# Tatar Speech Commands Dataset

**Repository:** [https://github.com/IS2AI/TatarSCR](https://github.com/IS2AI/TatarSCR)


This repository contains an open-source Tatar speech commands dataset and associated resources for training and deploying a keyword spotting model.  The dataset is designed for use in IoT and robotics applications.


## Dataset Summary

The Tatar Speech Commands Dataset (TSCD) consists of 3,547 one-second utterances of 35 commands commonly used in robotics, IoT, and smart systems.  The data was collected from 153 different speakers.  The dataset was preprocessed and augmented using the provided notebook.  A Keyword-MLP model was trained on this dataset, and the code for training, testing, conversion to ONNX format, and inference (both PyTorch and ONNX) is included.

## Dataset Statistics

| Statistic           | Value      |
|-----------------------|-------------|
| Number of Commands   | 35          |
| Number of Utterances | 3,547       |
| Number of Speakers   | 153         |
| Audio Length         | 1 second    |


## Model Details

The provided code uses the Keyword-MLP model ([https://github.com/AI-Research-BD/Keyword-MLP](https://github.com/AI-Research-BD/Keyword-MLP)).  Both PyTorch and ONNX runtime inference scripts are included.

## Citation

To cite this dataset and model, please use the following information:

> Askat Kuzdeuov, Rinat Gilmullin, Bulat Khakimov, and Huseyin Atakan Varol. An Open-Source Tatar Speech Commands Dataset for IoT and Robotics Applications. TechRxiv. October 18, 2024, DOI: 10.36227/techrxiv.172926779.98914732/v1.