Optimized FIR Filtering Engines for Real-Time Speech and Assistive Communication Devices

Abstract

Hardware Assistive communication aids such as hearing aids, speech-generating aids, and augmentative and alternative communication (AAC) software require real-time speech recognition with very strict limits on latency, power, and computation requirements. The popular choice of filters to use in such systems is Finite Impulse Response (FIR), because of its stability, and linear response in phase, which is important to maintain speech intelligibility. Nonetheless, standard FIR filter implementations do not easily meet the criteria of real time and energy efficiency of embedded assistive devices, especially in the wearable and battery operated context. The current paper introduces the best FIR filtering engine that is specifically designed to use in real-time speech enhancement of assistive communication devices. The suggested methodology involves the exploitation of the symmetry of coefficient, pipelined multiply-accumulate (MAC) architectures and optimization of fixed-point arithmetic to greatly minimise the computational complexity and processing latency. The filtering engine is provided in an embedded processor platform that is a representation of assistive communication hardware running at a standard speech sampling rate. The experimental performance proves that the optimized FIR architecture has processing latency specified at sub-milliseconds and cuts down on energy rate by about 30 40 percent than traditional FIRs and offers good talk quality. Objective speech analysis indicates significant signal/noise enhancement without audible distortion effects, indicating that it will be applicable to assistive speech devices. The above-mentioned results suggest that proposed FIR filtering engine is indeed an efficient, scalable, and realistic solution to speech processing in real-time in intelligent assistive communication systems. The architecture is also appropriate to be incorporated into the next generation assistive devices that have low latency, low power consumption, and improved speech intelligibility.