Criteria for database and tool design for speech timing analysis with special reference to mandarin

This position paper investigates some of the problems in modelling speech timing for the design of speech databases and corpus analysis tools for phonetics and speech technology. First we examine a selection of phonetic approaches to speech timing analysis, the so-called ‘rhythm metrics’, and focus on explaining (1) inconsistencies (varying results for the same language) and (2) the failure to model rhythmic alternation. To overcome these problems we present a new perspective on the phonetic identification of rhythm patterns as a special case of duration modelling, including the additional criterion of alternation. We describe the Rhythm Parser, a tool for identifying hierarchical alternating patterns, and discuss results from applying it.