A Frogman Speech Enhancement Algorithm with Pre-Enhanced DNN Joint Model
Xi Yan, YanFeng Xu, B.S. Liu · 2023
The paper proposes a joint speech enhancement algorithm that can significantly improve the intelligibility of underwater communication. Due to the influence of masks worn during underwater operations and ambient noise, the voice quality deteriorates, hindering normal communication. To address this issue, firstly, the paper designs an integrated speech enhancer that combines ten statistical-based speech enhancement algorithms which employs the optimal spectral amplitude estimation method to preprocess the noisy speech and reduce the noise amplitude. Secondly, the pre-enhanced signal is fed into the DNN model. The model uses multiple noise types for multi-condition training, and the experimental results show that the model has good generalization ability and can cope with unknown noise environments. Finally, the algorithm is applied to the complex unknown noise of underwater frogman communication. Experimental results prove that the algorithm is feasible and effective, and PESQ has been significantly improved.