Time domain acoustic parameters for speech recognition