Thesis or Dissertation Fast Search of Audio Fingerprint using K40 GPGPU

Nguyen, Mau Toan

Nowadays, there are millions of audio and video contents uploaded to the Internet, so the searching speed and database organization are the problems for the audio management system. Audio fingerprint is the digital fingerprint that can help to identify the audio content. With the advantages of audio fingerprint, we can reduce the size of data to hundreds of times less than storing original audio raw data. And with audio fingerprint, we have a standard format that supports to compare or structuralize the database. In this thesis, we propose a new hierarchy searching system that can detect the meta information for fingerprint in real time by using the advantages of K-modes and Locality Sensitive Hashing (LSH). The K-modes is used as Level 1 in our method and works in CPU. K-modes supports in clustering the big database into sub-databases that can store to GPGPU devices. In searching step, K-modes is responsible for finding the nearest centroid of every query and send this query to suitable GPGPU device. LSH will handle the data structure of GPGPU devices' sub-database and respond for management the kernel that is compatible with parallel in single GPGPU. Our method can combine the advantages of both CPU and GPGPUs by putting together in the same computer system. With the power of multiple GPGPU devices, we can obtain the meta information for a query within 2 milliseconds for 10 million songs' database.
Supervisor:井口 寧

Number of accesses :  

Other information