Introduction

India’s voice-agent stack just got a sharper TTS option. On September 23, 2026, Mumbai-based Navana.ai unveiled Bodhi TTS and opened public API access—built for real Indian languages, not English-first models with a translation afterthought.

The pitch is blunt: more than 50 natively Indian voices across 10 languages, first audio in under 100 milliseconds, and a single list price of ₹12 per 10,000 characters (about ₹0.72 per minute of speech). Against ElevenLabs Multilingual/v3 list pricing (~₹95 / 10k chars) and Sarvam Bulbul v3 (₹30 / 10k), that is roughly one-eighth the global list and well under the nearest local comparator Navana cites.

Pro Tip

If you already cover Indic TTS in a creator or BFSI stack, keep Bodhi separate from open finetunes like Pravega/CosyVoice. This drop is a sovereign commercial API with on-prem path and India-native pronunciation control—not another community GGUF repack.

What Shipped Today

Navana’s homepage banner reads “NAVANA LAUNCHES STT & TTS APIs TODAY.” Trade coverage from CXOToday (Sep 23, ~18:51 IST) and the company’s platform page frame the same day as Bodhi TTS going live for signup, with ₹1,000 in free credits on account creation at platform.navana.ai.

Bodhi STT—speech recognition for 10 Indian languages and 40+ dialects—opens by invitation the same day, with benchmarks pointed at Hugging Face. Banks, insurers, and lenders can apply for early access to both models; Navana says the first 100 approved teams get 100 hours of usage credits valid for 90 days.

One caveat for builders: the marketing site still surfaces a waitlist CTA beside the launch banner, while the platform page shows a live countdown and “We’re live.” Treat signup as rolling public access—try the platform first, and expect enterprise gates on STT / on-prem.

Specs That Matter for Creators and Agents

Languages live at launch: Hindi, Telugu, Tamil, Marathi, Bengali, Odia, Malayalam, Kannada, English, and Gujarati.

Voices are tuned for phone-conversation delivery rather than studio narration. Every voice sits on the same price—no premium tier. Features called out as standard (not add-ons):

  • Sub-100 ms time-to-first-audio for streaming responses
  • Zero-shot cloning from a few seconds of reference audio, in all 10 languages, without a redeploy
  • Sound-level pronunciation control—upload BFSI dictionaries (product names, branches, customer names) or mark names inline so the model obeys them
  • Native reading of Indian text quirks: ₹1,00,000 → “one lakh rupees”, plus PAN, phone numbers, dates, EMIs, and scheme names the way a bank would say them

Navana also claims Bodhi is about 5× smaller than the nearest comparable architecture, which is how they argue mid-tier on-prem GPUs and regulated-data residency stay realistic.

Sovereign Stack, Not Just “Hosted in India”

Bodhi is positioned as sovereign by architecture: Navana trains and owns the speech stack end-to-end, with no customer audio routed through a third-party model provider. That story lines up with ISO 27001 / SOC 2 Type II certification and BFSI clients already running high-volume voice agents (Navana cites ₹1,000+ crore in monthly loan disbursals across its banking footprint, plus logos like Bajaj Finserv, Ujjivan, and Jana).

Co-founder & CEO Raoul Nanavati framed the gap as language plus price: global models that mangled Marathi/Tamil everyday words, at prices that only worked for pilots. Co-founder Jai Nanavati pushed the size angle—quality for Indian speech does not automatically scale with parameter count if the model cannot live inside the bank’s own data centre.

How This Differs from Pravega (and Why We Covered Both)

ArtRealmAI already covered Dheeyantra Pravega—an open CosyVoice-based Indic TTS finetune path for builders who want weights and self-hosting DIY. Bodhi is a different shape: managed API + enterprise on-prem, India-native pronunciation/dictionary tooling, and aggressive list pricing aimed at call-centre volumes.

Use Pravega when you want open weights and research/community iteration. Reach for Bodhi when you need latency, clone-from-seconds, and regulated-data packaging without assembling the stack yourself.

Try Path / What to Watch

  1. Sign up at platform.navana.ai and burn the ₹1,000 credits on your target languages (especially code-mixed Hindi–English and Tamil/Telugu agent scripts).
  2. Stress-test numbers, PANs, and product names—that is the differentiator Bodhi markets hardest.
  3. If you need STT too, apply for the invitation track; do not assume TTS signup unlocks recognition.
  4. Compare total cost against Sarvam Bulbul and your current ElevenLabs Multilingual usage at your character volumes, not just list rates.
  5. For on-prem / sovereign RFPs, ask for the mid-tier GPU footprint and data-residency runbook early—those claims are central to the pitch.

Docs surfaces to bookmark: docs.dev.navana.ai and the Bodhi GitBook at navana.gitbook.io/bodhi.

Conclusion

Bodhi TTS is a real Sep 23 product drop for Indic voice: ten languages, fifty-plus native voices, sub-100 ms first audio, zero-shot clone, and a ₹12 / 10k-character list price aimed at India’s call volumes. Pair that with a sovereign/on-prem story and India-native pronunciation control, and it is more than a press-release clone of global TTS APIs. STT remains invite-only, and waitlist vs platform messaging is still a little fuzzy—so verify access before you promise a ship date—but the TTS door is open today.

Primary: Navana.ai, platform.navana.ai, CXOToday launch coverage.

—Anabel ♡