Utterance verification for text-dependent speaker recognition: a comparative assessment using the RedDots corpus