It seems to me that the audio space is still quite undeveloped when it comes to Machine Learning. I mean we have visual style transfer (look at the Deep Video

It seems to me that the audio space is still quite undeveloped when it comes to Machine Learning. I mean we have visual style transfer (look at the Deep Video Portraits shown at SIGGRAPH, it's insane..) yet the equivalent in speech isn't even usable. I think largely it is just a harder problem as you can break a video into frames and each frame can be treated independently, of course you may wish to take the sequential nature into account but there is a lot of information in each static frame. Contrast this with audio where a very small clip (analogous to a video frame) is insufficient to get even phonemes let alone words, or to identify if the speaker is male/female or if there is even any speech at all... That said - I think it is the area with the most potential and the one which interests me the most.