linguistic approach to pitch range modelling

Abstract

Pitch range is currently characterised in a number of different ways across research disciplines and is often treated as a simple measurement. Pitch range has been defined as the difference between minimum and maximum fO (Cosmides 1983). This data alone conveys no information about the distribution of fO values within that range. Similarly the mean and standard deviation does not adequately capture important differences in the pitch range of different speakers (Ladd et al. 1985). Ladd (1996) describes pitch range using two partially independent dimensions of variation, that of overall level and span. This idea has been further developed by Shriberg et al. (1996), in a study based on a large corpus of Dutch speech. Given this two parameter model, it is possible to predict target fO values for when speakers raise their voices from fO values at corresponding locations in speech produced normally. This thesis reports on three studies of pitch range variation across speakers. The experiments examine the relation between a two dimensional model of pitch range based on pitch level and pitch span with the perception of various speaker characteristics. The key to our measure of pitch range is that it is based on average data taken from clearly defined linguistic targets in speech. These targets included sentence-initial peaks, accent peaks, post-accent valleys and sentence-final lows. The results show that a pitch range model based on linguistic dimensions of variation better captures variation in listeners' judgements than the well established measures based on speakers' long term distributional properties of fO, such as 4 standard deviations around the mean, 95th-5th percentile and 90th-10th percentile. Most importantly this thesis shows that pitch range can and should be treated as the same entity across various research disciplines - extralinguistic, paralinguistic and linguistic - rather than the current situation in which pitch range has multiple definitions depending on the particular interest of the respective research discipline.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 56 citations exist upstream.