Running Spectrum Filtering in Speech Recognition