studies.