Uzbek Asr Curated 701H: Uzbek Speech Dataset for ASR

Uzbek speech data comprising approximately 701 hours of audio across 337,920 utterances, curated by uzinfocom-edu-ai. The dataset is formatted as 16 kHz mono WAV files with NeMo JSONL manifests and is split into training, validation, and test subsets.

Use Cases

• Train Uzbek speech recognition models based on the 701 hours of audio data.
• Evaluate ASR model performance on a dedicated Uzbek test split.
• Fine-tune multilingual speech models for the Uzbek language based on the curated multi-source data.
• Benchmark speech processing techniques on Uzbek audio using the provided train/val/test splits.

Strengths

• Large scale with 337,920 utterances.
• Substantial duration of approximately 701 hours of speech.
• Curated multi-source collection likely improving data diversity.
• Structured splits for machine learning (94% train, 3% val, 3% test).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• License information is unknown, which may restrict usage.
• Freshness should be verified as the last update timestamp is in the future (2026-06-26).

Provenance

Source

uzinfocom-edu-ai on Hugging Face

Collection Method

Curated from multiple sources, specific method unknown.

Freshness

Last updated 2026-06-26 06:19:06

Geography

Uzbek language focus, specific geographic coverage unknown.