Instructions to use CUAIStudents/Adapt-Ar with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use CUAIStudents/Adapt-Ar with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("automatic-speech-recognition", model="CUAIStudents/Adapt-Ar")# Load model directly from transformers import AutoProcessor, AutoModelForSpeechSeq2Seq processor = AutoProcessor.from_pretrained("CUAIStudents/Adapt-Ar") model = AutoModelForSpeechSeq2Seq.from_pretrained("CUAIStudents/Adapt-Ar", device_map="auto") - Notebooks
- Google Colab
- Kaggle
| library_name: transformers | |
| tags: [] | |
| # Adapt-Ar | |
| This model is adapted from the **[DeepAr](https://huggingface.co/CUAIStudents/DeepAr)** model. While structurally identical, it was trained for additional epochs on augmented data to improve robustness against noisy audio, silence, and variations introduced by augmentation. | |
| The purpose of this model is to address the main limitations of the original **[Whisper-Large-v3-Turbo](https://huggingface.co/openai/whisper-large-v3-turbo)** model: | |
| * Silent or low-volume segments → often caused the model to hallucinate text. | |
| * Noisy environments and variable speech speed → reduced transcription accuracy and stability. | |
| For details on model usage and the dataset, please refer to **[DeepAr](https://huggingface.co/CUAIStudents/DeepAr)**. **Adapt-Ar** is architecturally identical to DeepAR, but trained for additional half an epoch on augmented data to improve robustness. | |
| Both models share the same usage; the only difference lies in the training process and naming. | |