Environmental Noise Reduction Method Based on Formant Analysis-Synthesis Principle.

Abstract

A method of noise reduction based on formant analysis-synthesis principle is described. In this method, pitch and formants are extracted from noise superimposed speech, and noise reduced speech is synthesized from extracted parameters. To improve the accuracy of parameter extraction, the extraction procedure consists of three steps. First, vowel and consonant intervals are detected and rough estimates of the parameters in each interval are calculated. Secondly, the extracted parameters are checked to see if they are consistent in view of both global information such as parameter continuity and local information such as phonetic formant structure in order to make a plan for detailed reinvestigation. Lastly, the data are investigated again according to the plan in order to make a final decision of the value of the parameters. Experiments based on this method show that, in case of white noise superimposed speech, although the intelligibility of speech is rather decreased, the back ground noise is reduced and the easiness of hearing is fairly improved.