issai/Tatar_Speech_Commands_Dataset

TatarSCR: An Open-Source Tatar Speech Commands Dataset

This repository contains an open-source Tatar speech commands dataset designed for IoT and robotics applications, along with a trained Keyword-MLP model for inference. The dataset and model are described in the following paper:

Dataset Summary

This dataset comprises 3,547 one-second utterances of 35 commands commonly used in robotics, IoT, and smart systems. The data was collected from 153 different speakers.

Feature	Description
Number of Commands	35
Number of Utterances	3,547
Number of Speakers	153
Audio Length	1 second per utterance

Model Summary

A Keyword-MLP model was trained on this dataset. Pre-trained weights are provided, allowing for direct inference using either PyTorch or ONNX runtime. The model was adapted from the Keyword-MLP repository (https://github.com/AI-Research-BD/Keyword-MLP). Inference scripts are provided for both short (~1 second) audio clips and longer clips using a sliding window approach.

Citation

If you use this dataset or model, please cite the following paper:

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.

Repository Link

https://github.com/IS2AI/TatarSCR