Jie Zhao and Yufeng Wu
Institute for Signal and Information Processing
Mississippi State University
Phone/Fax: 601-325-8335/3149 Email: {zhao, wu}


In the past few months, many new features have been integrated into the ISIP public domain speech recognition system decoder.

The newest version ISIP decode is ISIP decoder v3.3. These new features include:

o N-Gram decoding, which can read the bigram and trigram language model and output the most likely hypothesis.

o Lattice generation, which can read the bigram and trigram language model and generate an N-best list of hypotheses. The N-best list is then converted into a lattice (word graph) format.

o HMM training, which can estimate the parameters of the HMMs from examples of the data sequences that they are intended to model.

o Grammar Compiler, which can read a regular expression grammar, and parse it into a lattice format.

o Extract_feature, which can transform a speech waveform into a sequence of parameter vectors.

In this talk, we will describe the new version ISIP decoder to be delivered and the way these new features were efficiently implemented in the ISIP recognition system.