The goal is to identify speech start/end times (if present at all) in a potentially noisy short recording. If time permits, further goals include separating words and identifying them, at least for a few basic words, which will require additional datasets and models. These techniques are widely used in voice-activated consumer electronics.
Also, audio analysis techniques in machine learning are basically image analysis techniques in disguise, so if you have experience in that or are willing to learn, you're welcome to join!
* Goals subject to change depending on how things go for the first week or so, though the final result will still be in speech analysis.