The invention relates to the field of sound recognition and discloses a construction method and device of a bird sound recognition model, a computer device and a storage medium. The construction method comprises: acquiring bird sound samples that include sound fragments, wherein each bird sound sample corresponds to one bird name tag; extracting Mel-frequency cepstrum features from the sound fragments to obtain multiple spectrums of the sound fragments; subjecting the spectrums to feature extraction and dimensionality reduction to obtain feature graphs corresponding to the sound fragments; subjecting the feature graphs to voiceprint feature encoding to obtain voiceprint feature vectors; inputting the voiceprint feature vectors into a capsule network for training, and training to acquire abird sound recognition model that includes the trained capsule network. The bird sound recognition model acquired herein is suitable for processing sound fragments containing bird sound and recognizing types of birds.