9th Annual Conference of the International Speech Communication Association

Brisbane, Australia
September 22-26, 2008

Utterance-Level Normalization for Relative Articulation Rate Analysis

Tuomo Saarni, Jussi Hakokari, Jouni Isoaho, Tapio Salakoski

University of Turku, Finland

This study describes a computational method for studying variation in articulation rate in a qualitatively mixed speech corpus. The method works within the scope of individual utterances, replacing each single speech sound's time information with a coefficient based on its duration relative to its environment. It can be used to generalize and determine points of acceleration and deceleration in articulation at the phone level, even when the general speaking rate varies greatly due to speaker, style, and utterance length related effects. To demonstrate the usability of the proposed method, we track observed deceleration of articulation rate (a form of final lengthening) towards the ends of utterances in a linguistically uncontrolled Finnish-language speech corpus with several speakers and styles.

Full Paper

Bibliographic reference.  Saarni, Tuomo / Hakokari, Jussi / Isoaho, Jouni / Salakoski, Tapio (2008): "Utterance-level normalization for relative articulation rate analysis", In INTERSPEECH-2008, 538-541.