Wav2vec2 Large Xls R 300m Kk With LM

🚀 wav2vec2-large-xls-r-300m-kk-with-LM

This is a fine - tuned model based on facebook/wav2vec2-xls-r-300m for Automatic Speech Recognition in the Kazakh (kk) language. It offers high - quality speech recognition performance on relevant datasets.

🚀 Quick Start

Evaluation Commands

1. Evaluate on mozilla - foundation/common_voice_8_0 with test split

python eval.py  --model_id DrishtiSharma/wav2vec2-large-xls-r-300m-kk-with-LM  --dataset mozilla-foundation/common_voice_8_0 --config kk --split test --log_outputs

2. Evaluate on speech - recognition - community - v2/dev_data Kazakh language isn't available in speech - recognition - community - v2/dev_data

✨ Features

• Multilingual Adaptability: Built on a large - scale pre - trained model, it can adapt well to the Kazakh language.
• High - Performance Metrics: Achieves good results in WER (Word Error Rate) and CER (Character Error Rate) on the evaluation datasets.

📦 Installation

No specific installation steps are provided in the original document, so this section is skipped.

💻 Usage Examples

No code examples are provided in the original document, so this section is skipped.

📚 Documentation

Model Information

Evaluation Results

This model has the following performance on different datasets:

• Common Voice 8 (ru):
  • Test WER: 0.4355
  • Test CER: 0.10469915859660263
  • Test WER (+LM): 0.417
  • Test CER (+LM): 0.10319098269566598
• Robust Speech Event - Dev Data (kk):
  • Test WER: NA
  • Test CER: NA
• Common Voice 8.0 (kk):
  • Test WER: 41.7
• Robust Speech Event - Test Data (kk):
  • Test WER: 67.09

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.000222
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9, 0.999) and epsilon = 1e - 08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 150.0
• mixed_precision_training: Native AMP

Training results

Framework versions

• Transformers 4.17.0.dev0
• Pytorch 1.10.2+cu102
• Datasets 1.18.2.dev0
• Tokenizers 0.11.0

🔧 Technical Details

No in - depth technical details are provided in the original document, so this section is skipped.

📄 License

This model is released under the Apache - 2.0 license.