Record
Thirty minutes of clean studio audio in the voice you want, read from our script in Hindi, English (US) and Hinglish. We help with the script and the session.
आधे घंटे की recording, बस.
First word in 27 milliseconds.
One stream, server-side: the first audio frame leaves the server 27 ms after the text arrives, and a minute of speech is made in eight seconds. Independent end-to-end measurements of rented voices land between 190 and 420 ms.
- 27 ms first audio
- 0.14 real-time factor
- Server-side, no queue
Rented voices cost about 2.5 cents a minute. Ours: a tenth of a cent per 1,000 characters, falling to a hundredth.
A rented voice is rented compute plus a vendor margin: 2.5 cents (₹2) or more a minute. Ours is 10 paise, about a tenth of a US cent, per 1,000 characters today, about a minute of speech; half that in six months and under a paisa, about a hundredth of a cent, in about a year, because the model is ours and it runs on ordinary servers.
- 10pper 1,000 characters, today
- 5pin six months
- <1pin about a year
Hindi, English (US), English (IN), Hinglish. Then the world's languages.
Trained on how people actually talk: switching languages mid-sentence, numbers, amounts and dates read the local way, codes and acronyms spelled out. More languages are in training now.
- हिंदी
- English (US)
- English (IN)
- Hinglish
- தமிழ்
- తెలుగు
- मराठी
- বাংলা
- ಕನ್ನಡ
No concurrency limit. Ever.
No queue and no scarce hardware: one four-core server makes 18 seconds of speech every second, and a server adds capacity in minutes, in our cloud or your data centre. The guard rails live inside the model: it speaks exactly the text it is given, reads names and amounts right, and stops cleanly at the end of the sentence.
- No cap on streams
- 18× real time per 4 cores
- Any server
- Your DC or ours
- Guard rails in the model
Give us the voice. We train the model for you.
Record your brand voice, an agent, an anchor, a founder, and we return a model trained on it: your timbre, your pronunciations, your pace. Same price, same speed, yours to run wherever you like.
- 30 min of recordings
- Your model
- Same price
- Runs anywhere
Five convictions
Vartaloop TTS · live · from 10 paise per 1,000 characters
Our first product: voices that start speaking before the sentence ends.
Natural, expressive Hindi, English (US), English (IN) and Hinglish, streamed from a plain API. Names, numbers and currency amounts read the way a person reads them. A tenth of a cent per 1,000 characters, no concurrency limit. Our own model on our own engine, so it is fast on a live call and cheap enough to put behind every message you send.
- 01Fast first audio in 27 ms server-side, 7× faster than real time
- 02Studio quality 24 kHz, code-switching mid-sentence, pace you set from 0.5× to 2×
- 03Reads it right $12.50 · ₹4,250 · 12 अक्टूबर · OTP digits · EMI, KYC
- 04Your voice record it, we train the model for you, same price
- 05No limits no concurrency cap, add a server, add capacity
10pper 1,000 characters, today
27 msfirst audio, server-side
0.14real-time factor
Vartaloop TTSStudioStreaming
नमस्ते, electricity repair के लिए आज सुबह बात हुई थी। Nivedita Xpress Distribution Company के case में मेरे अलावा alternate contact सिद्धार्थ भट्टाचार्य हैं। Callback +44 00000 19024 पर करें। Support app का version v7.07.1 है; processing estimate ₹17.99 है। इसे payment approval न मानें। मैं इन details की confirmation कर रही हूँ। आपने मेरा phone number groups में लिखने को कहा था; अब सही number है: +44 00000 19024। पहले वाला number गलत था; ऊपर लिखे सही number का हर digit जरूरी है। कृपया पुराने number पर call न करें, क्योंकि वह पड़ोसी का है और वे घर पर नहीं होंगे। शाम 6:30 के बाद मैं available रहूँगी। Meter की photo पहले ही भेज दी है, इसलिए काम शुरू करने से पहले दोबारा आने की जरूरत नहीं है। कृपया आने का estimated time message में बता दें। Reply के साथ +44 00000 19024 जोड़ दें। इस हिस्से के बिना confirmation incomplete रहेगी और फिर से check करना होगा।
“Hello, we spoke this morning about the electricity repair. For the Nivedita Xpress Distribution Company case, the alternate contact besides me is Siddharth Bhattacharya. Call back on +44 00000 19024. The support app is version v7.07.1; the processing estimate is ₹17.99. Do not treat this as payment approval. I am confirming these details. You asked me to write my phone number in groups; the correct number is +44 00000 19024. The earlier number was wrong; every digit of the number above matters. Please do not call the old number: it belongs to a neighbour and they will not be home. I will be available after 6:30 pm. I have already sent the meter photo, so there is no need to visit again before starting the work. Please message the estimated arrival time. Add +44 00000 19024 to the reply. Without that part the confirmation stays incomplete and will have to be checked again.”
Real outputAditya · English (IN)Production voice · 47 voices live, your own on request
Unedited model output, 24 kHz. Forty-seven production voices are live: twenty-one Hinglish — Saanvi and Sahil first, each with three expressions (excited, sad, whisper), and nineteen new voices from the deepest to the brightest register — eighteen English (US), nine female and nine male, Kiara and Kevin first among them; and eight English (IN), four female and four male, Ananya and Arjun first among them. Pick a language and a voice above to hear each one; your own voice trains on the same engine at the same price.
