Assamese is an important language with its own uniqueness concentrated primarily among a population of around 30 million in the North Eastern part of India. The work focuses on designing an optimal feature extraction block and a few ANN architectures so that the performance of the Speech Recognition System can be improved. The key part of the work is related to certain signal processing operations including adaptive filtering for designing of a set of speech corpus of Assamese numerals recorded with gender and mood variation. The work carried out with multiple ANN based architectures provides important insights to the development of language specific speech recognition tools. Experiments work carried out in this connection is reported here in this work so as to formulate a rudimentary platform for speech corpus generation using adaptive LMS filters and LPC cepstrum, as a part of an ANN based Speech Recognition System exclusively designed to recognize numerals of Assamese.