Proceedings Fonetik 2009 - Institutionen för lingvistik

Proceedings, FONETIK 2009, Dept. of Linguistics, Stockholm Universitybars used in the equalizers of sound systems:the current amount of variation is indicated bythe number of bars that are lit up in a stack ofbars, and the highest variation over the past twoseconds is indicated by a lingering top bar. Themeter has a short, constant latency of 100ms.The test group and the control group eachconsisted of 7 students of engineering, 4women and 3 men each. The participants wererecruited from English classes at KTH, andwere exchange students from China, in Swedenfor stays of six months to two years. Participants’proficiency in English was judged bymeans of an internal placement test to be at theupper intermediate to advanced level. The participantsspoke a variety of dialects of Chinesebut used Mandarin with each other and for theirstudies in China. They did not speak Swedishand were using English with their teachers andclassmates.Each participant began the study by givingan oral presentation of about five minutes inlength, either for their English classes or for asmaller group of students. Audio recordingswere made of the presentations using a smallclip-on microphone that recorded directly into acomputer. The presentations were also videorecorded,and participants watched the presentationstogether with one of the researchers,who commented on presentation content, deliveryand language. The individualized trainingmaterial for each subject was prepared from theaudio recordings. A set of 10 utterances, eachof about 5-10 seconds in length, was extractedfrom the participants’ speech. The utteranceswere mostly non-consecutive and were chosenon the basis of their potential to provide examplesof contrastive pitch movement within theindividual utterance. The researcher recordedher own (native-American speaking) versionsof them, making an effort to use her voice asexpressively as possible and making more pitchcontrasts than in the original student version.For example, a modeled version of a student’sflat utterance could be represented as: “AndTHIRDly, it will take us a lot of TIME and EFfortto READ each piece of news.”The participants were assigned to the controlor test groups following the preparation oftheir individualized training material. Participantswere ranked in terms of the global pitchvariation in their first presentation, as follows:they were first split into two lists according togender, and each list was ordered according toinitial global pitch variation. Participants wererandomly assigned pair-wise from the list to thecontrol or test group, ensuring gender balanceas well as balance in initial pitch variation.Four participants who joined the study at a laterdate were distributed in the same manner.Participants completed approximately threehours of training in half-hour sessions; someparticipants chose to occasionally have back-tobacksessions of one hour. The training sessionswere spread out over a period of fourweeks. Training took place in a quiet and privateroom at the university language unit, withoutthe presence of the researchers or otheronlookers. For the first four or five sessions,participants listened to and repeated the teacherversions of their own utterances. They were instructedto listen and repeat each of their 10 utterancesbetween 20 and 30 times. Test groupparticipants received the visual feedback describedabove and were encouraged to speak sothat the meter showed a maximum amount ofgreen bars. The control group was able to listento recordings of their production but receivedno other feedback.Upon completion of the repetitions, bothgroups were encouraged to use the system topractice their second oral presentation, whichwas to be on a different topic than the firstpresentation. For this practice, the part of theinterface designated for ‘free speech’ was used.In these sessions, once again the test participantsreceived visual feedback on their production,while control participants were only ableto listen to recordings of their speech. Within48 hours of completing the training, the participantsheld another presentation, this time aboutten minutes in length, for most of them as partof the examination of their English courses.This presentation was audio recorded.ResultsWe measured development in two ways: overthe roughly three hours of training per student,in which case we compared pitch variation inthe first and the second half of the training foreach of the 10 utterances used for practice, andin generalized form, by comparing pitch variationin two presentations, one before and oneafter training. Pitch estimations were extractedusing the same software used to feed the pitchvariation indicator used in training, an incrementalversion of the getF0/RAPT (Talkin,1995) algorithm. Variation was calculated in amanner consistent with Hincks (2005) by calcu-104