Leveraging Cross-Attention Transformer and Multi-Feature Fusion for
Cross-Linguistic Speech Emotion Recognition
0
Citation
0
Reference
10
Related Paper
Abstract:
Speech Emotion Recognition (SER) plays a crucial role in enhancing human-computer interaction. Cross-Linguistic SER (CLSER) has been a challenging research problem due to significant variability in linguistic and acoustic features of different languages. In this study, we propose a novel approach HuMP-CAT, which combines HuBERT, MFCC, and prosodic characteristics. These features are fused using a cross-attention transformer (CAT) mechanism during feature extraction. Transfer learning is applied to gain from a source emotional speech dataset to the target corpus for emotion recognition. We use IEMOCAP as the source dataset to train the source model and evaluate the proposed method on seven datasets in five languages (e.g., English, German, Spanish, Italian, and Chinese). We show that, by fine-tuning the source model with a small portion of speech from the target datasets, HuMP-CAT achieves an average accuracy of 78.75% across the seven datasets, with notable performance of 88.69% on EMODB (German language) and 79.48% on EMOVO (Italian language). Our extensive evaluation demonstrates that HuMP-CAT outperforms existing methods across multiple target languages.Keywords:
Feature (linguistics)
Keyhole
Energy density
Selective Laser Melting
Penetration (warfare)
Cite
Citations (32)
Abstract A new emotion recognition system based on speech is constructed to improve the ability of recognizing negative emotions. Multi-dimensional acoustic characteristics were tested and among them, short-term energy and Mel-frequency cepstral coefficients (MFCC) were selected to be used as parameters for recognition. The system consists two modes: single recognition and group recognition. Single recognition adopts BP neural network model based on MFCC, while group recognition adds support vector machine model based on short-term energy on the basis of single recognition which the group recognition rate of 20 speech can reach 97%. With the increase of the number of speech in each group, the recognition accuracy of negative emotion tends to 100%.
Mel-frequency cepstrum
Negative emotion
Cite
Citations (0)
A vast body of fusion data has been analyzed for different projectiles and target nuclei. It is indicated that the sub-barrier fusion depends on the fusion Q-value. In terms of a recently introduced fusion Q-value rule and an energy scaling reduction procedure, the experimental fusion excitation functions are reduced and compared with each other. It is found that the reduced fusion excitations of selected fusion systems show a similar trend. The fusion data for massive nuclei are in agreement with the Q-value rule. In the fusion process, the Q contribution should be considered. Within this approach, the sub-barrier fusion cross sections of most fusion systems can be predicted without involving any structure effects of colliding nuclei. Instances of disagreement are presented in a few fusion systems. The use of the energy scaling as a criterion of possible experimental data inconsistency is discussed. More precise experimental fusion data need to be measured.
Q value
Value (mathematics)
Cite
Citations (3)
Observed feature P3-Pl corresponds to model feature BOTTOM-SIDE In the model Observed feature P6 corresponds to model feature PZ In the model Observed feature P3 corresponds to model feature P6 Observed feature P5 corresp-onds to model feature P3 In the model
Feature (linguistics)
Feature model
Cite
Citations (0)
Abstract Fusion cross sections were measured for the systems 40 Ar+ 144 Sm, 40 Ar+ 148 Sm and 40 Ar+ 154 Sm above and under Coulomb barrier to understand the role of coupled channels effects involved in barriers fusion. The fusion barriers distributions and fusion probabilities were analyzed using the semiclassical mechanical code which called Sequential Complete Fusion (SCF) as well as Full Coupled Channel code (CCFULL). These calculations show that the observed fusion cross sections fusion barrier distribution and fusion probabilities for these systems are reproduced clearly in the semiclassical mechanical for all excitation states above and under Coulomb barrier.
Semiclassical physics
Cite
Citations (2)
Many researches for detecting feature interactions of telephony services have been discussed all over the world. But, there is no guarantee that all detected feature interactions can be resolved. In fact, services must be provided with unresolved feature interactions still remained. Consequently, a method for avoiding feature interactions at run-time is required. However, as far as the authors know, any effective solutions for avoiding feature interactions at run-time have not yet been proposed. In this article, it is clarified that even for services between which feature interactions occur, in some terminal assignments, feature interactions do not occur. Based on this fact, a method for avoiding feature interactions is proposed, where service initiation is prevented only when a terminal assignment causes feature interactions.
Feature (linguistics)
Feature model
Cite
Citations (2)
Many researches for detecting feature interactions of telephony services have been discussed all over the world. But, there is no guarantee that all detected feature interactions can be resolved. In fact, services must be provided with unresolved feature interactions still remained. Consequently, a method for avoiding feature interactions at run-time is required. However, as far as the authors know, any effective solutions for avoiding feature interactions at run-time have not yet been proposed. In this article, it is clarified that even for services between which feature interactions occur, in some terminal assignments, feature interactions do not occur. Based on this fact, a method for avoiding feature interactions is proposed, where service initiation is prevented only when a terminal assignment causes feature interactions.
Feature (linguistics)
Feature model
Cite
Citations (3)
We give an overall picture of our present understanding of the effect of the break-up of stable weakly bound nuclei on their fusion cross section with light, medium and heavy mass nuclei, at energies above the Coulomb barrier. The discussion is based mostly on recent data obtained by our group in collaborative experiments with ANU, USP and Tandar Laboratories. We conclude that there is complete fusion suppression for heavy targets, due to the loss of flux, corresponding to the occurrence of incomplete fusion of one of the break-up fragments. For medium and light mass targets, the incomplete fusion is negligible and therefore is no complete fusion suppression.
Section (typography)
Cite
Citations (8)
Speaker diarisation
Cite
Citations (32)
Complete fusion excitation functions for 9 Be + 208 Pb have been measured to high precision at near barrier energies. The experimental fusion barrier distribution extracted from these data allows reliable prediction of the expected complete fusion cross-sections. However, the measured cross-sections are only 68% of those predicted. The large cross-sections observed for incomplete fusion products support the interpretation that this suppression of fusion is caused by 9 Be breaking up into charged fragments before reaching the fusion
Cite
Citations (1)