Wav2vec2 Xlsr Estonian

🚀 sammy786/wav2vec2-xlsr-estonian

This model is a fine - tuned version of facebook/wav2vec2-xls-r-1b on the MOZILLA - FOUNDATION/COMMON_VOICE_8_0 - et dataset. It offers a solution for automatic speech recognition tasks, achieving specific results on evaluation sets.

🚀 Quick Start

This model is a fine - tuned version of facebook/wav2vec2-xls-r-1b on the MOZILLA - FOUNDATION/COMMON_VOICE_8_0 - et dataset. It achieves the following results on the evaluation set (which is 10 percent of the train dataset merged with other and dev datasets):

• Loss: 17.94
• Wer: 30.38

✨ Features

• Fine - Tuned Model: Based on facebook/wav2vec2-xls-r-1b, fine - tuned on the MOZILLA - FOUNDATION/COMMON_VOICE_8_0 - et dataset.
• Multiple Datasets Support: Can be evaluated on different datasets like mozilla - foundation/common_voice_8_0 and speech - recognition - community - v2 datasets.

📚 Documentation

Model description

"facebook/wav2vec2-xls-r-1b" was finetuned.

Intended uses & limitations

More information needed

Training and evaluation data

Training data - Common voice Finnish train.tsv, dev.tsv and other.tsv

Training procedure

For creating the train dataset, all possible datasets were appended and a 90 - 10 split was used.

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.000045637994662983496
• train_batch_size: 8
• eval_batch_size: 16
• seed: 13
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon = 1e - 08
• lr_scheduler_type: cosine_with_restarts
• lr_scheduler_warmup_steps: 500
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Framework versions

• Transformers 4.16.0.dev0
• Pytorch 1.10.0+cu102
• Datasets 1.17.1.dev0
• Tokenizers 0.10.3

Evaluation Commands

1. To evaluate on mozilla - foundation/common_voice_8_0 with split test

python eval.py --model_id sammy786/wav2vec2-xlsr-estonian --dataset mozilla - foundation/common_voice_8_0 --config et --split test

🔧 Technical Details

The model is based on the pre - trained "facebook/wav2vec2-xls-r-1b" and fine - tuned on the MOZILLA - FOUNDATION/COMMON_VOICE_8_0 - et dataset. During training, various hyperparameters are carefully selected to optimize the performance. A 90 - 10 split of the datasets is used for training and evaluation.

📄 License

This project is licensed under the Apache - 2.0 license.