Here’s a Git project from Deep Speech by @Tilman_Kamp
It seems to achieve the goal, but also seems too obvious an answer for others not to comment?
Also being discussed on forum here
Either way, great thought 
(and if I’m way off, kindly let me know 
)