This paper proposes an innovative solution for the conservation and understanding of avian biodiversity in forest ecosystems in Ecuador using deep learning techniques. Preprocessing techniques, including noise signal separation, were applied, and convolutional neural network models were used to extract significant features from Mel spectrograms using data from the forest “La Prosperina”. To train the model, an average of 400 samples of bird songs were utilized for each of the six species to be classified, achieving accuracy rates of 99.93% for training dataset and 96.27% for testing dataset, demonstrating exceptional abilities to identify threatened bird species, with positive results represented in a confusion matrix. This approach not only benefits conservationists by providing them with effective and economically viable tools for monitoring bird species and behaviors but also has broader applications.