Abstract: Acoustic signals emitted by large yellow croaker during various stages of breeding period can generally reflect its physiological and behavioral states．However, acoustic signals collected in actual aquaculture environment are often mixed with a variety of noises, so noise reduction pre-processing is required to be performed on raw signals．An improved wavelet threshold-CEEMDAN noise reduction algorithm was proposed, in which original signal was first decomposed into multiple IMFs, then each IMF was processed using improved wavelet threshold function, and finally processed IMFs were reconstructed．Results showed that SNR of detection system was increased to 14.53 dB and the RMSE was reduced to 0.00196 dB after using noise reduction algorithm proposed．Compared with traditional noise reduction algorithms, improved algorithm has a better noise reduction effect, which was more conducive to analyzing and studying vocal behaviors during breeding period of large yellow croaker．