In our previous works, a maximum likelihood training approach was developed based on the concept of stochastic vector mapping (SVM) that performs a frame-dependent bias removal to compensate for environmental variabilities in both training and recognition stages. Its effectiveness was confirmed by evaluation experiments on Aurora2 and Aurora3 databases. In this paper, we present an extended ML formulation to entertain some new SVM functions that are piecewise linear transformations and are more flexible than the frame-dependent bias removal. Evaluation results on Finnish Aurora3 database show that in comparison with the performance of a baseline system based on ML-trained CDHMMs without feature compensation, the previous and the new SVM-based feature compensation approaches achieve a relative word error rate reduction of 15.7% and 26.1% respectively for well-matched condition.