Speaker Recognition in an Unsupervised Environment using MFCC
In Speaker Recognition the main objective is to determine who is speaking i.e., recognize the unknown speaker. Speaker Recognition is the process of automatically identifying the identity of an individual based on the target source. It is classified into two: Speaker Identification and Speaker Verification. In this paper, Speaker Identification is been carried out with text independent type, as text independent has the advantage of being more flexible than text dependent and it can also be used for non-cooperating individuals. The aim is to locate and identify the speaker in an unsupervised environment. The recording of the individual speaker is done by accessing the microphone of the system in the program. Each recorded pre-processed digitized audio signal passes through the MFCC computation in the frequency domain. The audio signal of the individual speaker is vectored and smoothed with the help of the hamming window in the MFCC. The clustered vectors are been normalized with Vector Quantization technique which are stored in the database of the system. When the recognition of the unknown speaker goes on, the normalized database which is been created is allowed to pass through the searching procedure which is done by the Euclidean Distance. The unknown speaker audio signal is allowed to pass through the MFCC computation modelling and vector quantization process. The unknown speaker clustered signal compares the vectors with the already stored speakers and declare which most matches. Hence, the speaker is identified.
Keywords - Speaker Recognition, MFCCs, Hamming Window, Vector Quantization, Euclidean Distance