Quantifying Language Variation Acoustically with Few Resources