Is this really necessary ?
I'll refer you to an answer I gave in another thread on this topic:
Maybe. Maybe not. The numbers given so far in this thread discuss the Shannon entropy, but have you calculated the min-entropy you would achieve from doing this? What randomness extractor algorithm are you planning to use to turn those dice rolls in to usable entropy? How are you converting those dice rolls to binary without introducing modulo bias? It's not as simple as just "roll the dice more" - it's a very complex topic which most people do not fully understand (and I do not profess to either), which is why whenever the topic of manually generating entropy comes up, I always suggest von Neumann's coin flips to simply, quickly, and most importantly verifiably generate 128 or 256 bits of provably unbiased entropy.
If the answer to generating true random numbers was as simple as "Take any old non-random and biased process and just repeat it a bunch of times", there would not be an entire field of research dedicated to it.
We have methods were are provable and verifiable. Why risk everything by coming up with your own
ad hoc scheme?
I would use casino grade dices and test it for its bias before using it. To convert it to binary without bias i would use this model : number1 : 1 number2: 0 number3: 00 number4: 01 number5: 10 number6: 11.
on an old offline pc to get to 24 words and then burn the pc.