Challenge

Challenge tracks

To ensure the maximum participation from research communities across the world and bring out various novel approaches, we have kept 4 different tracks that are shown below in the table. Track 1 encourages participants to utilize only the data shared through this special session and try out different modelling techniques and architectures. In tracks 2 and 3, we encourage participants to utilize the existing (beyond what is shared in this special session) resources or model checkpoints for either the acoustic model (AM) or the language model (LM) while building the other using in-corpus data (shared in this special session) only. This might be helpful in understanding the role of both AM and LM in building a reliable ASR in a low-resource setting. Finally, track 4 encourages participants to use any existing resources along with the in-corpus audio and text to get the best performance on the blind test set. All 4 tracks are summarized in the table below:

In corpus = approx 1000 hours dialect balanced supervised audio + text, along with approx 100k dialect labelled sentences 

Challenge Timeline

Registration opens - 27 May 2023

Dataset (train+dev) shared - 27 May 2023

Baselines release - 28 May 2023

Dataset (test) sharing - 28 June 2023

Challenge submission opens - 30 June 2023

Paper submission deadline -   03 July 2023  10 July 2023

Final challenge submission deadline -  5 July 2023 10 July 2023

Challenge acceptance results - 07 July 2023 12 July 2023

Paper revision deadline -   10 July 2023  17 July 2023

Note: Anywhere on Earth (AoE) time is used for deadlines