Recording of the speech samples is the first step in speech recognition and related tasks. For English, there are a bunch of readily available data sets. But standard data sets with regional dialect and mood variations are not available and the need to create our own data set for our experimental works has been faced. We have considered Assamese language for our case study and since it is less computationally aware, there is a need to develop the speech corpus having dialect and mood variations. Also, the development of corpus is an ongoing process and the initial task is reported in this paper.