Search

Speech is one of the most fundamental mediums for human-to-human interaction, thereby playing a pivotal role in shaping the landscape of next-generation human-computer interaction (HCI). The development of an accurate speech emotion recognition (SER) system for human conversation is a critical yet challenging task. Most state-of-the-art research work in SER predominantly centers around the individual modeling of vocal attributes within each discrete speech utterance, often overlooking the integration of transactional cues intrinsic to the broader interactive context. In this paper, we introduce an innovative dual-level framework designed for the recognition of speech emotions, which leverages the complementary attributes of MFCC features and Mel-spectrograms. Furthermore, we propose a hierarchical attention mechanism designed to effectively include contextual information, hence improving the accuracy of emotion recognition. Our experimentation, conducted on the widely recognized IEMOCAP emotional benchmark dataset, yields promising results. Compared to state-of-the-art methods in four-class emotion recognition, our model demonstrates a substantial advancement, achieving a weighted accuracy of 75.0% and an unweighted accuracy of 75.9%. This marks a notable enhancement of 5.8% in terms of unweighted accuracy, underscoring the efficacy of our approach. This work contributes to the advancement of SER by effectively utilizing multiple audio representations and contextual information. The significant improvements underscore the efficacy of our approach, promising more accurate emotion recognition in human-computer interaction and affective computing. [ABSTRACT FROM AUTHOR]

The use of machine learning and artificial intelligence enables us to create intelligent systems. Speech emotion recognition system analyzes the speaker's speech to determine his/her emotional state. Speech emotion recognition is a challenging pattern recognition task. This paper proposes a new robust and lightweight speech emotion recognition system based on a dual-stream CNN-Transformer fusion network that effectively captures spatial and temporal information disentangled in long-distance raw features by exploring both MFCCs and Mel-spectrograms in a parallel mechanism. Experiments are performed on the widely used emotional benchmark datasets. Our approach has shown to be quite efficient. It outperforms the best-known state-of-the-art models with an accuracy of 97.64%, 99.42%, and 97.53% on RAVDESS, TESS, and EMO-DB datasets respectively. The results demonstrate the significant advantages of the proposed model and the ability of this architecture to learn to recognize emotional data features accurately. [ABSTRACT FROM AUTHOR]

Introduction: Newer generation bioresorbable scaffolds (BRSs) with thinner struts and improved deliverability are expected to enhance safety and efficacy profiles. Bioheart (Bio-Heart, Shanghai, China) BRS is constructed from a PLLA (poly-l-lactic acid) backbone coated with a PDLLA (poly D-l-lactic acid) layer eluting sirolimus. We report 2-year serial intracoronary imaging findings.Methods: In this first-in-human study, 46 patients with single de novo lesions in native coronary vessels (vessel size 3.0-3.75 mm, lesion length ≤ 25 mm) were enrolled at a single institution. Baseline intravascular ultrasound (IVUS) and post-implantation IVUS and optical coherence tomography (OCT) examinations were mandatory. After successful implantations of BRS, the 46 patients were randomized to two different follow-up cohorts in a 2:1 ratio. Thirty patients in cohort 1 had to undergo angiography, IVUS, and OCT follow-ups at 6 and 24 months, respectively. The 16 patients in cohort 2 underwent the same types of imaging follow-ups at 12 and 36 months, respectively. Clinical follow-ups were scheduled uniformly in both cohorts at 1, 6, and 12 months and annually up to 5 years for all patients.Results: Between August and November 2016, a total of 54 patients were assessed. However, 8 patients could not meet all the inclusion criteria; thus, the remaining 46 patients (age 57.5 ± 8.7 years, 34.8% female, 50.0% with unstable angina, 26.1% diabetics) with 46 target lesions were enrolled in this study. All patients in both cohorts were required to complete clinical follow-up uniformly and regularly. In cohort 1, one patient had definite scaffold thrombosis within 6 months of follow-up; thus, after 6 months, cohort 1 had 96.7% patients . Imaging follow-up was available in 24 patients, and in-scaffold late loss was 0.44 ± 0.47 mm; intracoronary imaging confirmed the late loss was mainly due to to neointimal hyperplasia, but not scaffold recoil.Conclusions: Serial 2-year clinical and imaging follow-up results confirmed the preliminary safety and efficacy of Bioheart BRS for treatment of simple coronary lesions. [ABSTRACT FROM AUTHOR]