A crowdsourcing approach to labelling a mood induced speech corpora: 4th International Workshop on Corpora for Research on Emotion Sentiment & Social Signals (ES³ 2012)

This paper demonstrates the use of crowdsourcing to accumulate ratings from nave listeners as a means to provide labels for a naturalistic emotional speech dataset. In order to do so, listening tasks are performed with a rating tool, which is delivered via the web. The rating requirements are based on the classical dimensions, activation and evaluation, presented to the participant as two discretised 5-point scales. Great emphasis is placed on the participant's overall understanding of the task, and on the ease-of-use of the tool so that labelling accuracy is reinforced. The accumulation process is ongoing with a goal to supply the research community with a publicly available speech corpus.