Dataset Viewer (First 5GB)
Auto-converted to Parquet Duplicate
Search is not available for this dataset
The dataset viewer is not available for this split.
Server error while post-processing the rows. Please report the issue.
Error code:   RowsPostProcessingError

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.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Emotion Voice References for QwenTTS

Emotion-wise organized voice WAV files for Hindi (Male + Female) and English (Male + Female), ready to use as reference audio for Qwen3-TTS voice cloning.

🎯 Purpose

Reference audio clips for Qwen3-TTS voice cloning with emotion-aware synthesis. Organized consistently as {language}/{gender}/{emotion}/*.wav.

πŸ“ Structure

hindi/
  male/{emotion}/       β†’ 4,895 WAVs (HIN_M_* from Rasa Hindi + URD_M_* from Rasa Urdu)
  female/{emotion}/     β†’ 2,327 WAVs (URD_F_* from Rasa Urdu)

english/
  male/{emotion}/       β†’ 720 WAVs (RAVDESS, 12 actors)
  female/{emotion}/     β†’ 720 WAVs (RAVDESS, 12 actors)

hindi_best_clips/
  male/{emotion}/       β†’ 3 curated ~5s clips per emotion
  female/{emotion}/     β†’ 3 curated ~5s clips per emotion

english_best_clips/
  male/{emotion}/       β†’ 3 curated ~5s clips per emotion
  female/{emotion}/     β†’ 3 curated ~5s clips per emotion

πŸ“Š Statistics

Hindi Male

Emotion Files
anger 839
disgust 815
fear 801
happy 857
sad 745
surprise 838
Total 4,895

Sources: AI4Bharat Rasa Hindi (HIN_M) + Rasa Urdu (URD_M) β€’ 48kHz WAV β€’ CC BY 4.0

Hindi Female

Emotion Files
anger 395
disgust 417
fear 343
happy 401
sad 384
surprise 387
Total 2,327

Source: AI4Bharat Rasa Urdu Female (URD_F) β€’ 48kHz WAV β€’ CC BY 4.0

English Male

Emotion Files
neutral 48
calm 96
happy 96
sad 96
angry 96
fearful 96
disgust 96
surprised 96
Total 720

English Female

Emotion Files
neutral 48
calm 96
happy 96
sad 96
angry 96
fearful 96
disgust 96
surprised 96
Total 720

Source: RAVDESS (12M + 12F actors) β€’ 48kHz WAV β€’ CC BY-NC-SA 4.0

Grand Total: 8,662 emotion-tagged WAV files

πŸš€ Quick Usage with QwenTTS

from huggingface_hub import hf_hub_download

REPO = "sumedhu/hindi-emotion-voice-references"

# Hindi Male - happy
hf_hub_download(repo_id=REPO, repo_type="dataset",
    filename="hindi_best_clips/male/happy/HIN_M_HAPPY_00057.wav")

# Hindi Female - angry
hf_hub_download(repo_id=REPO, repo_type="dataset",
    filename="hindi_best_clips/female/anger/URD_F_ANGER_00237.wav")

# English Male - sad
hf_hub_download(repo_id=REPO, repo_type="dataset",
    filename="english_best_clips/male/sad/03-01-04-01-01-02-21.wav")

# English Female - surprised
hf_hub_download(repo_id=REPO, repo_type="dataset",
    filename="english_best_clips/female/surprised/03-01-08-02-01-01-14.wav")

πŸ—ΊοΈ Emotion Coverage Matrix

Emotion Hindi Male Hindi Female English Male English Female
happy βœ… 857 βœ… 401 βœ… 96 βœ… 96
sad βœ… 745 βœ… 384 βœ… 96 βœ… 96
anger/angry βœ… 839 βœ… 395 βœ… 96 βœ… 96
fear/fearful βœ… 801 βœ… 343 βœ… 96 βœ… 96
disgust βœ… 815 βœ… 417 βœ… 96 βœ… 96
surprise/surprised βœ… 838 βœ… 387 βœ… 96 βœ… 96
neutral ❌ ❌ βœ… 48 βœ… 48
calm ❌ ❌ βœ… 96 βœ… 96

⚠️ Notes

  • Best clips are ~5 seconds each (ideal for QwenTTS reference: 3-10s)
  • All clips are single-emotion per file
  • The reference audio IS the emotion carrier β€” QwenTTS clones both voice and emotion
  • Hindi Female uses Rasa Urdu Female speaker (nearly identical phonology to Hindi)
  • Hindi Male combines Rasa Hindi Male + Rasa Urdu Male speakers
  • English clips have 2 intensity levels (normal + strong) per emotion

License

  • Hindi clips: CC BY 4.0 (AI4Bharat Rasa)
  • English clips: CC BY-NC-SA 4.0 (RAVDESS)
Downloads last month
46