Transiems as dynamically defined, sub-phonemic speech units are constructed via a feature-assimilation process involving phonologically motivated major articulatory feature(s). A computational model representing transiems and the associated statistical estimation theory enable speech recognizers to directly employ transiems as the speech primitive in place of the conventional phonemic representation of the lexicon. The main advantages of the transiems for potential use in speech recognition are that they naturally match the intrinsically dynamic pattern of speech and that co-articulation effects spanning several phonetic segments are automatically embedded in the construction of transiems without use of acoustic data of speech.
Deng, L. (1996). Transiems as dynamically defined, sub-phonemic units of speech: A computational model. Signal Processing, 49(1), 25–35. https://doi.org/10.1016/0165-1684(96)83106-9