Speaker-independent spoken digit recognition in noisy environments using dynamic spectral features and neural networks