metadata

language:
  - gu
license: apache-2.0
tags:
  - whisper-event
  - generated_from_trainer
metrics:
  - wer
model-index:
  - name: Whisper Gujarati Base - Vasista Sai Lodagala
    results:
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: google/fleurs
          type: google/fleurs
          config: gu_in
          split: test
        metrics:
          - type: wer
            value: 30
            name: WER

Whisper Gujarati Base

This model is a fine-tuned version of openai/whisper-base on the Gujarati data available from multiple publicly available ASR corpuses. It has been fine-tuned as a part of the Whisper fine-tuning sprint.

Training and evaluation data

Training Data: ULCA ASR Corpus, OpenSLR, Microsoft Research Telugu Corpus (Train+Dev), Google/Fleurs Train+Dev set.

Evaluation Data: Google/Fleurs Test set, Microsoft Research Telugu Corpus Test .

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 3.3e-05
train_batch_size: 80
eval_batch_size: 88
seed: 22
optimizer: adamw_bnb_8bit
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 4000
training_steps: 4250 (terminated upon convergence. Initially set to 21250 steps)
mixed_precision_training: True

Acknowledgement

This work was done at Speech Lab, IITM. The compute resources for this work were funded by "Bhashini: National Language translation Mission" project of the Ministry of Electronics and Information Technology (MeitY), Government of India.