Frequently asked questions

1. 1. How to Participate and/or access the data base?

Ans) Enrol yourself by registering on this link: Register Here!!!

1. 2. How to submit the challenge results?

Ans) Individual submission <LINK>

Team submission <LINK>

1. 3. Can participants use pre-trained acoustic and/or language models?

Ans)

1. Closed Challenge: One cannot use any pretrained models or data (Can use 100 hours Gramvaani data only).

2. Self Supervised Closed Challenge: One can use the 1000 hour Gramvaani data we are sharing to develop a pretrained model that is built only on Gramvaani data. No other pretrained models or data can be used. (Can use 100 hours Gramvaani data + 1000 hours Gramvaani data here)

3. Open Challenge: Open to all other external data, model etc

1. 4. Can participants use external data to train the model?

Ans) Yes, participants are allowed to use external data to train the model only on the Open Challenge track

1. 5. Can participants use different techniques such as data augmentation, speech enhancement etc with the ASR?

Ans) Yes, any approaches can be used with ASR

1. 6. Can participants use the data provided, outside the challenge?

Ans) After the completion of the challenge, one can use this data for other research activities. But data citation has to be done.

1. 7. What is the composition of sampling rate in test set?

Ans) The sampling rate distribution follows that of the training set.

1. 8. Can a participant build their own lexicon?

Ans) Yes, they can build their own lexicon.

1. 9. Can participants use external unlabelled data in Self Supervised Closed Challenge?

Ans) External data of any type can only be used in open challenge track.

1. 10. Can participants use pseudo-labelling in Self Supervised Closed Challenge?

Ans) Yes

1. 11. Availability of Dictionary?

Ans) Dictionary is Available in the baseline Github page. Please check the baseline tab webpage.

1. 12. Is the 100 hours training data transcription noisy?

Ans) Yes it is noisy.

1. 13. Can I use some other challenge data for training?

Ans) For open challenge track one can use, but not for the closed ones.

1. 14. Can participants use lexicon file present in baseline model for challenge 1 and 2 (closed challenges). Lexicon file contains more words than Train+Dev ?

Ans) Yes they can.

1. 15. How many submissions allowed from an Organization?

Ans) Not more than 3 submissions from an organization are allowed for each of the tracks.

1. 16. Is it allowed to add Dev set data to Train set?

Ans) Yes you can. You can use Dev and the Train data to train a model and evaluate the Eval. But make sure that you will not use that model to evaluate the Dev data.