Browsing by Author "Tademe Nadew"

Now showing 1 - 2 of 2
  • No Thumbnail Available
    Item
    Formant Based Speech Synthesis for Amharic Vowels
    (Addis Ababa University, 2008-01) Tademe Nadew; H/Mariam Sebsibe (PhD)
    Speech synthesis, a process of making artificial speech, has been one of a research area for the last many decades. But, even if there are several techniques of producing synthetic speech, yet it is still challenging to find one that overcomes all the limitations. Although there are challenges, synthetic speech is pertinent in many areas like in teaching and learning process, telecommunication, as aid for handicaps and so on. This study introduces a new method of formant-based speech synthesis for Amharic vowels. The model described in this work has two major parts. The analysis part: that handles the text analysis (transcription of the input word) and extraction of the speech parameters. The synthesis part: that generates the artificial speech. In the new model, smaller speech units like phonemes and the like are not stored in the database rather the speech parameters are stored. This highly minimizes the memory requirement of the speech synthesizer than other synthesizing methods. At run time, the CART machine selects best parameters from the database that exactly or closely matches to the contexts of the input vowel. At the same time the acoustic parameters of the file will be passed to the Formant Synthesizer to synthesize vowels from the given context of a word. The system is evaluated using MOS, one of the most popular testing techniques in speech synthesis. Experimental results showed that, the system has provided high flexibility of a speech with low memory and also data requirements. On average, 88.85% of the vowels are correctly recognized by the listeners. Keywords: Speech Synthesis, Formant, CART, Transcription, MOS
  • No Thumbnail Available
    Item
    Formant Based Speech Synthesis for Amharic Vowels
    (Addis Ababa University, 2008-01) Tademe Nadew; H/Mariam Sebsibe (PhD)
    Speech synthesis, a process of making artificial speech, has been one of a research area for the last many decades. But, even if there are several techniques of producing synthetic speech, yet it is still challenging to find one that overcomes all the limitations. Although there are challenges, synthetic speech is pertinent in many areas like in teaching and learning process, telecommunication, as aid for handicaps and so on. This study introduces a new method of formant-based speech synthesis for Amharic vowels. The model described in this work has two major parts. The analysis part: that handles the text analysis (transcription of the input word) and extraction of the speech parameters. The synthesis part: that generates the artificial speech. In the new model, smaller speech units like phonemes and the like are not stored in the database rather the speech parameters are stored. This highly minimizes the memory requirement of the speech synthesizer than other synthesizing methods. At run time, the CART machine selects best parameters from the database that exactly or closely matches to the contexts of the input vowel. At the same time the acoustic parameters of the file will be passed to the Formant Synthesizer to synthesize vowels from the given context of a word. The system is evaluated using MOS, one of the most popular testing techniques in speech synthesis. Experimental results showed that, the system has provided high flexibility of a speech with low memory and also data requirements. On average, 88.85% of the vowels are correctly recognized by the listeners. Keywords: Speech Synthesis, Formant, CART, Transcription, MOS