Upload README.md with huggingface_hub

README.md

@@ -0,0 +1,23 @@
+# Tatar Speech Commands Dataset
+
+This dataset contains 3,547 one-second utterances of 35 commands commonly used in robotics, IoT, and smart systems.  The data was collected from 153 speakers.
+
+## Dataset Statistics
+
+* **Number of commands:** 35
+* **Number of utterances:** 3,547
+* **Number of speakers:** 153
+* **Audio length:** 1 second per utterance
+
+## Data Download
+
+The dataset can be downloaded from [Google Drive](https://drive.google.com/file/d/1CBmVeAYgNrkNKhL1wtG7KUKuLJ9hOfHL/view?usp=sharing).
+
+## Related Work
+
+The provided Keyword-MLP model ([https://github.com/AI-Research-BD/Keyword-MLP](https://github.com/AI-Research-BD/Keyword-MLP)) was used for training and testing on this dataset.  The TatarSCR repository ([https://github.com/IS2AI/TatarSCR.git](https://github.com/IS2AI/TatarSCR.git)) contains the code and configurations used in this work.  Preprocessing and augmentation can be performed using the `data_preprocessing_augmentation.ipynb` notebook, which requires the ESC-50 dataset ([https://github.com/karolpiczak/ESC-50](https://github.com/karolpiczak/ESC-50)).
+
+
+## Model Inference
+
+Inference can be performed using PyTorch or ONNX.  PyTorch offers two scripts: `inference.py` for short audio clips and `window_inference.py` for longer clips using a sliding window approach.  ONNX inference is handled by `onnx_inference.py`.  The `label_map.json` file is required for inference.