Post-processings for PDAs in corrupted speech
Merouane Boudraa, Halim Sayoud, B. Boudraa, B. Guerin · 2002
In pitch detection, algorithms commit, often, some errors because the signal of speech is noisy and nonstationary. The errors are accordingly more important. In this paper, some new algorithms of post-processing have been developed. These techniques are destined to correct the errors in an optimal way. These post-processings have been tested in noisy environments at -6 dB of SNR. Improvements in precision of each algorithm have been objectively evaluated by elaborating some new error rates well adapted to this problem. The rates are called: absolute pitch error (APE) and three-dimensional spectral error (3DSE). The statistical results obtained after processing of 60 Arabic sentences by the logical filtering algorithms (LFA) revealed that the enhancements brought by the new algorithms are about 50%.