Skip to content
Fan Zeng (Frank)
Back to resources

LibriCSS

LibriCSS records overlapping playback of LibriSpeech utterances through a seven-channel array in a meeting room.

Visit official source (opens in a new tab)

Research use

Measure the effect of overlap and speaker distance on streaming transcription and compare mono and multichannel inputs.

Limitations

The benchmark uses replayed read speech. The original turns were arranged for controlled overlap conditions.

License and access

Dataset license
Unconfirmed for this release
Access status
Direct access

Download the public ZIP from the authors’ Google Drive. The project repository documents preparation and scoring.

Additional sources