Performance of a Wavelet-Based Frontend Under Typical Noisy Environments for Continuous Speech Recognition