Wav2vec2 Xls R Sl A2

🚀 wav2vec2-xls-r-sl-a2

This is a fine - tuned model based on facebook/wav2vec2-xls-r-300m for automatic speech recognition in the Slovenian language. It provides high - quality speech recognition performance on multiple datasets.

🚀 Quick Start

Evaluation Commands

• Evaluate on mozilla - foundation/common_voice_8_0 with test split:

python eval.py --model_id DrishtiSharma/wav2vec2-xls-r-sl-a2 --dataset mozilla-foundation/common_voice_8_0 --config sl --split test --log_outputs

• Evaluate on speech - recognition - community - v2/dev_data:

Votic language not found in speech-recognition-community-v2/dev_data

✨ Features

• High - quality Speech Recognition: Achieves low WER and CER on multiple datasets, demonstrating excellent performance in automatic speech recognition.
• Fine - tuned for Slovenian: Specifically optimized for the Slovenian language, suitable for related speech recognition tasks.

📦 Installation

The installation process is not provided in the original document, so this section is skipped.

💻 Usage Examples

The original document does not provide code examples, so this section is skipped.

📚 Documentation

Model Information

Evaluation Results

This model has achieved the following results on different datasets:

• Common Voice 8:
  • Test WER: 0.21695212999560826
  • Test CER: 0.052850080572474256
• Robust Speech Event - Dev Data (vot):
  • Test WER: 0.560722380639029
  • Test CER: 0.2279626093074681
• Robust Speech Event - Dev Data (sl):
  • Test WER: 56.07
• Robust Speech Event - Test Data (sl):
  • Test WER: 56.19

Training Hyperparameters

• 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
• Number of 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

🔧 Technical Details

Training Process

This model is a fine - tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA - FOUNDATION/COMMON_VOICE_8_0 - SL dataset. During training, a series of hyperparameters were used, including a learning rate of 7e - 05, a batch size of 32, and a linear learning rate scheduler with 1000 warm - up steps. The training lasted for 100 epochs, using Native AMP for mixed - precision training.

📄 License

This model is released under the Apache 2.0 license.