Mohit GedamCV
Lab

Hindi Text-to-Speech with XTTS-v2

A fine-tune of XTTS-v2 for Hindi speech on about 10 hours of audio, lowering eval loss by 12.7%.

  • Source codecoming soon
eval loss
−12.7%
clips · about 10 h of audio
4,586

Problem

Hindi speech synthesis with an open text-to-speech model.

What I built

A fine-tune of XTTS-v2 on roughly 10 hours of Hindi audio, split into 4,586 clips.

Results

  • Eval loss down 12.7%

Stack

XTTS-v2.