Note that I don't think I'll be integrating this into my compression protocol for at least another year, but it's a very exciting development that I think will only improve with time...
What's always been important about voice recordings isn't that it's an accurate recording of the person speaking, but that what's being said is understandable. Most compression tools handle this through subtraction: taking out everything that's not within the rather thin part of the sound spectrum scope occupied by the human voice. For example, a high-pitched noise in a voice recording might not be part of the final output.
Now, a new system has come about that records sound in an exceedingly basic (and small) format and then uses an intelligent algorithm to "improve" it. This combination has given rise to a codec that could fit a whole podcast on a Floppy Disk:
Codec2: https://auphonic.com/blog/2018/06/01/co ... oppy-disk/
The technology is a neat idea on it's own, but the provided link is very well written and describes the tech with live samples. This could have huge applications for bandwidth, audio archives, poor network connections, and the use of technology in the 3rd world by a non-literate populace.
More info: http://www.rowetel.com/?p=5966