Under the laboratory environment, speaker recognition has made great progress. But in real life, the performance of speaker recognition system is vulnerable to various factors, especially environmental noise and healthy condition. This paper studies the performance of speaker identification system when the tester suffers from the cold. The cold tends to induce inflammation and swelling of the nasal cavity, and then changes the modulation of nasals to sound source excitation signal and makes the speaker’s voice changed. R.G. Tull also found the speaker recognition system’s performance significantly decreases when taking normal speech speaked by the healthy persons as train speech, while cold speech by persons who are catching cold as test speech.
In this paper, through studying the composition of nasal and comparing the frequency domain properties of normal speech and cold speech, we find that the cold makes the low frequency components larger and high-frequency components smaller. So we propose the method using different pre-emphasis filter process normal speech and cold speech. Experimental results show that this method can improve the performance of the speaker identification system by 6% compared to general method all speeches are processed with the same filter.