# 90th Place Solution for the Bengali.AI Speech Recognition

I would like to thank the hosts for the wonderful task! I feel so bad that I had only 3 days for this competition and didn't join earlier.
Context
Business context:
[https://www.kaggle.com/competitions/bengaliai-speech/overview](https://www.kaggle.com/competitions/bengaliai-speech/overview)
Data context:
[https://www.kaggle.com/competitions/bengaliai-speech/data](https://www.kaggle.com/competitions/bengaliai-speech/data)
Overview of the Approach
As almost everyone else I used pretrained on bengali speech model. Main parameters for fine-tuning are listed below
=,
=,
=,
=,
=,
=,
=,
=,
=,
=e-,
=
and the following configuration was used
.freeze_feature_extractor()
.freeze_feature_encoder()
Details of the submission
It took me 16h to fine-tune a model on 10% of train data for 3 epochs on my GPU — thats where I felt the need for time.
I found that increasing
beam_width
in decoder from default to 2048 lead to 0.003 boost both on public and private (and inference time is ~8h, so we cant really increase further).
Sources
[https://www.kaggle.com/code/takanashihumbert/bengali-sr-wav2vec-v1-bengali-training](https://www.kaggle.com/code/takanashihumbert/bengali-sr-wav2vec-v1-bengali-training)
— notebook that I used for fine-tuning
[https://www.kaggle.com/competitions/bengaliai-speech/discussion/435300](https://www.kaggle.com/competitions/bengaliai-speech/discussion/435300)
— datasets and model checkpoints for resource efficient training