ISCA Archive ICSLP 1992
ISCA Archive ICSLP 1992

Towards synthesis of Hindi consonants using KLSYN88

Shyam S. Agrawal, Kenneth N. Stevens

This paper presents results of synthesis of Hindi consonants using KLSYN88 speech synthesizer. All frequently occuring 29 consonants of Hindi were synthesised in the initial position of CVC syllables. The central vowel /a/ and the final consonant III was always used to make the syllables into meaningful Hindi words. The words spoken by a standard Hindi male speaker were digitised at 10K samples per second using a VAX 750 computer system. Techniques employed for analysis include short term DFT magnitude spectrum, variations in formant frequencies, fundamental frequency and amplitude, and display of digital spectrograms. Quantitative acoustic parameters required for synthesis of phonetic features of phonemes were determined. The consonants were synthesised in combination with vowel Id to generate CV syllables and concatenated with the syllable /al/ to form CVC type synthetic words. A number of synthsiser control parameters were interactively varied for each sound till a satisfactory quality of synthetic speech and distinction among all the consonants was achieved. Special attention was paid to the synthesis of stops and affricates with various voicing and aspiration features. These sounds required careful selection and timing of source parameters. The spectral characteristics of synthesised and original sound segments were also compared to further improve the quality of synthetic speech. The results indicate that all the consonants of Hindi speech can be synthesised with natural quality. The perception tests indicate that the intelligibility scores for both types of speech are nearly the same. It is envisaged that further improvements are possible by careful control of synthesizer parameters and improvements in the aspiration source.