SMASH Corpus: A Spontaneous Speech Corpus Recording Third-person Audio Commentaries on Gameplay

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 6571–6577 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 6571 SMASH Corpus: A Spontaneous Speech Corpus Recording Third-person Audio Commentaries on Gameplay Yuki Saito, Shinnosuke Takamichi, Hiroshi Saruwatari Graduate School of Information Science and Technology, The University of Tokyo, Japan. 7-3-1 Hongo, Bunkyo-ku, Tokyo, Japan. {yuuki saito, shinnosuke takamichi, hiroshi saruwatari}@ipc.i.u-tokyo.ac.jp Abstract Developing a spontaneous speech corpus would be beneficial for spoken language processing and understanding. We present a speech corpus named the SMASH corpus, which includes spontaneous speech of two Japanese male commentators that made third-person audio commentaries during the gameplay of a fighting game. Each commentator ad-libbed while watching the gameplay with various topics covering not only explanations of each moment to convey the information on the fight but also comments to entertain listeners. We made transcriptions and topic tags as annotations on the recorded commentaries with our two-step method. We first made automatic and manual transcriptions of the commentaries and then manually annotated the topic tags. This paper describes how we constructed the SMASH corpus and reports some results of the annotations. Keywords: spontaneous speech corpus, gameplay, third-person audio commentary, transcription, topic tagging 1. Introduction Spontaneous speech, typically uttered by one person with- out planning what to say in advance, has many aspects to be studied for deeper understanding of our speech commu- nication. Various techniques for spontaneous speech pro- cessing and understanding have been proposed coinciding with the progress in machine learning research (e.g., recog- nition (Furui, 2003; Toyama et al., 2017), synthesis (Werner et al., 2004; Sz´ekely et al., 2019), and summarization (Fu- rui et al., 2004; Szasz´ak et al., 2017)). Speech corpora in- cluding a large variety of spontaneous speech are required to develop these techniques further. Not only the corpora themselves but also methodologies for recording sponta- neous speech are important. The largest spontaneous speech corpus in Japanese is the Corpus of Spontaneous Japanese (CSJ) (Maekawa et al., 2000), which contains about 660 hours of sponta- neous speech uttered by 1,417 Japanese speakers with var- ious speaking styles. Various situations for recording di- verse spontaneous speech were organized for construct- ing this enormous corpus (e.g., actual academic presen- tation, simulated public speaking, and task-oriented dia- logues). Although the CSJ is undoubtedly beneficial to both speech and natural language researchers, such a large-scale methodology for the data collection is naturally difficult for other researchers to follow. One easier way to compile nat- urally uttered spontaneous speech is to make a recording situation where the speakers have many opportunities to see things in their daily lives and to express their thoughts and emotions induced by the situation with no hesitations. The Online Gaming Voice Chat (OGVC) corpus (Arimoto et al., 2012) is one example that utilizes such a situation. A voice chat system was adopted for it as a natural communi- cation tool for speakers to talk with each other while play- ing well-known massively multiplayer online …

Content truncated to 500 words