Wav2vec2 Large Xls R 300m Sl With LM V2

🚀 wav2vec2-large-xls-r-300m-sl-with-LM-v2

This is a fine - tuned model based on facebook/wav2vec2-xls-r-300m for automatic speech recognition in the Slovenian language. It is trained on the MOZILLA - FOUNDATION/COMMON_VOICE_8_0 - SL dataset, achieving excellent performance on multiple evaluation metrics.

✨ Features

• Automatic Speech Recognition: Specialized for Slovenian speech recognition.
• Trained on High - Quality Dataset: Utilizes the Mozilla Foundation's Common Voice 8.0 dataset.
• Robust Performance: Demonstrates good results on both standard and robust speech event datasets.

📦 Installation

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

💻 Usage Examples

Basic Usage

The following are the evaluation commands for different datasets:

Evaluate on mozilla - foundation/common_voice_8_0

python eval.py --model_id DrishtiSharma/wav2vec2-large-xls-r-300m-sl-with-LM-v2 --dataset mozilla-foundation/common_voice_8_0 --config sl --split test --log_outputs

Evaluate on speech - recognition - community - v2/dev_data

python eval.py --model_id DrishtiSharma/wav2vec2-large-xls-r-300m-sl-with-LM-v2 --dataset speech-recognition-community-v2/dev_data --config sl --split validation --chunk_length_s 10 --stride_length_s 1

📚 Documentation

Model Information

Evaluation Results

The model has the following evaluation results:

• Common Voice 8:
  • Test WER: 0.21695212999560826
  • Test CER: 0.052850080572474256
  • Test WER (+LM): 0.14551310203484116
  • Test CER (+LM): 0.03927566711277415
• Robust Speech Event - Dev Data:
  • Dev WER: 0.560722380639029
  • Dev CER: 0.2279626093074681
  • Dev WER (+LM): 0.46486802661402354
  • Dev CER (+LM): 0.21105136194592422
• Robust Speech Event - Test Data:
  • Test WER: 46.69

Training Hyperparameters

The following hyperparameters were used during training:

• learning_rate: 7e - 05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9, 0.999) and epsilon = 1e - 08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 100.0
• 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

📄 License

This model is licensed under the Apache - 2.0 license.