1 ADAPT Algorithm for Dynamic Alignment of Phonetic Transcriptions Bram Elffers, Christophe Van Bael, Helmer Strik bramelf@lands.let.ru.nl c.v.bael@let.ru.nl h.strik@let.ru.nl Department of Language and Speech Radboud University Nijmegen The Netherlands 1. Introduction ADAPT (Algorithm for Dynamic Alignment of Phonetic Transcriptions) is a dynamic programming algorithm that computes the optimal alignment between two strings of phonetic symbols. The optimal alignment is computed from a matrix in which the distances between phonetic symbols are defined in terms of articulatory features. The optimal alignment yields a distance measure, expressing the phonetic similarity of the transcriptions. This document discusses the purpose of using ADAPT, provides a technical description of its alignment algorithm, and documents its usage. 2. Purpose ADAPT can be used to determine the phonetic similarity of two transcriptions of a spoken utterance. These transcriptions, provided as strings of phonetic symbols, typically represent a reference transcription (describing the optimal pronunciation of the utterance) and a hypothesis transcription (describing an actual pronunciation of the utterance). A smaller distance between the hypothesis transcription and its reference transcription indicates a greater phonetic similarity, usually interpreted as a higher quality assessment of the hypothesis transcription. Given two strings of phonetic symbols, ADAPT produces two types of output. Firstly, it produces the optimal alignment between the symbols in the two transcriptions. Secondly, it produces two statistic measures given this optimal alignment: the number of substitutions, deletions, and insertions at the phone level, and the overall phonetic distance between the reference and hypothesis transcriptions.