Custom Arithmetic for High-speed, Low-resource ASR Systems

With the skyrocketing popularity of mobile devices, new processing methods tailored for low-resource systems have become necessary. We propose the use of custom arithmetic, arithmetic logic tailored to a specific application. In a system with all parameters quantized to low precision, such arithmetic can be implemented through a set of small, fast table lookups. We present here a framework for the design of such a system architecture, and several heuristic algorithms to optimize system performance. In addition, we apply our techniques to an automatic speech recognition (ASR) application. Our simulations on various architectures show that on most modern processor designs, we can expect a cycle-count speedup of at least 3 times while requiring a total of only 59kB of ROMs to hold the lookup tables.