Skip to content

Golos

~1,240 h of manually annotated open Russian speech split between crowd-sourced (~1,106 h) and farfield/smart-device (~134 h) recordings.

Recommendation

A strong default for Russian ASR training and benchmarking, especially for farfield/voice-assistant conditions alongside clean crowd speech. Licensing is a custom Sber document (not a standard identifier) — review before commercial use.

Getting the data

Obtain from the dataset homepage.

License is the Sber document in the repo; also mirrored on OpenSLR #114 and HuggingFace (SberDevices/Golos).

Suggested processing

A recommended VoxKitchen pipeline ships in the repository at voxkitchen/templates/pipelines/asr-training-data.yaml — run it with vkit docker run.