Next we looked at vocalizations w/ 6,000 hrs of RFID box audio (thanks @openacousticdevices.info!) + ML to find vocal events

Takeaways:
1. Vocalization is seasonal: most in spring/summer when social groups are small
2. Wild mice use both audible squeaks & ultrasonic calls, often close together

4/8

Finally, we used a wireless ultrasonic mic to record vocalizations. We observed an amazing diversity of calls which have acoustic features that lay outside of the standard distribution reported in the lab. Also, we see interesting usage of the classically negative affect 22 kHz call during foraging.

It's hard to interpret observations like this because in the field we cannot control environmental factors that might be causing movement. Instead, we built 3D reconstructions of the environment using gaussian splatting with the goal of building causal models that can link environment + behavior.

We noticed periods of coordinated collective movement where some, but not all, animals appeared to participate. We implemented multi-animal tracking and a size-estimation model to show that larger rats move faster and are more likely to participate in collective bursts than smaller animals.