Human Voice Abstraction

Abstract

Human voice abstraction is one of the basic problems in voice transmission. Also it becomes a more practical issue when doing voice recording with background noises. An adequate separation of pure human voice from the outside noise is crucial, as speech audio file can include many outside frequencies along with variable amplitudes. Voice filtering should be done more carefully as there can be many frequencies which blend with the human voice. In this work we have successfully developed a voice filtering technique. Based on reducing the background noise to abstract the human voice, we use a hybrid approach with optimization to reduce environment interference. Wavelet de-noising technique has been used to remove the noise from speech. To cut off bad frequencies and amplitudes, the soft thresholding mechanism has been used. This dissertation details previous techniques of voice abstraction and describe the solution with overall system structure, the implementation method along with the evolution of the approach.