TEXT-TO-SPEECH SYNTHESIZER BASED ON PHONETIC AND VOICE PROCESSING IN MARATHI AND HINDI LANGUAGE

The paper proposes an approach to implement a Text-To-Speech synthesizer in two Indian languages: Hindi and Marathi. This approach is based on unit selection process using synthesis-by-rules. It comprises phonetics normalizer and phones preparation. In order to enhance the selection process of text and voice signals, phonetics normalization was developed. In natural phonetics normalizer, a couple of Indian languages were normalized such as the vowels, consonants, abbreviation and word except conjunct characters. For phone preparation, the phone information has been recorded in voice form and segmented into the pieces of phones. The recorded voice with noise and noise-free signals were analyzed in the prosodic parameter such as formant frequencies. For objective measurement, formants (1st-4th) frequencies were evaluated on each generation of the sound sample. The model was tested on persons using subjective listening test. The test was applied on the syntactic generation of speech using MOS. The satisfactory result of MOS test was achieved 3.51 (between high and very high) out of 5. Ultimately, this system is well-organized for voice output like mankind voice.