Voice AI infrastructure in India: TTS, STT, LLM and agent APIs.

What are the four layers?

Speech to text turns the caller’s voice into words. A language model decides what to say. Text to speech turns the reply back into audio. The agent platform runs everything around them: the phone line, turn-taking, interruptions, tools, and what happens after the call.

Buy each layer separately and you pick every piece. Buy a bundle and you get one bill and fewer hops between the caller and the reply. Two questions narrow the list fast. Does it speak the way your callers speak? And where does the audio go?

Who sells each layer?

Here’s what each company published on the pages we read on 16 October 2026. The text-to-speech and speech-to-text comparisons have every price, each linked to its source.

Layer	Who sells it	How it is priced
Text to speech	Frequency (Fourier TTS), Sarvam (Bulbul), ElevenLabs, Google Cloud, Azure, Amazon Polly, OpenAI, Cartesia	Per character for most; per token for OpenAI’s and Google’s newest models; monthly credits on Cartesia
Speech to text	Frequency (Fourier Transcribe for recordings), Sarvam (Saaras), ElevenLabs (Scribe), Deepgram (Nova-3), Google Cloud, Azure, Amazon Transcribe, OpenAI	Per minute or per hour of audio
Language model	Frequency, Sarvam (Sarvam 105B and hosted open models), OpenAI, Google, Azure	Per million input and output tokens
Voice agents	Frequency, Sarvam Voice Agents, ElevenLabs, Deepgram Voice Agent API, Cartesia Managed Agents, OpenAI GPT-Live, and agent platforms such as Vapi, Retell AI, Bland AI and Bolna	Per minute of call, with telephony and model usage included or billed on top depending on the vendor

The per-minute agent prices we read the same day:

Agent product	Published price
Sarvam Voice Agents	“As low as ₹3 a minute”, models included
ElevenLabs Speech Engine	$0.08 a minute
Deepgram Voice Agent API	$0.075 a minute, standard pay-as-you-go tier
Cartesia	$0.06 a minute, plus $0.014 a minute for a Cartesia phone number
OpenAI GPT-Live	$0.05 a minute, with backend model and tool usage charged separately

Our platform comparison lists Vapi, Retell AI, Bland AI, Bolna, Gnani.ai and Skit.ai as each published them on 8 October 2026.

Who says the audio stays in India?

This is where vendors’ pages differ most. Storing data in a region isn’t the same as processing it there, so read the exact words.

Provider	What its own pages say
Frequency	Calls run on our own GPUs in India; our terms name the backup providers and where recordings are stored
Azure Speech	Central India supports neural text to speech and real-time, fast and batch transcription
Amazon Polly	Neural voices run in Asia Pacific (Mumbai), among other regions
OpenAI	The India region lists storage, not processing, for the audio endpoints, and needs approval through sales
ElevenLabs	An India isolated environment for storage, for Enterprise customers; processing may still happen outside it
Google Cloud	TTS endpoints are global, US and EU, with Chirp 3: HD marked for Tokyo and Singapore and Mumbai listed without a model; Hindi speech to text is listed in US, EU and Singapore regions
Deepgram	A dedicated EU endpoint; the pricing page names no India region
Cartesia	The pricing page names no hosting region
Sarvam	The pricing and model pages we read do not name a hosting region

If your compliance team needs a firm answer, ask each vendor in writing, and name the exact API you’ll call.

Where does Frequency fit?

We sell all four layers on our own models, from one rupee wallet with one sk_live_ key.

What you buy	Price, before GST
Phone agents on T1, the volume plan	₹1 a minute, with a ₹1,00,000 monthly commitment
Phone agents on T3	₹3 a minute, with a ₹50,000 monthly commitment
Phone agents on T5	₹5 a minute, pay as you go
Fourier TTS	₹1.80 per 1,000 characters
Transcription of recorded calls	₹18 an hour of audio
LLM	₹15 input, ₹9.30 cached input and ₹55 output per 1,000,000 tokens

Agent calls bill in 10-second blocks. T5 and T3 include Mirai phone calling, and your own telco, WhatsApp or browser calling works on every plan.

Fourier TTS, our own voice, is sold on its own through OpenAI’s speech protocol. Fourier STT, our own speech recognition, runs inside every agent; recorded calls go through the public API today, and streaming is in private preview. The LLM allows 1,000 requests a minute on pay as you go once we switch it on, and a standalone endpoint is on the roadmap.

Two gaps to plan around. We speak Hindi, English, Gujarati, Tamil and Telugu today, so Bengali, Marathi, Kannada and other Indian languages still need someone else. And transfer to a person isn’t available yet. If you need either now, the companies above cover them.

How should I choose a provider?

Start from the calls you’ll make. If they’re Hindi with English mixed in, shortlist the vendors that say they handle it, then test them on your own recordings and scripts. Voice Arena’s leaderboards and our benchmark tables are a place to start, not a verdict.

If you already run an orchestrator like Pipecat, buy the layers. If you want calls going out this week, an agent platform is less work. Either way, check the data location for the exact API, and price the whole minute, phone line and model included.