Voice-Based Personalization

Abstract

Nowadays,most large-scale commercial and social websites uses recommendation sys- tem,through that they connect product or people to the user. Here We propose Rec- ommendation system based on voice input.In this thesis, we will discuss why voice based personalization is still a very challenging task. Then we will go through the previous research work done in this domain. Further, we will see the two important portions of voice-based personalization, Speaker Identification, and Recommendation system re- spectively. Then we will see the important feature extraction methods and classification methods required for Speaker Identification. The feature extraction method that we discussed is Mel frequency cepstral coeficient. And we will also discuss two machine learning algorithms that are required for speaker Identification, Vector Quantization, and K-nearest Neighbors respectively. For the Recommendation Systems, we discussed two techniques Collaborative ltering approach and Content-Based approach. Then we will explain our implementation. First, we will discuss the implementation of speaker identidication with MFCC and Vector quantization in MATLAB and also with K-Nearest Neighbors in Python. After that, we will explain the implementation of our recommenda- tion system using PredictionIO framework. Following this, we will explain the problem related to the integration of the two different modules of our project for it seamless working as a full- edged system.