The paper presents a system for manner-based labelling of the speech signal for isolated words from total energy profile. The manner-based classes selected are sibilant, vowel, nasal murmur, semi vowel and lateral, trill, unvoiced unaspirated interrupt, unvoiced aspirated interrupt, voiced unaspirated interrupt , voiced aspirated interrupt and consonant clusters. The results of labelling for 100 words uttered by one male and one female informants are presented. The relavance of this approach in very large vocabulary phoneme-based spoken word recognition system is discussed.