Text independent neural network/rule based hybrid, continuous speech recognition