The objective evaluation of sibilants

摘要

BACKGROUND: The inaccurate production of 's-sounds', i. e. sibilants represents the most common speech disorder. An objective description of such inaccurate speech sound production is mandatory. Several tools have been proposed and might be of great value, amongst them spectral analysis. The latter can nowadays be performed easily without the need for expensive equipment. METHOD: Different sibilants were uttered either as isolated speech sounds or embedded in test words and sentences. These were then digitized and spectral analysis was performed. Since several protocols have been put forward on how to execute the spectral analysis of sibilants, we replicated those protocols and evaluated how they exert influence on the analysis results. The selection of time windows or frequency bands regarding the computation of spectral moments will be considered in particular. RESULTS: Sibilants are characterised by a band-pass noise. Usage of power spectral density functions for the extraction of features from temporal and spatial signals is a standard approach. Thus, we calculated spectral moments to describe noise signals derived from sibilant productions. We demonstrate differing results based on varying frequency and time analysis windows. CONCLUSION: Unfortunately to date there is no generally accepted "gold standard" for the spectro analysis of speech sounds. For clinical routine, we recommend using a time window of 40 ms around the temporal midpoint, extracted from utterances or words. It is important, however, that recording conditions and analysis parameters are identical pre- and post therapy.