Abstract: We explore the properties of the pronunciation variation (PV) models as an approach for an automatic speech recognition accuracy improvement. The PV model is formulated as well as the methods to find out PV parameters and include the model into the search procedures. We show that utilizing of the PV models could substantionally increase the accuracy of automatic recognition of natural speech.

Keywords: automatic speech recognition, acoustic modeling, speech pronunciation variation modeling, explicit models, hidden markov models.

ACM Classification Keywords: I.2.7 Natural Language Processing - Speech recognition and synthesis

Link:

IMPROVING AUTOMATIC SPEECH RECOGNITION ACCURACY BY MEANS OF PRONUNCIATION VARIATION MODELING

Vladimir Chuchupal, Anton Korenchikov

http://www.foibg.com/ijicp/vol01/ijicp01-04-p08.pdf