This study proposes a deep learning model to identify and correct English speech distortion caused by factors like background noise, channel distortion, and compression distortion. The model, which integrates a convolutional neural network (CNN) with a long short-term memory network (LSTM), performs better in identifying distortions and repairing signals. Experimental results show the model's detection accuracy exceeded 92% on multiple public speech datasets, and the signal-to-noise ratio (SNR) improved by about 8 dB after signal repair, verifying the method's effectiveness under different distortion types. Comparison with traditional autoencoders and WaveNet methods shows the proposed model achieves significant improvements in both detection accuracy and signal quality repair. The findings offer a new solution for addressing distortion issues in speech signal processing, with potential applications in high-quality speech communication, speech recognition, and speech enhancement.