metadata
language:
- ja
configs:
- config_name: default
data_files:
- split: train
path: all.parquet
japanese-asr/whisper_transcriptions.reazon_speech_all without audio
align.tsv with turbo
segment.tsv split ~20 chars segments with timestamps, filtered ir < 0.1 & cer < 0.5