Automatic clinical speech recognition for CLEF 2015 eHealth challenge

In this working notes report/paper, we describe the details of two submis-sions for CLEF 2015 eHealth challenge for Task 1a, with details of methods and tools developed for automatic speech recognition of NICTA synthetic nursing handover dataset. The first method involves a novel zero-resource approach based on unsuper-vised acoustic only modeling of speech involving word discovery, and the second method is based on combination of acoustic, language, grammar and dictionary models, using well known open source speech recognition toolkit from CMU, the CMU Sphinx[7]. The experimental evaluation of the two methods was done on Challenge dataset (NICTA synthetic nursing handover dataset).