DESIGNING AUTOMATIC SPEECH SYNTHESIZER FOR TIGRIGNA LANGUAGE

AKALU, TEKIEN

dc.contributor.author	AKALU, TEKIEN
dc.date.accessioned	2020-03-16T09:28:05Z
dc.date.available	2020-03-16T09:28:05Z
dc.date.issued	2020-03-16
dc.identifier.uri	http://hdl.handle.net/123456789/10357
dc.description.abstract	Speech synthesis is a computer-based system that has the ability to read any text and convert it to speech that resembles as native speaker of the language. Tigrigna is one of the under resourced language, in which discovering or preparing pronunciation dictionaries, letter to sound rules, transliteration tool and other resources that can be used as basic input for synthesizer is challenging. This paper explores model-based TTS system for the spoken language idom of Tigrigna language by integrating the transliteration tool to Clustergen synthesizer for transcribing the Unicode character in to International Phonetic Alphabet. The conversion process from input text into acoustic waveform is performed in a number of steps consisting of useful components. Both text and speech corpora are developed and 90% was used for training and 10% for testing among the developed corpus according to the synthesizer that Clustergen had implemented. The overall process started at modules of Natural Language Processing like transliterating, phone-set preparation, pruning speech corpus and text corpus preparation to generate the phones; then, labeling was done using the Ergodic Hidden Markov model (EHMM) labeler by applying generated phones and extracted features (spectral and excitation). After that modeling was done using CART (Classification And Regression Tree) to generate the trained model. At the end of those processes, the synthesized speech produced from the parameter, vocoder and trained model. Finally, the performance of the TTS system is measured in two ways: objectively and subjectively using the MCD (Mel cepstral distortion) which is built in the festvox/festival tool and MOS (mean opinion score) respectively in terms of intelligibility and naturalness. The overall performance of the system evaluated by MCD was 5.38 mean and by using MOS among 20 participants for randomly selected 25 sentences was 4.6 for Naturalness and 4.4 for intelligibility out of 5 score. As a general, we can conclude that, the developed model-based system is marvelous according to the obtained result	en_US
dc.language.iso	en	en_US
dc.subject	Computer Science	en_US
dc.title	DESIGNING AUTOMATIC SPEECH SYNTHESIZER FOR TIGRIGNA LANGUAGE	en_US
dc.type	Thesis	en_US