Speaker Identification through Gender Detection
Mir Md Taosif Nur, Sumaiya Sultana Dola, Apurba Kishore Banik, Tanzeem Akhter, Nafees Hossain, A. B. M. Alim Al Islam, Jannatun Noor · 2022 International Conference on Innovation and Intelligence for Informatics, Computing, and Technologies (3ICT) · 2022
In modern technological advancement, voice recognition has played an integral part in many machine learning algorithms, having diversified application areas such as speech recognition, building better access control systems and security systems, etc. The goal of this research is to compare the results of two voice recognition methods, the first method involves identifying a speaker by first determining his/her gender using various machine learning techniques and then cross-matching the voice sample with the detected genders in the dataset using a pattern recognition algorithm. The second method uses the same pattern recognition algorithm on the voice sample without performing gender detection. To do so, we have utilized Mel Frequency Cepstral Coefficients (MFCC) for extracting audio features and used machine learning algorithms such as MLP, RBFN, Random Forest, KNN, Gradient Boosting, Decision Tree, Naive Bayes, Logistic Regression, and SVM for gender detection then compared the results between the nine classifiers. Then, pattern recognition algorithm GMM is applied for the detection of an individual. A dataset of our own and a readymade dataset from Kaggle were used for the research. Gradient Boosting and Random Forest showed high performance in both datasets in case of gender detection. Subsequently, combining gender detection algorithm with GMM led to improved accuracy in some instances.