Low-Complexity DNN-Based End-to-End Automatic Speech Recognition using Low-Rank Approximation
2020
Targeting the on-device speech-to-text application for streaming inputs, this paper presents an efficient way to reduce the computational complexity of deep neural networks (DNNs) for attention-based speech processing. The proposed technique applies the singular value decomposition (SVD) to the large-sized matrix multiplications, removing less important computations by utilizing the low-rank approximation. The clipping thresholds are carefully adjusted to relax the computing costs as well as the memory overheads while maintaining the recognition accuracy.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
8
References
0
Citations
NaN
KQI