| Mandarin Chinese as a language spoken by the largest number of people in the world has a long history and rich connotations. With a very strong rhythm, mandarin is tonal languages from the pronunciation, pronunciation intonation. Compared to most languages especially indo-european language, mandarin has its own distinctive features. This paper studies mandarin in the aspect of voice, voice feature extraction with Chinese characteristics, and focus on advanced features to apply it to the Chinese language in the field of speaker recognition. The purpose of our research is to improve the recognition rate of Chinese speaker recognition.At the beginning of the study focuses on the basis of feature extraction, especially the description of pitch which is used to describe the tone of voice and the stress which is used to describe short-time energy. Using a variety of methods of pitch tracking, reduce bad point in the track pitch period. Contrast loudness and short-time energy extraction principle, draw lessons from the extraction of PLP method, this paper proposes a new feature can represent energy characteristics of perception. Considering the features of the auditory, the ability to distinguish the speaker of perceptual energy is superior to short-term energy.After get the reliable basic characteristics, the pitch and short-time energy will be used to generated description rhythm notation. Model of prosodic symbol using n-grams, described the voice with very good Markov property, use of sound and advanced features of information on the speaker. Then, further analyses the characteristics of Chinese rhythm, the rhythm words rhyme phrases to join rhythm notation. Complete Chinese prosodic feature extraction and modeling, to improve the recognition efficiency.Finally, we completed the open set speaker recognition system and provide five features for system. Through the experiment of scores fusion, reached the use of Chinese prosodic features to improve speaker recognition effect. |
PDF Full Text Request