Spoken QA
Svitlana Vakulenko · UvA-DARE (University of Amsterdam) · 2021
Spoken questions automatically generated via Google Translate API for the following datasets (test splits only): Natural Questions https://ir-datasets.com/beir.html#beir/nq MS MARCO https://ir-datasets.com/beir.html#beir/msmarco/dev Simple Questions (WikiData) https://github.com/askplatypus/wikidata-simplequestions Script used for speech generation The automated transcriptions for all spoken questions were made using the Facebook wav2vec2 large ASR model (wav2vec2-large-960h-lv60-self) More info on the project: github.com/svakulenk0/spoken_qa