Prosodic Features ForLanguage Identirication

Inthis paper, weexamine theeffectiveness syllables orwordsaremademoreprominent than of prosodic features forlanguage identification. others, resulting instress. Theinformation gleaned Prosodic differences amongworldlanguages include fromthemelody, timing, and stress inspeech variations inintonation, rhythm, andstress. These increases theintelligibility ofspoken message, and variations arerepresented using features derived from alsoconveys information suchaslexical tone, accent, fundamental frequency (FO) contour, duration, and andemotion. The characteristics thatmake us energy contour. Forextracting theprosodic features,perceive theseeffects arecollectively referred toas speech signal issegmented intosyllable-like units by prosody. Prosody hasa greatdealtoofferfor locating vowel-onset points(VOP)automatically. effective humanlanguage identification. Various parameters arethenderived torepresent Fo contour, duration, andenergy contour characteristicsMuchoftheLIDresearch sofarhasplaced its foreachsyllable-like unit. Thefeatures obtained by emphasis on spectral information, mainly using the concatenating theparameters derived fromthree consecutive syllable-like units areusedtorepresent the acoustic atrofson unitst(reerred as prosodic characteristics ofalanguage. Theprosodicphonetics), and theiralignment (referred as features thusderived fromdifferent languages areused phonotactics). Suchsystems may perform wellin totrain a multilayer feedforward neural network similar acoustic conditions (2). Buttheir performance (MLFFNN)classifier forlanguage identification. The degrades dueto noiseand channelmismatch. effectiveness oftheproposed approach isverified using Prosodic features derived frompitchcontour, OregonGraduate Institute (OGI)multi-language amplitude contour andduration arerelatively less telephone speech corpus andNational Institute of affected by channel variations andnoise.This Scienceand Technology (NIST)2003 language motivated us to explore prosodic features, for identification database. language identification. Though thesystems based on spectral features outperform theprosody-based LID