Auditory decoding of temporal features is a complex process in which vocalizations such as birdsongs are segmented into distinct auditory units, a crucial step in transforming sound into meaning. The songbird auditory system offers a valuable model for studying the neural representation of complex sounds. The hierarchical structure of the Zebra Finches' song makes it ideal for in-depth analysis of auditory decoding. We use stacked Bidirectional Long Short-Term Memory (BiLSTM) deep neural networks to decode the amplitude envelope and the time-locked envelope features of zebra finch songs. To assess the neural activity's efficacy at segmenting continuous songs into units and decoding amplitude, the network was trained with local field potential (LFP) and multi-unit activity envelope (MUAe). In ensemble responses, both the amplitude envelope and time-locked features could be accurately decoded using LFP and MUA. The performance of LFP and MUAe was very similar, but MUAe gave slightly better results for envelope decoding. It was observed that temporal information might not have been present everywhere in the brain/auditory pallium, and this segmentation function could be modulated by other factors (such as attention). Notably, the envelopes of the introductory notes and the first motif were significantly better decoded than the second motif. This result suggested that these specific parts of songs were receiving more attention. Additionally, network accuracy and inter-trial phase coherence exhibited a positive linear relationship in LFP and MUA signals, indicating the importance of neural synchrony. High-performance decoding of temporal features has shown how neural representations of these features facilitate or reflect the segmentation of songs. It provides valuable insights for future research into the intricate processes involved in vocal communication.