Text → Speech

TTS

IndicParler-TTS — 22 Indian languages + English. Type, pick a voice (described or recorded), generate.

Language

Model

Damrooh-Indic-Dialects LoRA loads after fine-tune completes.

Text · 66/600

Voice description

Be specific: gender, pitch, tempo, mood. Damrooh's prompt-based control.

District preset (optional) · 0 loaded

Building presets in background — refresh in ~50 min.

Use a saved voice (optional)

Falls back if no preset is picked.

Long-form mode

Auto-splits text on `।`/`.` boundaries into ≤200-char chunks, generates each, and stitches with 200 ms gaps. Use for narration / paragraph-length copy.

First call after server restart loads the model (~90 s). Subsequent gens are ~10–15 s.

Record a reference voice

voice cloning

Record up to 30 s of speech. We extract a tone-color embedding (~15 s), then your TTS gens use OpenVoice timbre transfer over IndicParler — output sounds like you, in any of the 22 Indian languages.

Label

Saved voices · 0

No voices yet

Record one with the recorder on the left.

Queue

No speech yet

Type something and hit generate.