Multi-stages decision voice print recognition in telecommunication

Abstract

Using the cepstrum and pitch parameter, a VQ voice print recognition (i.e. speaker identification) method was realized and used in telecommunication. In this paper, we analyzed the characteristic of long-time LPCCEP and the pitch parameter for text-independent speaker identification. Considering the different contribution of each LPC cepstrum component, the weight was added to modify the distortion measurement. Based on the improved vector quantification (VQ), we constructed the speaker individual model and the system was training shorter, the response real time and the data memory smaller. A simple pitch measurement was introduced and applied in multi-stage decision to raise system robustness. The test results show that using about 10 seconds voice, in 60 speakers set, the correct identification rate is above 95%.

Keywords

• Voice print recognition
• Vector quantification(VQ)
• LPC cepstrum
• Pitch