Audio signal processing has always been plagued with the problem of acoustic howling, especially when it comes to real-time and open microphone applications. In this paper, we present HiGT-Fuse, a novel Hierarchical Graph-Transformer Fusion based howling detection and suppression with self-supervised learning to achieve robust and real-time howling detection and suppression. We develop graphs-multi-resolution ones-on time-frequency spectrograms and these are reflections of fine and coarse-grained spectral dynamics. In order to solve the limitations of the fixed graph structures, we propose a meta-learned edge function that is able to dynamically modify the graph structure. Individual graphs are encoded via a deep Graph Neural Network (GNN) which is enhanced with a Transformer encoder to enable long range temporal dependencies. The learning strategy is a contrastive curriculum, which combines incremental augmentation and dual-loss training, and enables the model to learn the abnormal patterns without the need of labeled data. Learned embeddings are used to calculate anomaly scores and a dynamically configurable thresholding module, which uses a Gated Recurrent Unit (GRU), is used to identify howling events. We design a two-stream GRU controller which, in real time, parameterizes a digital notch filter which is used to reject the detected anomalies. Learned embeddings are used to compute the anomaly scores and a real-time thresholding module based on GRU is used to localize events. A two-stream GRU controller is then used to adapt a digital notch filter to suppress the detected howling. The average F1-score, Perceptual Evaluation of Speech Quality (PESQ) improvement, Signal-to-Distortion Ratio (SDR) and Real-Time Factor (RTF) of HiGT-Fuse are 0.91, 3.03, 3.03 dB and 0.37x respectively, which means that the system is approximately 2.7 times faster than real time; also, it outperforms the traditional baselines, including AFC, and recent deep learning models, including Temporal-Spectral Transformer (TST) and Bootstrap Your Own Latent (BY The system is highly generalizable to both real-world and synthetic data sets including LibriSpeech, GWA, VCTK, REVERB and VoiceBank-DEMAND, in both Signal-to-Noise Ratio (SNR) and reverberation settings. HiGT-Fuse is a novel standard in autonomous and low-latency audio stabilization and can be widely applied to anomaly detection in the spectral domain.

Pulse repetition interval modulation (PRIM) recognition is a critical task in electronic intelligence (ELINT) and electronic support measure (ESM) systems for detecting radar threats accurately. However, PRI recognition is a complex issue due to missing and spurious pulses, resulting in noisy PRI pattern changes in real environments. To address this problem, this paper proposes a novel approach that recognizes the five common types of PRIM through a four-phase process. In the first phase, a deep convolutional neural network (DCNN) is used as a feature extractor. Then, extreme learning machines (ELMs) are used for real-time recognition of the PRIM patterns in the second phase. In the third phase, we employ the biogeography-based optimizer (BBO) to enhance the network’s robustness by optimizing the connection weights and biases. To address the increasing complexity of the model, we introduce an optimized variable-length internet protocol-based BBO (VBBO) in the fourth phase. In this approach (i.e., DCNN-VBBO-ELM), each layer of DCNN is encoded by an IP address into a habitat of VBBO in the same sequence as the DCNN layers. To evaluate the proposed method, we develop a real experimental dataset consisting of five common PRI patterns. Our approach achieves a final accuracy of 97.05%, which is better than other ELM-based benchmark models. Moreover, the proposed model requires only 27 s of training time to process 50,000 training images, confirming its real-time capabilities. In conclusion, our proposed approach improves PRI recognition by leveraging DCNN, ELM, and VBBO, resulting in a more accurate and robust real-time radar PRI classifier.

In this paper, a radial basis function neural network (RBF-NN) automatic sonar target recognition system is proposed. For the RBF-NN training phase, a whale optimisation algorithm (WOA) developed with a fuzzy system has been used (which is called FWOA). The reason for using the fuzzy system is the lack of correct identification of the boundary between the two stages of exploration and exploitation. Thus, the tuning of the effective parameters of the WOA is left to the fuzzy system of the Mamdani type. RBF-NN was trained by chimp optimisation algorithm (ChOA), genetic algorithm (GA), Evolution Strategy (ES), league championship algorithm (LCA), grey wolf algorithms (GWO), gravitational search algorithm (GSA), and WOA to compare the proposed algorithm. The measured criteria are convergence speed, ability to avoid local optimisation, and classification rate. The simulation results showed that FWOA with 97.49% classification accuracy rate in sonar data performed better than the other seven benchmark algorithms.