Pilot Corpus for Multisensor Speech Processing

摘要

The estimation of speech parameters and the intelligibility of speech transmitted through low-rate coders are severely degraded when there are high levels of acoustic noise in the speaking environment. The application of nonacoustic and nontraditional sensors, which are less sensitive to acoustic noise than the standard microphone, is being investigated to address this problem, under the DARPA Advanced Speech Encoding program. Sensors investigated include the General Electromagnetic Motion Sensor (GEMS) and the Physiological Microphone (P-mic). In order to enable this research, a Pilot Corpus with simultaneous recordings from multiple sensors has been collected by ARCON Corporation, under subcontract to MIT Lincoln Laboratory. This report describes the corpus collection, including: corpus structure, acoustic noise environments, speech materials, the sensors, and baseline intelligibility evaluations. The corpus includes Diagnostic Rhyme Test (DRT) word lists, sentence lists, and Consonant Vowel Consonant (CVC) nonsense words. Noise environments include: M2 Bradley Fighting Vehicle (M2), Military Operations in Urban Terrain (MOUT), UR- 60 Blackhawk Helicopter (BH), and a Military Command Enclosure (MCE). This pilot corpus has been utilized by a number of DARPA-sponsored research teams for R AND D on advanced speech encoding exploiting multiple sensors in the military noise environments.