Kazakh Speech Commands Dataset Card
Repository: https://github.com/IS2AI/Kazakh-Speech-Commands-Dataset
Summary: This dataset contains Kazakh speech commands generated using a combination of synthetic speech generation and speech corpus scraping. Synthetic data was created using the Piper text-to-speech system, leveraging its Kazakh language models. Real-world speech commands were extracted from a large-scale speech corpus using the Vosk Speech Recognition Toolkit. Data augmentation techniques were applied to expand the dataset. The accompanying repository provides Jupyter notebooks detailing the data generation and augmentation processes, as well as information on model training, validation, and testing using a Keyword-MLP approach. Video tutorials are available on YouTube.
Dataset Details:
The dataset creation involved three main steps:
- Synthetic Speech Generation: Leveraged Piper, a neural text-to-speech system, to generate synthetic Kazakh speech commands. Five Kazakh voices were utilized from the Piper system.
- Speech Corpus Scraping: Utilized the Vosk Speech Recognition Toolkit to automatically extract speech commands from a large-scale speech corpus (the source of this corpus is not specified).
- Data Augmentation: Audio augmentation techniques were applied to both the synthetic and scraped datasets to increase the overall size and diversity of the data.
Tools and Technologies:
| Tool/Technology | Description | Link |
|---|---|---|
| Piper | Fast, local neural text-to-speech system | https://github.com/rhasspy/piper |
| Vosk Speech Recognition | Toolkit used for speech command extraction | https://github.com/alphacep/vosk-api/tree/master |
| Keyword-MLP | Model used for training, validation, and testing | https://github.com/IS2AI/Kazakh-Speech-Commands-Dataset/tree/main/Keyword-MLP |
Citation:
@article{Kuzdeuov2023,
author = "Askat Kuzdeuov and Shakhizat Nurgaliyev and Diana Turmakhan and Nurkhan Laiyk and Huseyin Atakan Varol",
title = "{Speech Command Recognition: Text-to-Speech and Speech Corpus Scraping Are All You Need}",
year = "2023",
month = "5",
url = "https://www.techrxiv.org/articles/preprint/Speech_Command_Recognition_Text-to-Speech_and_Speech_Corpus_Scraping_Are_All_You_Need/22717657",
doi = "10.36227/techrxiv.22717657.v1"
}