Data and Python scripts for analyzing, evaluating, and visualizing data and fitted models for a noise-masked speech experiment in which 11 listeners identified the consonants in the non-word syllables 'ta', 'da', 'sa', and 'za' produced by 20 talkers and embedded in multi-talker babble.
Trial-by-trial identification-confusion data is contained in the file tdsz_grt_data.csv. Columns indicate trial, ISI, stimulus category (stimcat), response, block, talker, talker index (tkidx), stimulus category index (stidx), (within-talker) token index (tnidx), and listener.
PyMC model fitting scripts are named pm2_grt_*_tdsz.py. The acronym lts in some file names stands for "listener talker scaling," reflecting components of the models wherein listener- and talker-specific parameters scale a subset of the shared group-level parameters. A script for fitting these models using the ipyparallel module is in ipyparallel_grt_fits.py.
The remaining .py files are for model evaluation and model and data visualization.
The *.hdf5 files contain chains of samples from the posterior distributions for different models.
Finally, the file silbert_zadeh_grt_tdsz_2017_preprint.pdf gives a detailed account of the experiment and analyses.