The numbers · one stream · measured, not promised
Server-side, one stream, ordinary servers.
Everything below was measured on 17 October 2026 on one small cloud server, engine-side, before the network. The rented-voice figures are independent end-to-end leaderboard medians from the same month. We publish the method with the numbers.
27 mstime to first audio, one stream
0.14real-time factor: a minute of speech in 8 s
51 msfirst audio with four streams on the same small server, each at 0.22 of real time
24 kHzstudio sample rate, streamed in 80 ms frames
Time to first audio
lower is better · ours engine-side, one stream · theirs independent, end-to-end, Oct 2026Price per minute of speech
list prices, October 2026, converted at ₹84 a dollar · ours is per 1,000 characters, about a minute of speechHow: one synthesis stream on one small cloud server, our engine, 24 kHz output, eight Hinglish sentences, medians. First audio is the time from text in to the first 80 ms frame out. Real-time factor is wall time divided by audio produced. Four streams on the same four-core server, one process sharing one copy of the model: 51 ms first audio, each stream at 0.22 of real time, 18 seconds of speech made every second, no gaps.
Your voice · trained for you
Give us the voice. We give you the model.
Our stock voices are good. Yours is better for your brand: the agent your customers already know, the anchor from your ads, the founder. Record it once; we train a model on it and hand it back at the same price and the same speed, to run in our region or on your own servers.
We train
The model learns the timbre, the pace and the pronunciations of that voice while keeping everything it already knows about names, numbers and code-switching.
Two to three weeks, end to end.
You ship
Your voice behind the same API at the same 10 paise per 1,000 characters, in our cloud or inside your data centre. The model is yours.
Same price. Same speed. Your voice.
Your model, yours to runData stays in your regionConsent on recordSame 10p per 1,000 characters
Talk about your voice
Pricing · in paise · per 1,000 characters
Three plans. Every one only goes down.
We own the model and it runs on ordinary servers, so the price is arithmetic and not a negotiation. Per 1,000 characters of text you send, which is about a minute of speech, about a tenth of a US cent, billed in rupees (₹100 ≈ $1.20 per million characters) with a tax invoice. No per-seat licence, no setup fee, no card needed to start.
Vartaloop TTSStreaming API
The voice, streamed.
10pper 1,000 characters₹100 per million characters · about a minute of speech · streamed, 24 kHz, any of our voices or yours
- 10ptoday
- 5pin six months
- <1pin about a year
- English (US), English (IN), Hindi and Hinglish voices; names, numbers and amounts read right
- Streaming WebSocket and REST, first audio in 27 ms server-side; pace adjustable per request
- Unlimited concurrency, in our cloud or your data centre
- Your own voice, trained for you, at the same price
- More languages as they leave training, same price
Pay as you go10p per 1,000 charactersprepaid credits from ₹100 (10 lakh characters), no card on file
Commit ₹10,000 / month12.5 crore characters included8p per 1,000 effective · then 8p per 1,000
Commit ₹25,000 / month40 crore characters included6.25p per 1,000 effective · then 6p per 1,000
Launch offer: the first 1,000 sign-ups get 10 lakh characters free, once the e-mail is verified.
A character is a Unicode character of the text you send, trimmed and counted before normalisation: every मात्रा, halant and nukta counts as one (नमस्ते is 6), and spaces and punctuation count. Pay as you go is prepaid: top up from ₹100 by UPI, card or net banking through Razorpay, 18% GST extra; when credits run out the API answers 402 until you top up, so there is never a surprise bill. A commitment is charged on the 1st of every month whatever the usage, unused characters do not roll over, and usage beyond the block draws on credits first, then goes on the next invoice. Larger commitments: talk to us.
For anyone whose product speaks
Every message, said out loud — in your customers' languages.
Bring the sentences your product needs to say and the volume you need to say them at. Thirty minutes is enough to hear them in our voices, talk about yours, and walk through the paise on your numbers.
10pper 1,000 characters, about a minute of speech, today
27 msfirst audio, server-side
∞streams, no concurrency cap
Your regiondata residency, your DC or ours
30-minute intro call
Hear it live, then talk numbers.
With Puru Chaudhary, founder & CTO. Pick a slot that suits you and the invite lands in your calendar with a video link.
- Your sentences, in our voices, live
- Your own voice: what to record, how long it takes
- Unit economics, the stack and the roadmap
The founders
Two founders, one stack.
One of us runs the company, the customers and the pilots. The other trains the speech recognition, the language model and the voices, and builds the platform they run on. Both have been here since the first call, and both have their own money in it.
Rajat MangotraFOUNDER & CEO
Runs Vartaloop and scales it, from the company and its customers to the pilots and the money. He has spent eight years building AI and ML products across retail, e-commerce, fintech, education and hospitality, with a record of making systems faster, cheaper and bigger, often cutting compute costs by 90% or more. Most recently, as co-founder of QuickDial AI, Vartaloop’s US sibling, he helped bring a CPU-only voice-agent stack down to a quarter of a cent a minute. An alumnus of IIT Kanpur.
Purushottam (Puru) ChaudharyFOUNDER & CTO · AI INFERENCE ENGINEER
Trained the voices and the speech recognition, wrote the engine that runs them on ordinary servers, and built the platform above it. Fifteen years of software for GE HealthCare, S&P Global, Bristol Myers Squibb and State Street; three years training models; three companies founded. Author of The Inference Mechanic.