Kazakh Speech MFA Punctuation: 1,110 Hours of Annotated Audio

Kazakhstan's largest open Kazakh speech corpus, an extended version of the ISSAI KSC2 dataset from Nazarbayev University. It contains approximately 1,110 hours of audio across 595,690 recordings, enhanced with punctuation and word-level timestamps from MFA alignment. The dataset is published and maintained by Jeti Labs.

Use Cases

• Train automatic speech recognition (ASR) models based on the 1,110 hours of Kazakh audio.
• Develop punctuation restoration models for transcribed speech based on the added punctuation annotations.
• Build text-to-speech alignment systems using the word-level temporal metadata (MFA alignment).
• Conduct linguistic research on Kazakh phonetics and prosody based on the aligned speech corpus.

Strengths

• Contains 595,690 audio recordings, providing a substantial volume of data.
• Offers approximately 1,110 hours of speech, making it a large-scale resource for a low-resource language.
• Enhanced with punctuation and word-level timestamps, adding valuable structure beyond raw audio.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Extended from the ISSAI KSC2 corpus by Nazarbayev University, published by Jeti Labs.

Collection Method

Likely derived from a speech corpus with automated alignment (MFA) and punctuation annotation.

Freshness

Last updated 2026-04-16 05:37:14; freshness should be verified.

Geography

Kazakhstan (implied by Kazakh language focus).