We are currently engaged in the development of speech recognition systems for continuous speech in Indian languages. In our opinion the most important block in the system is the phonetic recogniser. We propose to exploit the phonetic nature of Indian languages to design the phonetic recogniser. We consider characters of an Indian language as symbols for the design of speech signal-to-symbol transformation module of our system. In this paper we discuss the limitations of conventional techniques in speech signal-to-symbol transformation and show the importance of knowledge-based signal processing.