Kazakh Speech Corpus 2: An Industrial-Scale Open-Source Audio Dataset

1.2k hours of high-quality Kazakh speech audio form the KSC2 corpus. The dataset, created by author 'issai', consolidates and supplements previous Kazakh speech and text-to-speech corpora with data from sources like TV programs, radio, senate sessions, and podcasts. It was last updated on the Hugging Face platform on February 28, 2025.

Use Cases

• Train automatic speech recognition (ASR) models based on the described high-quality audio.
• Develop text-to-speech (TTS) systems based on the consolidation of previous TTS corpora.
• Conduct linguistic analysis of Kazakh using diverse audio sources like TV, radio, and podcasts.
• Benchmark audio processing models on an industrial-scale, open-source Kazakh dataset.

Strengths

• Industrial-scale size of approximately 1,200 hours of audio.
• High-quality audio content, as explicitly stated in the description.
• Consolidates multiple previous Kazakh speech datasets and supplements them with diverse sources.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and exact license are unknown, which may limit suitability assessment.

Provenance

Source

Hugging Face dataset by author 'issai'.

Collection Method

Consolidates previous Kazakh speech and TTS corpora and supplements data from TV programs, radio, senate, and podcasts.

Time Range

null

Freshness

Last updated 2025-02-28 13:05:39; freshness should be verified.

Geography

Likely focused on Kazakhstan, given the Kazakh language content.