Aiming at the shortcomings of EEG emotion recognition models in feature representation granularity and spatiotemporal dependence modeling, a multimodal emotion recognition model integrating multi-scale feature representation and attention mechanism is proposed. The model consists of a feature extraction module, feature fusion module, and classification module. The feature extraction module includes a multi-stream network module for extracting shallow EEG features and a dual-scale attention module for extracting shallow EOG features. The multi-scale and multi-granularity feature fusion improves the richness and discriminability of multimodal feature representation. Experimental results on two datasets show that the proposed model outperforms the existing model.

Speech signals are nonlinear chaotic time series. This paper proposes a novel speech signal nonlinear prediction model with the hidden phase space reconstruction method. The parameters, embedding dimension m, time delay τ and model structure are solved simultaneously, breaking the restriction of phase space, which needs to be reconstructed before modeling for the existing prediction method. Subsequently, an explicit speech signal prediction model is generated. Meanwhile, the introduction of the frame length parameter k effiectively extends the prediction length. Experimental results show that the values of m and τ solved by the proposed method are consistent with the values addressed by the Cao method and mutual information method, respectively. In addition, the optimal value of k is further discussed. The prediction results obtained using the proposed model are more accurate than those of linear prediction coding, the radial basis function neural network model and the long short-term memory network.

Music emotion recognition is becoming an important research direction due to its great significance for music information retrieval, music recommendation, and so on. In the task of music emotion recognition, the key to achieving accurate emotion recognition lies in how to extract the affect-salient features fully. In this paper, we propose an end-to-end spatial-temporal feature extraction method named FFA-BiGRU for music emotion classification. Taking the log Mel-spectrogram of music audio as the input, this method employs an attention-based convolutional residual module named FFA, which serves as a spatial feature learning module to obtain multi-scale spatial features. In the FFA module, three group architecture blocks extract multi-level spatial features, each of which consists of a stack of multiple channel-spatial attention-based residual blocks. Then, the output features from FFA are fed into the bidirectional gated recurrent units (BiGRU) module to capture the temporal features of music further. In order to make full use of the extracted spatial and temporal features, the output feature maps of FFA and those of the BiGRU are concatenated in the channel dimension. Finally, the concatenated features are passed through fully connected layers to predict the emotion classification results. The experimental results of the EMOPIA dataset show that the proposed model achieves better classification accuracy than the existing baselines. Meanwhile, the ablation experiments also demonstrate the effectiveness of each part of the proposed method.

Music generation by AI algorithms like Transformer is currently a research hotspot. Existing methods often suffer from issues related to coherence and high computational costs. To address these problems, we propose a novel Transformer-based model that incorporates a gate recurrent unit with root mean square norm restriction (TARREAN). This model improves the temporal coherence of music by utilizing the gate recurrent unit (GRU), which enhances the model’s ability to capture the dependencies between sequential elements. Additionally, we apply masked multi-head attention to prevent the model from accessing future information during training, preserving the causal structure of music sequences. To reduce computational overhead, we introduce root mean square layer normalization (RMS Norm), which smooths gradients and simplifies the calculations, thereby improving training efficiency. The music sequences are encoded using a compound word method, converting them into discrete symbol-event combinations for input into the TARREAN model. The proposed method effectively mitigates discontinuity issues in generated music and enhances generation quality. We evaluated the model using the Essen Associative Code and Folk Song Database, which contains 20,000 folk melodies from Germany, Poland, and China. The results show that our model produces music that is more aligned with human preferences, as indicated by subjective evaluation scores. The TARREAN model achieved a satisfaction score of 4.34, significantly higher than the 3.79 score of the Transformer-XL + REMI model. Objective evaluation also demonstrated a 15% improvement in temporal coherence compared to traditional methods. Both objective and subjective experimental results demonstrate that TARREAN can significantly improve generation coherence and reduce computational costs.