Semantically Corrected Amharic Automatic Speech Recognition

This repository is still in the process of being organized.

This repository houses the Test Dataset, references to models, and code necessary to replicate the results presented in the paper 'Here'.

Content

1. Scripts for preparing and fine-tuning Facebook's Wav2Vec model.
2. Scripts for preparing and training Seq2Seq models, such as T5, from scratch with various token sizes.
3. Dataset of semantically corrected Amharic sentences used for evaluating output performance.
4. Demonstration script showcasing its practical application.