Research on speech characteristics based on compressed sensing theory

Due to fewer researches and applications of speech compressed sensing, the sparsity of speech signals was studied first. Then the Fourier orthogonal transform method and the Orthogonal Matching Pursuit (OMP) algorithm were used to compress and reconstruct speech signals. The relationship between the signal reconstruction and its performance, such as the speech signal compression ratio, the periodicity of reconstructed signals, and the frame size etc. is also discussed here. Experiments have shown that: (1) voice signal is sparse and compressible; (2) the speech reconstruction of integral period and regular periodicity signals performs better than that of non integral period and irregular periodicity signals; (3) the best frame size of reconstructed speech is about 10ms.