We study the problem of privacy amplification with an active adversary in the information theoretic setting. In this setting, two parties Alice and Bob start out with a shared $n$-bit weak random string $W$, and try to agree on a secret random key $R$ over a public channel fully controlled by an active and unbounded adversary. Typical assumptions are that these two parties have access to local private uniform random bits. In this paper we seek to minimize the requirements on the local randomness used by the two parties. We make two improvements over previous results. First, we reduce the number of random bits needed for each party to $Θ(\ell+\log n)$, where $\ell$ is the security parameter, as long as $W$ has min-entropy $n^{Ω(1)}$. Previously, the best known result needs to use $Θ((\ell+\log n)\log n)$ bits. Our result is also asymptotically optimal. Second, we generalize the problem to the case where the two parties only have local weak random sources instead of truly uniform random bits. We show that when each party has a local weak random source with min-entropy $> n/2$, there is an efficient privacy amplification protocol that works nearly as good as if the two parties have access to local uniform random bits. Next, in the case where each party only has a weak random source with arbitrarily linear min-entropy, we give an efficient privacy amplification protocol where we can achieve security parameter up to $Ω(\log k)$. Our results give the first protocols that achieve privacy amplification when each party only has access to a local weak random source.