BrianatCambridge/SelectiveHearingBench

Test set main file

all_questions_descfilter_bg.json

This contains a list of Python Dictionaries in the following format

{
"question": The question,
"options": A list of options to choose from,
"answer": The answer to the question (corresponds to one option in the options field),
"speaker": main or bystander,
"audio": Relative path to the audio,
"scene": The scenario,
"main_speaker_desc": Description only based on speaker characteristics,
"main_speaker_desc_w_content": Description based on both speaker characteristics and some speech content clues,
"main_speaker_desc_w_content_filter": Removed any information leakage, used in the paper,
"idk_answer": The I don't know answer (corresponds to one option in the options field),
"type": scenario type, real or synthetic
}

Training set file

synth_audio_questions_train_filter.json with the same structure as the test one.