2021/08/26 by Luis Felipe Parra-Gallego, Parra-Gallego, Luis Felipe, Juan Rafael Orozco-Arroyave +2 · 1 citation
Computer Science · Engineering · Psychology · #Articulation (sociology) #Artificial intelligence #Audio and Speech Processing (eess.AS) #Computer science #Customer satisfaction #Emotion and Mood Recognition #FOS: Computer and information sciences #FOS: Electrical engineering #Feature (linguistics) #Feature vector #Linguistics #Machine Learning (cs.LG) #Natural language processing #Phonation #Phone #Prosody #Set (abstract data type) #Sound (cs.SD) #Speech Recognition and Synthesis #Speech and Audio Processing #Speech recognition #cs.LG #cs.SD #eess.AS #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2108.11981
arxiv created 2021/08/26 · openalex publication_date 2021/08/26 · arxiv updated 2021/08/30 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
This paper focuses on finding suitable features to robustly recognize\nemotions and evaluate customer satisfaction from speech in real acoustic\nscenarios. The classification of emotions is based on standard and well-known\ncorpora and the evaluation of customer satisfaction is based on recordings of\nreal opinions given by customers about the received service during phone calls\nwith call-center agents. The feature sets considered in this study include two\nspeaker models, namely x-vectors and i-vectors, and also the well known feature\nset introduced in the Interspeech 2010 Paralinguistics Challenge (I2010PC).\nAdditionally, we introduce the use of phonation, articulation and prosody\nfeatures extracted with the DisVoice framework as alternative feature sets to\nrobustly model emotions and customer satisfaction from speech. The results\nindicate that the I2010PC feature set is the best approach to classify emotions\nin the standard databases typically used in the literature. When considering\nthe recordings collected in the call-center, without any control over the\nacoustic conditions, the best results are obtained with our articulation\nfeatures. The I2010PC feature set includes 1584 measures while the articulation\napproach only includes 488 measures. We think that the proposed approach is\nmore suitable for real-world applications where the acoustic conditions are not\ncontrolled and also it is potentially more convenient for industrial\napplications.\n