Whisper Tamil Large V2

🚀 Whisper Tamil Large-v2

This model is a fine - tuned version of openai/whisper-large-v2 on Tamil data from multiple publicly available ASR corpuses. It was fine - tuned as part of the Whisper fine - tuning sprint, offering enhanced performance for Tamil automatic speech recognition.

NOTE: The code for training this model can be reused from the whisper-finetune repository.

🚀 Quick Start

✨ Features

• Fine - tuned on multiple Tamil ASR corpuses.
• Code for training and evaluation is publicly available for re - use.
• Supports faster inference with whisper - jax.

📦 Installation

The installation steps are not explicitly provided in the original README. However, you can refer to the whisper-finetune repository for relevant installation codes and scripts.

💻 Usage Examples

Basic Usage

>>> 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="vasista22/whisper-tamil-large-v2", chunk_length_s=30, device=device)
>>> transcribe.model.config.forced_decoder_ids = transcribe.tokenizer.get_decoder_prompt_ids(language="ta", task="transcribe")

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

Advanced Usage

>>> import jax.numpy as jnp
>>> from whisper_jax import FlaxWhisperForConditionalGeneration, FlaxWhisperPipline

>>> # path to the audio file to be transcribed
>>> audio = "/path/to/audio.format"

>>> transcribe = FlaxWhisperPipline("vasista22/whisper-tamil-large-v2", batch_size=16)
>>> transcribe.model.config.forced_decoder_ids = transcribe.tokenizer.get_decoder_prompt_ids(language="ta", task="transcribe")

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

📚 Documentation

Training and evaluation data

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.75e - 05
• train_batch_size: 8
• eval_batch_size: 24
• seed: 22
• optimizer: adamw_bnb_8bit
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 22000
• training_steps: 52500 (Initially set to 76000 steps)
• mixed_precision_training: True

🔧 Technical Details

This model is a fine - tuned version of openai/whisper-large-v2 on Tamil data. The fine - tuning was carried out as part of the Whisper fine - tuning sprint. The training and evaluation codes are available in the whisper-finetune repository.

📄 License

This model is licensed under the Apache - 2.0 license.

Model Performance

Acknowledgement

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