Dataset Viewer
Auto-converted to Parquet Duplicate
The dataset viewer is not available for this split.
Parquet error: Scan size limit exceeded: attempted to read 381843958 bytes, limit is 300000000 bytes Make sure that 1. the Parquet files contain a page index to enable random access without loading entire row groups2. otherwise use smaller row-group sizes when serializing the Parquet files
Error code:   TooBigContentError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Meddies ASR Small Test

Small Vietnamese golden-test slice for ASR evaluation, packaged as Parquet with embedded audio in a datasets.Audio column.

Configs

  • vi_faptv_human_gold: 20 FAPTV examples sourced from Meddies/meddies-asr-human-labels, clipped to the first 10 minutes of each original human-labeled episode.

Subset Status

  • vi_faptv_human_gold is directly scorable.
  • text and reference_text are identical and come from the clipped VTT transcript window.

Row Schema

column meaning
audio embedded 16 kHz mono WAV clip
text primary transcript field for scoring
reference_text explicit reference transcript
example_id stable benchmark row id
source_dataset, source_config provenance back to the source human-label dataset
title, youtube_video_id, youtube_url original episode metadata
clip_start_s, clip_end_s clipped time window inside the original source
original_duration_s full source episode duration before clipping

Build Notes

  • Selection is deterministic: the first 20 rows in local processed-audios-wav/manifest.csv with duration_seconds >= 600.
  • Each row uses the [0, 600) second window when the source is at least 10 minutes long.
  • Build metadata for this staged snapshot lives in metadata/build_summary.json.
Downloads last month
115

Collection including Meddies/meddies-asr-small-test