Dataset Viewer
Auto-converted to Parquet Duplicate
audio_id
stringclasses
3 values
whisper_count
int64
3.51k
6.84k
gemini_count
int64
0
0
rel_diff
float64
1
1
tolerance
float64
0.05
0.05
matched
bool
1 class
klar__155598
3,512
0
1
0.05
true
klar__155436
3,794
0
1
0.05
true
klar__154666
6,835
0
1
0.05
true

Yiddish TTS Aligned v0

Derived Yiddish text-to-speech training pairs produced by the Yiddish TTS pipeline: matched audio clips, dual (Whisper + Gemini) transcripts, and American-Hasidic-Yiddish IPA phonemes. The corpus is a formatted derivative of Kohn-AI/yiddish24-wav; raw source URLs are intentionally excluded.

Layout

  • audio/ — matched clip WAVs (16 kHz mono PCM).
  • transcripts/whisper.jsonl, gemini.jsonl, matched.jsonl.
  • phonemes/ — one {audio_id}.ipa.txt per clip (plain IPA).

Provenance & audit metadata

field value
git-sha 1587352df3b4779dbb83e88db0010949ebd8bbba
pipeline_run_id m2-data-20260716171657
tolerance 0.05
matched_count 3
total_count 3
Downloads last month
29