1 results listed
The core of kinship verification is to automatically
determine if two individuals are from the same family, in other
words, to verify if two or more individuals are in a kin
relationship by analyzing the given dataset. Kinship verification
becomes an important research area in the computer vision field.
Many previous datasets were collected from the internet, and
various factors affect them, such as image lighting, low internet
quality images or videos, facial expressions, and others. We
assume that the human voice can contain some features that can
be used to verify kinship. In this paper, we establish a novel
kinship dataset called Kinship Families Voices (Kin-FVs)
contains many families’ voices. Kin-FVs consists of 25 families of
94 individuals, where each individual has six voice records;
therefore, the total voice records is 564 recorded in various
languages such as Arabic, Assyrian, Kurdish, Turkmen, Turkish,
Swedish, Greek, and German. We performed data pre-processing
and analysis, then used MFCC, Delta, and Double Delta to
extract features and save them in csv file. We perform a
preliminary experiment by using a multilayer perceptron (MLP)
model, which achieves 78.5% accuracy. We reshape the.
extracted features and fed it to a 2D CNN, which achieves 87.0%
accuracy. These results prove the effectiveness of the new dataset
to be a new direction in the field of kinship verification based on
the features extracted from the human voice using machine
learning techniques
International Conference on Advanced Technologies, Computer Engineering and Science
ICATCES
Olivia Khalil Oraha
Yusra Faisal Mohammad