CCA based feature selection with application to continuous depression recognition from acoustic speech features

Human-Centered Video Feature Selection via mRMR-SCMMCCA for Preference Extraction

“…Then the effectiveness of the use of the canonical correlation for the fea- Fig. 1 Projection results of video features and viewing behavior features based on our method (mRMR-SCMMCCA algorithm), mRMR-CCA algorithm [4] and SLPCCA-OC algorithm [5]. The horizontal and vertical axes correspond to the projection results of video features and viewing behavior features, respectively.…”

“…(3), we derive a new feature selection algorithm, i.e., the mRMR-SCMMCCA algorithm. Specifically, we perform the optimal feature selection one-by-one in the same manner as the previously reported feature selection algorithms [3], [4]. Specifically, the selection of kth optimal video feature is performed as…”

“…The study of feature selection has been intensively carried out, and many benchmarking and state-of-the-art methods such as mRMR (minimum Redundancy Maximum Relevance) algorithm [3] and mRMR-CCA (minimum Redundancy Maximum Relevance-Canonical Correlation Analysis) algorithm [4] above feature selection algorithms only monitor the relationship between two modalities, e.g., video features and their corresponding rating scores which represent preference degrees of videos. Although we have proposed feature selection algorithm using SLPCCA-OC (Supervised Locality Preserving Canonical Correlation Analysis with Ordinal Classes) algorithm [5], it is difficult to simultaneously use correlations between video features and "viewing behavior features and rating scores of users".…”

State-of-the-art tasks and achievements of paralinguistic speech analysis systems

“…Данный конкурс соревнования был выигран коллективом турецкого Босфорского универ-ситета, г. Стамбул. В разработанной ими системе для оценки степени физической нагрузки (контролиро-валась посредством частоты сердцебиения), накладываемой на диктора, использовались многопоточные признаки аудиосигнала (multi-view discriminative projection), вычисляемые из LLD-наборов признаков [47,49]. Разработанная система позволила превзойти базовую систему распознавания, показав наилучшее среднее значение полноты 75,4% (Unweighted Average Recall, UAR=75,4%).…”

Implementing Fusion Techniques for the Classification of Paralinguistic Information

“…This includes feature selection in Support Vector Machine (SVM) classifier [13] for emotion recognition, ensemble feature selection in [14] for model adaptation, and recent openSMILE [3]. Canonical correlation analysis (CCA) is employed in [15] for selecting apt features from a set of speech features for depression recognition. Acoustic, linguistic and psycholinguistic features are employed in [16] for learning the personality traits from the spoken data.…”