Advanced Speech Recognition for Naval Systems and Ships
By Cynthia Hoyt
Incorporating an Advanced Speech Recognition (ASR) solution into Naval Command, Combat, and Ship Systems, as well as Future Programs, would provide innovative command and control, as well as advanced communications capabilities to US Navy and potentially all military branches supported by ASNE. SPEAR®, is a proprietary, noise robust ASR engine built through IR&D funding by Think-A-Move (TAM). The SPEAR solution provides a natural user interface for hands free-heads up control, and works well with multi-modal systems. SPEAR was developed to fill two large gaps in speech recognition: noise robustness and domain specific language models. Additionally, unlike many other speech recognition solutions, SPEAR runs locally on any device and does not require connection to a network. SPEAR is hardware agnostic and is compatible with all equipment, devices, and headsets. The noise robust front end is the key to SPEAR’s performance. Commercial Off the Shelf (COTS) speech recognition engines do not perform well in any noise above an office environment. SPEAR outperforms standard COTS speech recognition engines in noise. For commands, SPEAR maintains 97.74% accuracy in 90 dbA (Babble Noise) and 94% accuracy in 105 dbA (Black Hawk Helicopter). The leading COTS speech engines only test at 50% accuracy in 90 dbA. For Naval command and communications, TAM will develop a custom language model specific to naval verbiage. TAM will expand on the current military language model that has been developed for the US Army under past and current contracts. The ASR solution will provide command and control of multiple Naval systems, as well as integrate with current systems. The ASR solution is designed to run locally on the device on which it is installed. This gives the user true mobility for use in remote environments or where there are limited or no communication networks. This solution will improve capabilities for ASNE.