tiny-voca-ja / README.md

Commit History

Yesterday

  1. v2: fine-tune on pyopenjtalk-plus re-annotated data (step 114,000); 16-sentence eval normalized CER 1.6% -> 1.0%, UTMOS 3.73

    Verified signature
    0e57796

2 days ago

  1. Samples table: show sample 3's corrected score (0.0% after numeral normalization); keep the original 29.2% in the note and label the 4-sentence row's scoring method

    Verified signature
    5a21cc6
  2. Fix Evaluation intro: describe both the 4-sentence Whisper-small check and the 16-sentence large-v3 formal evaluation

    Verified signature
    d2ef91a
  3. Correct the s03 CER claim (scoring artifact) and add the formal 16-sentence evaluation

    Verified signature
    71753cd

3 days ago

  1. tiny-voca-ja: Japanese TTS model (4.5M params, retrained from the tiny-voca recipe)

    Verified signature
    cc743ee