whisper_hindi

This repository contains the implementation of Whisper-Large ASR model on the Gramvaani and Kathbath dataset for evaluation

• The notebooks used for running the evaluation for the datasets are present here :

Kathbath here

Gramvaani here

The model used for the evaluation purposes is the Whisper-Large by OpenAI

Times taken to evaluate on the datasets :

Dataset	Time Taken for evaluation	Overall WER
Kathbath	~4.5 hours	0.2971
Gramvaani	~2 hours	0.5017

Specs used for evaluation (Kaggle basic):

RAM : 30GB

GPUs : T4x2