Detecting and identifying emotions expressed in speech signals is a very complex task that generally requires processing a large sample size to extract intricate details and match the diversity of human expression in speech. There is not an emotional dataset commonly accepted as a standard test bench to evaluate the performance of the supervised machine learning algorithms when presented with extracted speech characteristics. This work proposes a generic platform to capture and validate emotional speech. The aim of the platform is collaborative-crowdsourcing and it can be used for any language (currently, it is available in four languages such as Spanish, English, German and French). As an example, a module for elicitation of stress in speech through a set of online interviews and other module for labeling recorded speech have been developed. This study is envisaged as the beginning of an effort to establish a large, cost-free standard speech corpus to assess emotions across multiple languages.
CITATION STYLE
Palacios-Alonso, D., Lázaro-Carrascosa, C., López-Arribas, A., Meléndez-Morales, G., Gómez-Rodellar, A., Loro-Álavez, A., … Gómez-Vilda, P. (2019). Assessing an Application of Spontaneous Stressed Speech - Emotions Portal. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 11486 LNCS, pp. 149–160). Springer Verlag. https://doi.org/10.1007/978-3-030-19591-5_16
Mendeley helps you to discover research relevant for your work.