README.md

metadata

base_model: vasista22/whisper-gujarati-small
datasets:
  - 1rsh/gujarati-openslr
language:
  - gu
license: apache-2.0
metrics:
  - wer
  - cer
tags:
  - hf-asr-leaderboard
  - generated_from_trainer
model-index:
  - name: Whisper Small Gujarati OpenSLR
    results:
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: Gujarati OpenSLR
          type: 1rsh/gujarati-openslr
          args: 'split: train'
        metrics:
          - type: wer
            value: 35.325794291868604
            name: WER
          - type: cer
            value: 22.3685
            name: CER
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: Google FLEURS
          type: google/fleurs
          args: 'config: gu_in; split: test'
        metrics:
          - type: wer
            value: 46.596808306094985
            name: WER
          - type: cer
            value: 22.69041389733006
            name: CER
          - type: nwer
            value: 44.01335002085941
            name: Normalized WER
          - type: ncer
            value: 18.702293460048406
            name: Normalized CER

Whisper Small Gujarati OpenSLR

This model is a fine-tuned version of vasista22/whisper-gujarati-small on the Gujarati OpenSLR dataset. It achieves the following results on the evaluation set:

• Loss: 0.0472
• Wer: 35.3258
• Cer: 22.3685

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.0018	4.9505	1000	0.0472	35.3258	22.3685

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1

Usage

In order to infer a single audio file using this model, the following code snippet can be used:

>>> import torch
>>> from transformers import pipeline

>>> # path to the audio file to be transcribed
>>> audio = "/path/to/audio.format"
>>> device = "cuda:0" if torch.cuda.is_available() else "cpu"

>>> transcribe = pipeline(task="automatic-speech-recognition", model="1rsh/whisper-small-gu", chunk_length_s=30, device=device)
>>> transcribe.model.config.forced_decoder_ids = transcribe.tokenizer.get_decoder_prompt_ids(language="gu", task="transcribe")

>>> print('Transcription: ', transcribe(audio)["text"])