New Robust Speech Recognition By Using Nonlinear Running Spectrurn Filter
Noboru Hayasaka, Kham Khankhavivone, Yoshikazu Miyanaga, Kraisin Songwatana · 2006
This report proposes new design of nonlinear running spectrum filters. The running spectrum filters have been applied for robust speech recognition. Using these filters, the system can recognize speech accurately in low SNR. However, when we consider high SNR circumstances, these filters interfere sometimes with accurate speech recognition. In this report, another approach on robust speech recognition is introduced as nonlinear running spectrum filtering (NRSF). This method can be used in various cases of low-high SNRs. It is shown in this report that the new NRSF can be used for robust speech recognition with high accuracy and low calculation cost