Extracting semantically-coherent keyphrases from speech

Diana Inkpen, Alain Désilets


A method for extracting keyphrase from spoken audio is discussed. The approach is based on the extractor system developed for text by Turney. The extractor uses a supervised learning approach to maximize the overlap between machine extracted and human extracted keyphrases. Each key word is detected with low similarity to its closest semantic neighbor, using an algorithm.


Algorithms; Audio systems; Broadcasting; Error analysis; Semantics; Speech; Pointwise mutual information (PMI); Semantic coherence; Speech extractor systems; Word error rates (WER)

Full Text:



  • There are currently no refbacks.