Noise Cancellation to Improve the Quality of LPC Processed Speech Degraded by Noise

摘要

Methods for improving the quality of the speech resulting from linear predictive analysis/synthesis of speech degraded by background noise are discussed. A method of noise cancellation using Wiener filtering in the frequency domain with the short-time Fourier transform was chosen for implementation. Implementation was done on a Data General Nova/Eclipse digital signal processing system in FORTRAN 5. Speech degraded by white gaussian noise was processed through linear predictive analysis/synthesis with and without noise cancellation preprocessing. Preliminary laboratory listenings verified that an improvement in quality was achieved with noise cancellation preprocessing. Although improvement in quality was achieved, more effort is required to make this implementation more efficient and improve the quality of speech produced. (Author)