The Reserch And Application Of Speaker Detection And Tracking Technology

Speaker detection and tracking technology is an application of speaker recognition technology. Under the condition that the number and identity of speakers are unknown in the audio file, it relates to the problem of determining "who spoke when?" in order to solve speaker detection, segmentation and recognition effectively. Speaker detection and tracking technology has a wide application prospect. For example, for audio data, such as telephone conversations, broadcast news and meeting recordings, this technology is used to detect and track speakers’ voice segments, and then extract speakers’ information effectively. Speaker detection and tracking system mainly includes speech feature extraction, voice detection, speaker segmentation, speaker clustering and speaker recognition. Speech feature extraction, voice detection and speaker segmentation directly affect the performance of the system. In this paper, the following contents are studied:(1)Summarize current developing situation and basic technologies related to speaker detection and tracking system.(2)Summarize speech features used in speaker segmentation and clustering, and study the MFCC extraction and parameter setting. This paper combines the short-term energy and the pitch with MFCC respectively, and then selects the best feature after comparing the performance.(3)Summarize the common technology of voice detection algorithm and mainly study the voice detection algorithm based on the statistical model. This paper proposes onset detection algorithm in voice detection for mandarin, and attains the improvement and perfection of typical algorithm. In low SNR, voice detection algorithm can effectively reduce the error rate of onset voice. (4)Summarize the common technology of speaker segmentation algorithm and mainly focus on the speaker segmentation algorithm based on distance criterion. This paper compares BIC and DISTBIC speaker segmentation algorithm, and then selects the most appropriate segmentation algorithm for system.(5)Complete the design and implementation of speaker detection and tracking system, which includes feature extraction, voice detection, speaker segmentation, speaker cluster, speaker re-segmentation, speaker re-clustering and speaker recognition. This paper analyzes the function and performance of each technology used in the system.