It's a nice study with some interesting findings, but not as hype as that youtuber brings it (I get it he's running a channel). Plants emitting and reacting to sounds has been known for quite a while, but if I understand it correctly, they found non-stressed and stressed plants to "emit" sounds less frequent and more frequent respectively. Not that they make actually different sounds.
The implications of their machine learning also seem to be a bit limited right now. They obviously have great performance when comparing an empty greenhouse vs. isolated plants as they are very different. Differentiating between irrigated/non-irrigated plants is nice, but takes an hour (with their current methods). The condition-specific performance is poor, with most sub-70%, so it's not confident in telling you what exactly it's looking at. As most crops and greenhouse plants are being maintained with very strict schedules, these 'dry' and 'cut' conditions are not that interesting. Also, you have to weigh the investment of equipment and software to do this versus doing a quick manual check (if the area is not too large). However, if they can improve it towards pathological conditions which they describe in the discussion (and like the industry are currently training imaging methods), they got something really nice. Still a nice proof-of-concept though.