DENOISING SPEECH OF MARATHI NUMERALS USING SPECTRAL SUBTRACTION

This paper presents the denoising technique based on spectral subtraction for speech synthesis of the Marathi numerals. Numerals are recorded through mice and normalized the signals with PRAAT tools. Different form of speech signals were analyzed by added noise in original speech signals. The voice activity detection (VAD) algorithm estimate the noise spectrum of original speech signal. Spectral Subtraction technique is adopted to reduce the noise. It exploits the ability of actively unwanted signals of speech. Spectral Subtraction method has reduced noise and improved the quality of speech signals. This paper concentrates on the application of quality speech signals for speech synthesis and the results found to be satisfactory.