Application Study On Vector Quantization In Speaker Identification

This thesis studies the improvement and application of Vector quantization (VQ) based speaker identification. Speaker identification was a biometrics that identify people via their voice, and VQ was the best model in the of speaker recognition because the method has a ability of condensing a lot of data.The author discuss the VQ based speaker identification in variety aspects and fulfilled a identification system, including speech collection, feature extraction and gain the identification result. Some improvement have made based on the system.1. Two speech corpuses, which include 50 people: the male have 29 people, and the female have 21 people, are built.2. Most popular and useful speaker models, vector quantization (VQ) model, and group vector quantization (GVQ) model are studied respectively. Based on the above feature parameters and speaker models, some complete speaker identification systems are established.3. Various phonetic studies have analyzed that different parts of speech signal have unequal discrimination properties between speakers. And the experiment approved the recognition improver with the method.4. Patition normalized distance measure (PNDM) is one of the design methods to improved codes. The work suggests the method makes the system performance improver than VQ and GVQ.