Application of symbolic machine learning techniques to the problem of audio signal segmentation

crazymeasleAI and Robotics

Oct 15, 2013 (3 years and 10 months ago)

139 views

Application of symbolic machine learning techniques to the problem of audio signal
segmentation

Dr. Arimantas Raskinis, dr. Gailius Raskinis

Vytautas Magnus Universiyu, Kaunas, Lithuania

Abstract.

In this paper, we address the application of symbolic machi
ne learning techniques to
the problem of audio signal segmentation. The whole segmentation process is
subdivided into four steps: 1) series of non
-
linear transformations are used for
building first
-
order features that allow easy detection of candidate segm
entation time
instants 2) second
-
order features that describe sound properties in the neighborhood
of a candidate segmentation instant are developed 3) segmentation candidate set is
transformed into machine learning dataset of positive and negative segment
ation
instances by labeling candidates in accordance to the annotated speech corpus 4)
supervised symbolic machine learning methods are applied resulting in segmentation
rules. We discuss the application of these processing steps to the automatic
segmentat
ion of speech into phonemes and to the transcription of music into notes.