A comparative study of two area function derivation techniques for fricative synthesis

Abstract

It is still unclear to speech scientists how the brain is able to distinguish fricative sounds from the many cues within the speech signal. This thesis makes an acoustic and articulatory investigation mint the problem, using fricative utterances and magnetic resonance (MR) images.

The results suggest that the cues to a fricative’s identity lie in the higher frequency region. For the sibilant fricatives, these are the relatively higher amplitude levels about 3.0kHz, while the distinct peak in the vicinity of 2.8 kHz gives the sound a “pitch quality” that was significantly perceived by the listeners. For the nonsibilant fricatives, the cues within the frication period did not provide sufficient information as to the fricatives’ identity even though coarticulatory effects due to vowel context persisted; nevertheless, rough matches in amplitude levels were adequate to produce a synthetic version which listeners decided were compared to the natural sound. Accurate measurements of the region posterior to the constriction were not essential for synthesis, supported by the fact that pyriform sinuses do not play an important role in modelling. The sublingual cavity was observed to affect the amplitude of the significant peak in /S/ but it is sufficient to include it as an increase in area. The Mermelstein technique is proposed to be most suitable for fricative synthesis: it is easier to implement and side branches do not need to be modelled realistically.

More information

Identifiers

Catalogue record

Export record

ASCII CitationAtomBibTeXData Cite XMLDublin CoreDublin CoreEP3 XMLEndNoteHTML CitationHTML CitationHTML ListJSONMETSMODSMPEG-21 DIDLOpenURL ContextObjectOpenURL ContextObject in SpanRDF+N-TriplesRDF+N3RDF+XMLRIOXX2 XMLReferReference ManagerSimple Metadata