yuekai/

speechio_test

Parquet repackaging of the SpeechColab SpeechIO Mandarin ASR benchmark, re-exported from yuekai/speechio (Lhotse cuts) into standard HuggingFace parquet with embedded 16 kHz audio.

27 test sets: SPEECHIO_ASR_ZH00000 ... SPEECHIO_ASR_ZH00026, each a config with a single test split. ~43k utterances, ~66 hours total, evaluation only.

	Columns




	

column type note

segment_id string utterance id

speaker string speaker id… See the full description on the dataset page: https://huggingface.co/datasets/yuekai/speechio_test.