Speech Emotion Recognition Wav2vec2 Large Xlsr 53 240304 SER Fine Tuned2.0

🚀 SER_wav2vec2-large-xlsr-53_240304_fine-tuned_2

This model is a fine - tuned version for speech emotion recognition, leveraging a pre - trained model on English datasets and showing cross - linguistic capabilities.

🚀 Quick Start

This model is a fine-tuned version of hughlan1214/SER_wav2vec2-large-xlsr-53_240304_fine-tuned1.1 on a Speech Emotion Recognition (en) dataset.

The dataset includes the 4 most popular English datasets: Crema, Ravdess, Savee, and Tess, with over 12,000 .wav audio files in total. Each of these four datasets has 6 to 8 different emotional labels.

On the evaluation set, it achieves the following results:

• Loss: 1.0601
• Accuracy: 0.6731
• Precision: 0.6761
• Recall: 0.6794
• F1: 0.6738

✨ Features

Model description

The model was obtained through feature extraction using facebook/wav2vec2-large-xlsr-53 and several rounds of fine - tuning. It predicts 7 types of emotions in speech, aiming to lay the foundation for real - time inference of user emotions using human micro - expressions on the visual level and context semantics under LLMS.

Although trained on purely English datasets, post - release testing showed that it performs well in predicting emotions in Chinese and French, demonstrating the powerful cross - linguistic capability of the facebook/wav2vec2-large-xlsr-53 pre - trained model.

emotions = ['angry', 'disgust', 'fear', 'happy', 'neutral', 'sad', 'surprise']

Intended uses & limitations

More information needed

📦 Installation

No installation steps provided in the original document.

💻 Usage Examples

No specific code examples for usage are provided in the original document.

📚 Documentation

Training and evaluation data

70/30 of the entire dataset was used for training and evaluation.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e - 05
• train_batch_size: 8
• eval_batch_size: 4
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon = 1e - 08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 10

Training results

Framework versions

• Transformers 4.38.1
• Pytorch 2.2.1
• Datasets 2.17.1
• Tokenizers 0.15.2

🔧 Technical Details

The model uses the facebook/wav2vec2-large-xlsr-53 pre - trained model for feature extraction and then fine - tunes it on the speech emotion recognition dataset. The cross - linguistic performance shows the generalization ability of the pre - trained model.

📄 License

The model is released under the Apache 2.0 license.