In recent works, the use of phone class-conditional posterior probabilities (posterior features) directly as features provided successful results in template-based ASR systems. Moreover, it has been shown that these features tend to be sparse and orthogonal. Given such properties, new types of ASR may be investigated. In this work, we investigate the use of Self-Organizing Maps to transform sequences of posterior feature vectors representing words into sparse fixed-size patterns. We evaluate the ability of these patterns to discriminate between words in the context of template-based ASR using a simple histogram matching technique (i.e. without the use of dynamic programming). We present experiments on 75-word speaker- and task-independent isolated word recognition task.