RSR2015ΒΆ
RSR2015 is a database collected by the Institute for Infocomm Research (I2R),
A*STAR in Singapore to support the development and evaluation of
text-dependent speaker verification algorithms.
This database can be bought from I2R at a very low price.
For more details, visit RSR2015 website.
Metat data (Keys, IdMap and Ndx) can be downloaded from Download lists for standard datasets.
- Prepare to run experiments on RSR2015
- Run a GMM-UBM system
- Input/Output
- Set your own parameters
- Load IdMap, Ndx, Key from HDF5 files and ubm_list
- Train the Universal background Model (UBM)
- Compute the sufficient statistics on the UBM
- Adapt the GMM speaker models from the UBM via a MAP adaptation
- Compute all trials and save scores in HDF5 format
- Plot DET curve and compute minDCF and EER
- Run a SVM GMM system on the RSR2015 database