Abstract: A spectrogram-based reconstruction method is proposed to address the problem of poor voice transmission and reconstruction quality under conditions of high compression ratios and low signal-to-noise ratios. Speech signals are converted into spectrograms at the transmitter, which are later transmitted and denoised at the receiver. Then, the amplitude spectrum is restored from the denoised spectrogram image and the voice is reconstructed through the amplitude spectrum by the voice model. Experiments show that the perceptual evaluation of speech quality (PESQ) of the reconstructed speech exceeds 3 under noise-free environment with compression ratios of 10 and 40 respectively. The PESQ can also exceed 2 under the low signal-to-noise ratio with compression ratio of 10. The proposed method shows significant improvement in reconstructed speech quality at high compression ratios compared with the traditional algorithm.

Key words: speechtransmissionandreconstruction, imageenhancement, vocalreconstructionmodel, compressionratioandlowsignal-to-noiseratio