Open AccessGraph Learning Based Speaker Independent Speech Emotion Recognition
Author(s) -
Xiaoli Xu,
Chengwei Huang,
Chen Wu,
Qunbo Wang,
Li Zhao
Publication year - 2014
Publication title -
advances in electrical and computer engineering
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.254
H-Index - 23
eISSN - 1844-7600
pISSN - 1582-7445
DOI - 10.4316/aece.2014.02003
Subject(s) - computer science , speech recognition , speaker recognition , graph , speaker diarisation , emotion recognition , artificial intelligence , natural language processing , theoretical computer science
In this paper, the algorithm based on graph learning and graph embedding framework, Speaker-Penalty Graph Learning (SPGL), is proposed in the research of speech emotion recognition to solve the problems caused by different speakers. Graph embedding framework theory is used to construct the dimensionality reduction stage of speech emotion recognition. Special penalty and intrinsic graphs of the graph embedding framework is proposed to penalize the impacts from different speakers in the task of speech emotion recognition. The original speech emotion features are extracted by various categories, reflecting different characteristics of each speech sample. According to the experiments in speech emotion corpus using different classifiers, the proposed method with linear and kernelized mapping forms can both achieve relatively better performance than the state-of-the-art dimensionality reduction methods