Wav2vec2 Large XLSR 53 for Malayalam

This repository contains the code for Wav2vec2 Large XLSR 53 model fine-tuned on Malayalam (ml) language for Automatic Speech Recognition (ASR). We use multiple public datasets to fine-tune the model. This model was trained as part of XLSR fine-tuning week organized by Hugging Face. The fine-tuned model can be directly used for downstream tasks from Hugging Face Models repository : https://huggingface.co/gvs/wav2vec2-large-xlsr-malayalam

Datasets

• IIIT-H Indic Speech Dataset
• Indic TTS Malayalam Speech Corpus
• SMC Malayalam Speech Corpus
• Openslr Malayalam Speech Corpus

The datasets needs to be converted to format suitable for Hugging Face Datasets library. This can be done using the notebook provided here.

Fine-tuning

The model can be fine-tuned using this notebook. Additional details on the fine-tuning procedure is given in the notebook itself.