摘要：Spoken language is the most natural way used by human to communicate information. Speech signal conveys linguistic information as well as speaker information (e.g. Emotional, regional and physiological characteristics). Such human ability has inspired many researches to understand production for developing the system that automatically process the richness of information in speech, this speech technology has many applications to find out “who is speaking†means speaker recognition system. This paper gives an overview of major techniques developed in each stage of speaker recognition. This paper has a list of techniques along with their results, merits and demerits. This also includes comparative study of different techniques, which are helping us to choose the technique for developing different language speaker recognition system like Marathi, Hindi and English. Time alignment of different utterances is a serious problem for distance measures and small shift would lead to incorrect identification. Dynamic time warping (DTW) is effective method. Vector quantization (VQ) is the classical quantization technique from signal processing. HMM is used for pattern matching. This paper shows the comparative result of PLC, PLPC and MFCC. MFCC have 73.62% is better result for Marathi language and 64.69% for Hindi language than PLC and PLPC techniques.
