Develop wireless audio recording backpack

Over the last couple of months, I have established collaborations with engineers to develop wireless audio recording devices that can be used to study vocal behaviors and social communication of budgerigar parrots in a large natural group.

I have made several significant progresses:

1. One major challenge of audio recording in a large group of animals is that vocalizations from other individuals will contaminate recording of the focal individual. Such cross talking will render the audio data unusable. One solution is to use bone conduction microphones that relies on head/body vibration when the bird is vocalizing. I have thus tested many models of bone conduction microphones and found the model that work reliably. Pilot recording experiment showed that I can obtain high-quality audio data from the focal bird even when other birds in the same room are vocalizing loudly.

2. I have developed a mini-backpack system that can be worn by budgerigar parrots to carry the wireless recording electronics. Behavioral observation suggested that the bird is able to vocalize and socialize normally when wearing the mini-backpack. Next step on this thread is to integrate the bone conduction microphone to this mini-backpack system.

3. With my collaborators, we have brainstormed a design of the recording chip that controls the audio recording and handles data storage. We estimate that a newly charged button battery can support continuous recording of ~4 hours. Next step is to implement the design, test its functionality, and troubleshoot any issues.

4. I also made progress on the data analysis side. The vocalization of budgerigar parrot is highly complex, rending traditional analysis methods unsuitable. I have therefore developed deep learning-based methods to analyze the budgerigar vocalization. So far, the variational auto-encoder (VAE) method is giving promising results -- I am able to identify the stereotyped acoustic structure in the highly complex budgerigar song.