Intelligent Assessment Method of Communication Interference Speech Quality Based on End-to-end Network
摘要
Speech quality can reflect the interference in the environment during speech communications. This paper focuses on evaluating speech quality in communication interference environments, and introduces an innovative end-to-end network-based intelligent evaluation method. Utilizing a transformer network structure, the method involves segmenting interference speech into time frames, extracting Mel and amplitude spectrograms, and constructing feature maps for deep feature extraction and quality assessment. Tested on a communication interference speech dataset, this end-to-end approach achieved a remarkable 93% accuracy in evaluating interference speech quality, outperforming CNN-based methods by 5.5%. This significantly enhances the precision of assessing interference speech quality.