Detecting Sentences Types in the Standard Arabic
Language
Ramzi Halimouche and Hocine Teffahi
Laboratory of Spoken Communication and Signal
Processing, Electronics and Computer Science Faculty, University of Sciences
and Technology Houari Boumediene, Algeria
Abstract: The standard
Arabic language, like many other languages, contains a prosodic feature, which
is hidden in the speech signal. The studies related to this field are still in
the preliminary stages. This fact results in restraining the performance of the
communication tools. The prosodic study allows people having all the communication
tools needed in their native language. Therefore, we propose, in this paper, a
prosodic study between the various types of sentences in the standard Arabic
language. The sentences are recognized according to three modalities as the
following: declarative, interrogative and exclamatory sentences. The results of
this study will be used to synthesize the different types of pronunciation that
can be exploited in several domains namely the man-machine communication. To
this end, we developed a specific dataset, consisting of the three types of
sentences. Then, we tested two sets of features: prosodic features (Fundamental
Frequency, Energy and Duration) and spectrum features (Mel-Frequency Cepstral
Coefficients and Linear Predictive Coding) as well their combination. We
adopted the Multi-Class Support Vector Machine (MC-SVM) as classifier. The
experimental results are very encouraging.
Keywords: Standard arabic
language, sentence type detection, fundamental frequency, energy, duration, mel-frequency cepstral coefficients,
linear predictive coding.
Received January 19, 2017; accepted August 23, 2017