Analysis, Design and Implementation Considerations of a Speech Coder Based on LPCNet
Author
Roskær, Christian Bekhøi
Term
4. term
Education
Publication year
2020
Submitted on
2020-06-30
Pages
166
Abstract
In this project made in collaboration with RTX, we examine the newly proposed voice decoding neural network called LPCNet, with the intent of implementation on an embedded device. This work includes an analysis of the many methods and theories utilized by LPCNet. An error of the source code is corrected, and new models are trained using the TIMIT data set, yielding intelligible but unnatural sounding speech. The part of the LPCNet with the highest computational complexity is selected, mainly consisting of two GRU layers. This sub-algorithm of the LPCNet is then further analysed, in order to gain insights of its inner workings. This analysis have among others resulted in the proposed Block Compressed Sparse Column (BCSC) format, as a means to address the block-sparse matrices utilized by the network. In an FPGA based approach different data moment schemes where explored. In order to increase energy efficiency, while maintain the throughput of the solution. In a CPU based implementation different causes for processing stalls were investigated, and memory hierarchy was identified as the biggest reason for stalls of the sub-algorithm. Parallel processing have been emphasized in both approaches. Practical implementation results of these approaches have not been obtained within the time frame of the project.
Documents