Browse Prior Art Database

Hash-Code Acoustic Fast Match

IP.com Disclosure Number: IPCOM000058422D
Original Publication Date: 1988-Oct-01
Included in the Prior Art Database: 2005-Feb-15

Publishing Venue

IBM

Related People

Authors:
Merialdo, B [+details]

Abstract

This article describes a new method to select from an acoustic observation (utterance of a word), a list of possible words from a dictionary. The method is based on a hash-code formulation, and is characterized by a very low computation cost. An "Acoustic Fast Match" is a procedure that, given an acoustic observation A, produces with little computation a list of word candidates w that may match this observation (i.e., whose pronunciation may be A). The Fast Match is used to reduce the search in a large vocabulary to a smaller subset. Generally, a more precise (and expensive) procedure is then applied to the subset. Let A be a sequence of acoustic labels, A = l1, l2,...