Whisper Hindi Large V2

🚀 Whisper Hindi Large-v2

This model is a fine - tuned version of openai/whisper-large-v2 on Hindi data from multiple publicly available ASR corpuses, contributing to speech recognition in Hindi.

🚀 Quick Start

This model is a fine - tuned version of openai/whisper-large-v2 on the Hindi data available from multiple publicly available ASR corpuses. It has been fine - tuned as a part of the Whisper fine - tuning sprint.

NOTE: The code used to train this model is available for re - use in the whisper-finetune repository.

✨ Features

• Fine - tuned on multiple publicly available Hindi ASR corpuses.
• The training code is open - source and available for re - use.
• Supports evaluation on entire datasets and single audio file inference.
• Allows for faster inference using whisper - jax.

📦 Installation

The installation steps are not explicitly provided in the original README. However, relevant evaluation and inference codes rely on repositories like whisper-finetune and the transformers, torch, whisper - jax libraries. You may need to install these dependencies according to their official documentation.

💻 Usage Examples

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

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

Advanced Usage

For faster inference of whisper models, the whisper-jax library can be used. Please follow the necessary installation steps as mentioned here, before using the following code snippet:

>>> 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-hindi-large-v2", batch_size=16)
>>> transcribe.model.config.forced_decoder_ids = transcribe.tokenizer.get_decoder_prompt_ids(language="hi", task="transcribe")

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

📚 Documentation

In order to evaluate this model on an entire dataset, the evaluation codes available in the whisper-finetune repository can be used. The same repository also provides the scripts for faster inference using whisper - jax.

🔧 Technical Details

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: 25000
• training_steps: 57000 (Initially set to 116255 steps)
• mixed_precision_training: True

📄 License

This model is released under the Apache - 2.0 license.

Acknowledgement

This work was done at Speech Lab, IIT Madras. 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.