Kazakh Speech Dataset: ~726 Hours of Audio for ASR and TTS

Kazakh language speech data comprising approximately 726 hours of audio in FLAC format at a 16kHz sampling rate. The dataset is designed to support Automatic Speech Recognition and Text-to-Speech system development. It is an open-source corpus created by Flamme-VRM and was last updated on Hugging Face in January 2026.

Use Cases

• Train Automatic Speech Recognition (ASR) models based on the described Kazakh speech corpus.
• Develop Text-to-Speech (TTS) systems based on the described Kazakh audio data.
• Benchmark speech model performance for the Kazakh language based on the described large-scale corpus.

Strengths

• Approximately 726 hours of total audio duration provides a substantial resource.
• Audio is provided in a standard lossless format (FLAC) at a 16kHz sampling rate.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and total file size are unknown, which may limit suitability assessment.
• The description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Flamme-VRM

Freshness

Last updated 2026-01-07 14:09:49; freshness should be verified.

Geography

Kazakh language content, likely from Kazakhstan or Kazakh-speaking regions.