Recently,I study the query-by-example spoken term detection approach based on the phonetic posteriorgram.
reference the paper "Query-by-example spoken term detection using phonetic posteriorgram templates"
But what confused me is the posteriorgram feature,while the paper just said the posteriorgrams feature came from the lattice.
Obviously,the lattice was generated by the phonrec/HVite system.
I tried the forward-backward algorithm，but I have no idea how to do next.