aqvaylis/kabyle-speech-to-text

Kabyle speech to text (torch model) # Kabyle Speech-to-Text Model (PyTorch) This project provides a speech recognition model (STT - Speech To Text) for the Kabyle language. The pre-trained model `best_kabyle_asr_optim.pt` was trained on more than 700,000 audio sentences with their textual transcriptions, from Common-voice and Tatoeba. This model was obtained after 30 Epochs, with the best version saved in the Epoch 20 with 'Val Loss= 0.1513'. The script saves the best model version according to the 'Val Loss' value and after ten times without improvement of this value, the script will stop and keep the last best version. ## 📋 Table of Contents - Overview - Requirements - Installation - Usage - Testing the Pre-trained Model - Custom Training - Recommended CUDA Configuration - Best Practices and Tips ## 📚 Overview The Kabyle speech recognition model has been designed to transform audio recordings in Kabyle into written text. The optimized model is available under the name `best_kabyle_asr_optim.pt`. > **Note**: Using a GPU is strongly recommended for optimal performance. ## 🖥️ Requirements - Python 3.8+ - PyTorch - CUDA-compatible GPU (strongly recommended) - Sufficient disk space for training data ## 🔧 Installation 1. Clone this repository: ```bash git clone github.com cd kabyle-speech2text ``` 2. Install the required dependencies: ```bash pip install -r requirements.txt ``` ## 🚀 Usage ### Local testing the Pre-trained Model 1. Run the test script: ```bash python kab_audio_model_test.py ``` 2. A Gradio interface will be generated - follow the link displayed in the console 3. Upload your MP3 audio file via the interface to get the transcription #### You can test the model on Hugging Face here . ### Custom Training To train the model with your own audio files: 1. Place your MP3 files in the `audios/` folder 2. Update the `transcription.csv` file with your information, following this format: You can check the current "audios" folder and …