SOFTWARE BASED PARALLELIZATION OF SPEECH RECOGNITION SYSTEM ON EMBEDDED PLATFORMS
The task of large vocabulary speech recognition is computationally hard with the constant increase in complexity of acoustic and language models. Efficient speech recognition demands that processing is done in real time given the constraints of memory and power of embedded devices. With the incorporation of multi-core processors in embedded systems, it is desirable that the resources provided by multiprocessors are utilized to the fullest. Pocketsphinx is a speech recognition system written in C designed for the purpose of developing applications for embedded systems and software. The goal of the proposed work is to identify time consuming modules in the sequential algorithm implemented in Pocketsphinx and parallelize it efficiently on a shared memory multicore architecture using OpenMP.