Burst Detection in Plosives for ASR

Seminar Type: Speech Processing, Seminar (3 SE)
Student: Sebastian Grill, André Menrath

Diffferent realizations of a plosive. Where is the burst? Is there a burst?

Plosives are the speech sounds /t, p, k, d, b, g/. In Automatic Speech Recognition (ASR) they are highly relevant for the correct recognition of words. In very careful spoken speech, a plosive consists of a closure, a burst and release friction. In fast, casually spoken speech however, plosives are realized with a high degree of variabiligty (see Figure, cl = closure, fr = friction, mb = multiple bursts).  Relevant characteristics of a burst are whether a burst happens or not and at which point in time the burst happens.

The aim of this bachelor project is to use MATLAB to built a burst detector which makes use of the information about slope of the energy of the speech. For this purpose, German read and spontaneous speechdatabases will be used.

(READ Thesis)



thesis_FINAL.pdf1.03 MB