Chicken voice recognition method, system and equipment based on improved neural network model and medium

The invention discloses a chicken voice recognition method, system and device based on an improved neural network model, and a medium. The method comprises the following steps: acquiring voice data of chickens in a coop; noise reduction is performed on the input sound data by using a spectral subtraction method in which a smoothing mechanism is introduced; filtering low-frequency noise signals in the noise-reduced sound data; intercepting sound segments in the filtered data by adopting a dual-threshold endpoint detection algorithm; extracting sound features in the sound segments by adopting a plurality of feature extraction algorithms, and combining the extracted sound features; and inputting the combined sound features into a bidirectional long-short memory network-attention mechanism model for identification to obtain a sound classification result. According to the method, the voice of the chicken in the chicken coop can be recognized, the voice classification result is obtained, the accuracy and convenienc