Automatic Speech Recognition for Non-native Speakers

Adélaïde Couplet, Anaïs Tack, Anne-Catherine Simon

Track: Type D (Master/Bachelor Thesis Abstracts)

Keywords: Natural Language Processing, Automatic Speech Recognition, Non-native Speakers, Whisper

Abstract: This abstract addresses automatic speech recognition, specifically the recognition of non-native speakers' speech. For our research, we use a corpus created by Dr. Ann-Sophie Noreillie (Noreillie, 2019), which consists of recordings of Dutch-speaking students speaking French, as well as recordings of French-speaking students speaking French. To perform the transcriptions, we use on OpenAI's Whisper model. In our work, we compare the transcriptions obtained with the different versions of the Whisper model. There are six versions of the model, trained with different numbers of parameters: tiny, base, small, medium, large, and large-v2. Our aim is to determine whether Whisper is as effective at transcribing the speech of non-native speakers as it is for native speakers, and to identify which version of Whisper performs best on our data.

Serve As Reviewer: ~Anaïs_Tack1

Submission Number: 76