Speech Group Receives NSF Grant for Speaker Recognition

Speech Group receives NSF grant "Modeling Idiosyncrasies in Speaking Behavior." Barbara Peskin of ICSI's Speech Group will head a project to improve speaker recognition through incorporation of higher-level features - e.g. derived from prosody, pronunciation patterns, word usage, even distinctive laughs - to augment traditional models that rely on short-term 10-20 msec frames of speech. Incorporating higher-level features should improve overall speaker recognition as well as provide greater robustness to varying acoustic conditions. In addition to exploring new features guided by known linguistic constructs, the project also seeks to mine speaker data for novel "vocal performances" - spectro-temporal patterns that may contain high speaker-characterizing value - through unconstrained search for idiosyncratic acoustic intervals

top