The identification and localization of speaker using fusion techniques and machine learning techniques

The systems of identification and localization of speakers are being used newly in diverse applications such as smart environments, audio conferences, and security, and social robotics which need more accuracy. The objective of this work is to define the localization of the speaker in sealed spaces and identifying the speaker in parallel using sound speaker signals. This work proposed a simulation of speaker localization and identification simultaneously using a feature fusion technique by constructing a feature vector which contains the features of identification and features of localization. The fusion technique has been used in each step of the proposed system such as data, feature, and decision fusion technique. Four Models were proposed for classifying the speaker are the Random Forest, the decision fusion which contains Random Forest and Support Vector Machine, the Restricted Boltzmann Machine which implemented by using the TensorFlow library from Google, and the long short-term memory technique was used which implemented using Keras library. The accuracy of the results was 66.39%, 82.035%, 99.84%, and 99.15% respectively.