The INTERSPEECH 2010 Paralinguistic Challenge

摘要: Abstract Most paralinguistic analysis tasks are lacking agreed-uponevaluation procedures and comparability, in contrast to more‘traditional’ disciplines speech analysis. The INTERSPEECH2010 Paralinguistic Challenge shall help overcome the usuallylow compatibility of results, by addressing three selected sub-challenges. In Age Sub-Challenge, age speakers hasto be determined four groups. Gender Sub-Challenge,a three-class classification task has solved finally, theAffect Sub-Challenge asks for speakers’ interest ordinal rep-resentation. This paper introduces conditions, Challengecorpora “aGender” “TUM AVIC” standard feature setsthat may used. Further, baseline results given.Index Terms: Challenge, Age, Gender, Affect 1. Introduction resemble each other not onlyby means processing ever-present data sparseness, but bylacking agreed-upon evaluation comparability,in more traditional Atthe same time, this is a rapidly emerging field research, dueto constantly growing on applications fieldsof Human-Machine Communication, Human-Robot Communi-cation, Multimedia Retrieval. these respects, INTER-SPEECH 2010 bridging thegap between excellent research information inspoken language low address-ing tasks. AVIC”corpora provided organizers. first consists 46hours telephone speech, stemming from 954 speakers, andserves evaluate features algorithms detection ofspeaker gender. second 2 hours humanconversational recording (21 subjects), annotated 5differentlevelsofinterest. Thecorpusfurtherfeaturesauniquelydetailed transcription spoken content with word boundaries byforced alignment, non-linguistic vocalizations, single annotatortracks, sequence (sub-)speaker-turns. Both given