![]() | Only 14 pages are availabe for public view |
Abstract Machine translation systems for sign languages play a significant role in facilitating the communication between Deaf and hearing people, especially when signs interpreters are not available or their services are expensive. Using virtual human signers, to express the signing, make these systems more acceptable by Deaf and consequently more applicable. This thesis aims to study, develop, and implement a machine translation system from Arabic to Arabic Sign Language (ArSL). The developed system accepts the Arabic utterance as input via a microphone or audio files, and outputs the translation in an animated form by means of a signing avatar. It tries to help Deaf students attend courses and share materials with hearing students. Considering the Arabic Sign Language, many problems arise when developing such translation systems. First the Arabic Sign Language is a descriptive language; the signer describes the words to explain the meaning. Different signs may refer to the same word and different words may be expressed by the same sign. Second the two hands can be used interchangeably to express the same sign. Third the lack of Arabic Sign Language corpora, in this work a new corpus is built from scratch. Finally, the lack of linguistics studies on Arabic Sign Language, which limits the size of the corpus constructed in this work. The proposed system consists of three main modules: an Arabic speech recognizer, an Arabic machine translator, and a 3D signing avatar animator. The Arabic speech recognizer receives the Arabic speech either via a microphone or an audio file and converts it to its textual form. A speech recognition library is used to adapt an existing language model to the Arabic phones. The evaluation of this module separately gives an acceptable Word Error Rate (WER) using a test data set of real Arabic recorded sentences. The Arabic machine ii translator converts the Arabic text to its equivalent in Arabic Sign Language. |