Repository for the project done at LEAP Lab, Indian Institute of Science, under the guidance of Neeraj Sharma and Prof. Sriram Ganapathy. The aim was to develop a real time and speaker independent model to detect speaker change in a given wave file.
TIMIT Dataset was used. Three disjoint set of speakers was created, and only SX and SI sentences for each speaker were taken into consideration while creating the files.
- lists contains different lists like which files were used for training, testing and validation. It also has scripts to generate the lists.
- models contains different classifiers, like DNN and CNN for different features.
- scripts contains data processing codes, like for generating gammatone, fbank, and other features. It also has scripts for combining the features generated according to context.
- resources contains plots used to visualize the features, and some generic diagrams used in the presentation.
I can be contacted at sidm@iitk.ac.in