Bereket, Kasaye Tikui (2008) Developing a Speech Synthesizer for Amharic Language Using Hidden Markov Model. Masters thesis, Addis Ababa University.
![[img]](http://thesisbank.jhia.ac.ke/style/images/fileicons/application_pdf.png) |
PDF (Developing a Speech Synthesizer for Amharic Language Using Hidden Markov Model)
Bereket, tikui.pdf - Accepted Version Restricted to Repository staff only Download (815kB) | Request a copy |
Abstract
Speech synthesis systems are concerned with generating a natural sounding and intelligible speech by taking text as input. Speech Synthesizers are very important in helping impaired people, in teaching and learning process, for telecommunications and industries. Though it has many applications, generating intelligible and natural sounding synthetic speech has been a challenging task for years. To overcome these challenges, different techniques have been studied and implemented. Though speech synthesizers based on HMM are done for foreign languages, they are not applicable for Amharic language since the languages special characteristics are not considered in these synthesizers. Hence, in this thesis work Hidden Markov Model based speech synthesis for Amharic language (HTS-FA) is done. The HTS-FA has two phases: the training and synthesis phase. The main activities included in the training phase are preparation of the training dataset, language modeling, feature extraction and training the model. In the synthesis phase, models are selected according to the text to be synthesized, and then speech parameters are generated from them. Finally, the synthesized speech is generated from the speech parameters. A total of five hundred sentences are used for training the model from a corpus having a size of 11,670 sentences, and twenty sentences, which are not included in the training dataset, are used for testing the performance of the system. In this thesis, the Mean Opinion Score (MOS) evaluation technique is used. The results from the MOS were found to be 4.12 and 3.6 for intelligibility and naturalness respectively for speeches synthesized by HTS-FA. Using concatinative method the result obtained for intelligibility and naturalness are 3.54 and 3.25 respectively.
| Item Type: | Thesis (Masters) |
|---|---|
| Uncontrolled Keywords: | Speech synthesis, HMM, HMM based speech synthesis, Language Modeling |
| Subjects: | P Language and Literature > PL Languages and literatures of Eastern Asia, Africa, Oceania Q Science > QA Mathematics Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QA Mathematics > QA76 Computer software |
| Divisions: | Africana |
| Depositing User: | Selom Ghislain |
| Date Deposited: | 11 Sep 2018 12:27 |
| Last Modified: | 11 Sep 2018 12:27 |
| URI: | http://thesisbank.jhia.ac.ke/id/eprint/5267 |
Actions (login required)
 |
View Item |