Identification of English /r/ and /l/ in white noise by native and non-native listeners

ATRHumanInformationScienceLaboratories,2–2–2,Hikaridai,Seika-cho,Soraku-gun,Kyoto,619–0288Japan(Received14May2002,Acceptedforpublication10June2002)Keywords: Consonantidentification,Signal-to-noiseratio,WordperceptionPACSnumber: 43.71.Es,43.71.Gv,43.71.Hw,43.66.Dc1. IntroductionIt is known that second language learners often havedifficulty in distinguishing phonetic contrasts that do notoccur in their native languages [1,2], partly because theyperceptually assimilate those contrasts into their nativecategories [3,4]. For example, Japanese-native listeners finditveryhardtoidentify/r/and/l/correctly.Nonetheless,itisquite possible to train Japanese-native listeners so that theycan attain 90% accuracy in /r/-/l/ identification under quietconditions[5].It is also known that sentence recognition performancedeterioratesinanoisyenvironment[6],andtheperformancebecomesevenworsewhenthespeechtoberecognizedisinanon-native language than when it is in the listener’s nativelanguage. This seems to be partly because, in a noisyenvironment, native listeners can use contextual cues [7–9]moreeffectivelythannon-nativelisteners[10–13].Given that, it is still important to investigate consonantrecognitionperformanceinanisolatedword,becausethereisapossibilitythattheperceptualframeworksoflistenersaffecttheextentofnoisetoleranceinphonemeperceptionforanon-nativelanguage.TakataandNa´be˘lek[14]studiedtheeffectofbabble noise and reverberation on consonant recognition byEnglish-nativeandJapanese-nativelisteners.Theyfoundthatthe performance of the Japanese-native listeners was lowerwith the babble noise and reverberation than that of theEnglish-native listeners, although both groups of listenersperformedequallywellunderquietconditions.However,weareunawareofanyinvestigationthatsystematicallyexplorestheeffectofnoiseonconsonantidentificationbynativeandnon-nativelisteners.Thus, the purpose of our present investigation is tocompare English /r/-/l/ identification performance betweennative and non-native listeners, with systematically changedsignal-to-noiseratios.2. Experiment2.1. StimuliThe stimuli consisted of 53 pairs of English wordsminimally contrasting in /r/ and /l/ in one of five phoneticenvironments: initial singleton, initial consonant cluster,intervocalic,finalconsonantcluster,andfinalsingleton.Eachword was produced by one of four speakers of AmericanEnglish (2 males and 2 females). Speakers A (male) and B(female) produced 15 minimal pairs, speaker C (male)produced14pairs,andspeakerD(female)produced9pairs,foratotalof106words.Therecordingsweredigitizedat16-bit resolution and 22.05-kHz sampling frequency, and eachwordwassavedintoasoundfile.The stimuli were presented through an adapter (STAX,SRM-1/MK2 pp) and headphones (STAX, SR LambdaSignature). The sound pressure levels of the stimuli weremeasuredwithanIECcoupler(Bru¨elK (2) Two levels ofconsonant contrasts: /r/-/l/ and filler condition; and (3) Twolevels of the native languages of the participants: AmericanEnglish and Japanese. The experiment conditions wereformedwithafactorialcombinationofthesefactors.2.3. ParticipantsTwenty-six native speakers of American English, agedfrom 16 to 44 with an average age of 31.0, participated asEnglish-native listeners. Seventeen native speakers of Japa-nese, aged from 19 to 23 with an average age of 21.5,