An Anti-noise Speech Recognition Model Based on Improved Wiener Filter and PUM

摘要

The noise resistance is the key in the study of speech recognition technology, this paper proposed an improved Wiener filter combining with the new model of PUM. The model by using the modified Wiener filter to filter out of broadband noise, provides the PUM only partial band noise pollution of speech signal, makes up for the PUM does not apply to speech signal frequency band to be noise pollution. At the same time, the model using the PUM as post-processing of speech enhancement, speech distortion caused by effectively eliminating the speech enhancement. The experimental results show that the modified Wiener filter combining PUM new model under the condition of different noise than other models have better voice recognition rate, can improve the word speech recognition rate of nearly 15%.