Fast random/mutation algorithms (vector to vector)

I have been trying to create some very detailed procedural terrain recently and this has lead me to want to make my own custom noise functions. To do this at the centre I need a reliable random number generator and I was curious about what the most specialised algorithms for my use case is and how I can implement a version of it.

Essentially all I want is an algorithm that can take in a given string of data, of a predetermined size, say a vector, and output a different, random seeming string of data, again of a predetermined variant size.

The algorithm should be seedable and deterministic, and it's output shouldn't decay or produce any kind of bias or obvious looking pattern. basically it needs to be fairly robust when it comes to producing neat results for noise.

beyond this, my priority is speed, the fastest algorithm that can achieve these goals is the ideal. if you know of anything that can do this please let me know :)