Accent Rating

This is a collection of scripts and data I used when working on my dissertation.

Recording procedure (Praat directory)

• recording_procedure.praat - this is the script I used for collecting the audio data (it's a mess like most Praat scripts I've written, but it was good enough to do what I wanted)

PsyToolkit (v3.2.0)

• survey.sur - PsyToolkit survey for collecting metadata and running the experiment
• rating_experiment.psy - The experiment itself
• scale_l_to_r.txt - Code that draws a sliding scale in rating_experiment.psy with left-to-right orientation
• scale_r_to_l.txt - Code that draws a sliding scale in rating_experiment.psy with right-to-left orientation
• sounds.txt - List of audio files used in the experiment (audio files are not included due to license limitations)
• bitmaps/ - Collection of pictures used in rating_experiment.psy
• tables/ - Collection of sounds used in rating_experiment.psy. Each table has three columns: variable name, file name, duration

EER measurements

• measure_EER.py - script that draws DET curves and calculates EER values from the sid_results_table_individual_files_filtered.csv results table. Simply run python measure_EER.py. The script creates a folder named EER_measurements which contains temporary files to verify that same and different speaker labels are correct.
• sid_results_table_individual_files_filtered.csv - different-sex recordings have been filtered out

Language identification

• phonexia_lid_wrapper.sh - wrapper for the Phonexia LID technology (this is mainly here to show which subset of languages was used in the analysis)
• speechbrain_lid_wrapper.py - wrapper for the SpeechBrain LID technology it subclasses the EncoderClassifier to extract scores for English and Czech apart from the best-matching language (the audio files are not included due to license limitations so the script does not really produce the results!)

Speaker identification

• sid4_wrapper.sh - wrapper for the Phonexia SID4 technology. A simple helper script which extracts voiceprints from all wave files in one directory and then compares all voiceprints with each other. It also saves the amount of "net speech" in each voiceprint to a file.
• speechbrain_sid_wrapper.py - a wrapper for SpeechBrain's SpeakerRecognition. It has options for computing embeddings from audio files or loading previously created embeddings. It subclasses SpeakerRecognition in order to be able to perform verification on embeddings instead of on audio files. The script creates embeddings from all audio files in a directory and then performs speaker verification either on all combinations or all permutations of the embeddings set.