Bodhan AIIndic-Speak
Indic-Speak
Indic-Speak turns text into speech in English and 22 Indian languages. Bodhan and AI4Bharat emphasize educational material and sentences that mix English with an Indian language.
Restricted access
The voice library and decoder affect output
The card requires a speaker name used in training. An unknown name can produce a weaker averaged voice without an error. The model focuses on explanatory speech, including equations, units and code-mixed lessons. It does not generate laughter, sighs or other sound effects.Indic-Speak official model card
Review the text that will actually be read aloud
Practical guidance. A lesson can contain unfamiliar names, equations and acronyms that ordinary prose does not cover. Review those items in the selected voice before using long-form generated audio.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Restricted access |
| Access | Downloadable, API |
| Model identifiers | indic-speak |
| Ancestry | Llama-3.2-3B speech language model; SNAC 24 kHz audio codes and a fine-tuned Vocos decoder.Indic-Speak official model card |
| Voice selection | 95 listed voices from consenting, contracted artists. Voice cloning is not supported.Indic-Speak official model card |
| Style | Emotion and descriptive style controls are preview quality.Indic-Speak official model card |
| Access and terms | Gated weights under Indic Open Model License v1.0; upstream terms also apply.Indic Open Model License v1.0 |
| Hosted access | Bodhan documents an authenticated API operation for this model.Bodhan API reference |