MANGO TTS

About Dataset

Key Features:

• 255,150 human ratings of TTS-generated outputs and ground-truth human speech.
• Covers two major Indian languages: Hindi & Tamil, and English.
• Based on the MUSHRA (Multiple Stimuli with Hidden Reference and Anchor) test methodology.
• Ratings are provided on a continuous scale from 0 to 100, with discrete quality categories:
  • 100-80: Excellent
  • 80-60: Good
  • 60-40: Fair
  • 40-20: Poor
  • 20-0: Bad
• Includes evaluations involving:
  • MUSHRA: with explicitly mentioned high-quality references.
  • MUSHRA-NMR: without explicitly mentioned high-quality references.
  • MUSHRA-DG: with detailed guidelines across fine-grained dimensions
  • MUSHRA-DG-NMR: with detailed guidelines across fine-grained dimensions and without explicitly mentioned high-quality references.

Available Splits

The dataset includes the following splits based on the test type and language.