iVaani - Fine-tuned ASR model for Hindi Language

Model Description

This is iVaani model, specifically optimized for the Hindi language. The fine-tuning process has led to an improvement in accuracy by 2.5% compared to the original Whisper model.

Performance

After fine-tuning, the model shows a 2.5% increase in transcription accuracy for Hindi language audio compared to the base Whisper medium model.

How to Use

You can use this model directly with a simple API call in Hugging Face. Here is a Python code snippet for using the model:

from transformers import AutoModelForCTC, Wav2Vec2Processor

model = AutoModelForCTC.from_pretrained("rukaiyah-indika-ai/iVaani")
processor = Wav2Vec2Processor.from_pretrained("rukaiyah-indika-ai/iVaani")

# Replace 'path_to_audio_file' with the path to your Hindi audio file
input_audio = processor(path_to_audio_file, return_tensors="pt", padding=True)

# Perform the transcription
transcription = model.generate(**input_audio)
print("Transcription:", transcription)

Additional Language Models

Indika AI has also fine-tuned ASR (Automatic Speech Recognition) models for several other Indic languages, enhancing the accuracy by 2-5% for each language. The word error rate has also been significantly reduced.

The additional languages include:

Language	Original Accuracy
Bengali	88%
Telugu	86%
Marathi	87%
Tamil	88%
Gujarati	90%
Kannada	86.5%
Malayalam	87.5%
Punjabi	89%
Odia	88.5%

BibTeX entry and citation info

If you use this model in your research, please cite it as follows:

@misc{whisper-medium-hindi-fine-tuned,
  author = {Indika AI},
  title = {iVaani},
  year = {2024},
  publisher = {Hugging Face},
  journal = {Hugging Face Model Hub}
}

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 2
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• training_steps: 1000
• mixed_precision_training: Native AMP

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu121
• Datasets 2.16.0
• Tokenizers 0.15.0