leenag's picture
Update Readme.md
5f3a6c9 verified
|
raw
history blame
1.12 kB

Malayalam ASR Predicted output & Reference Samples

This repository contains evaluation results from the Malayalam ASR model "vrclc/Whisper_small_malayalam" using the "google/fleurs" dataset.

Model Details

Evaluation Results

  • Dataset Description: The test set of google/fleurs dataset which consists of Malayalam speech data.
  • Model Training: vrclc/Whisper_small_malayalam was trained with 50 hours of Malayalam speech data.
  • Evaluation Metric:The unnormalized evaluation result of 500 samples from google/fleurs and the WER is 60.4%.

This repository serves to provide an insight in error analysis which helps to identify general mistakes and areas for improvement in Malayalam speech recognition.