Compare conversational voice AI providers
Every vendor says yes.This is how you find out.
A directory of 24 voice AI providers organised around the four things that actually decide the project — how fast it goes live, who does the implementation, whether you can pilot it, and whether the language survives a real caller.
Go straight to
Go straight to
- 24
- providers listed
- 20
- languages tracked
- 6
- claimed by their team
- 16
- articles and hard facts
Free · No account needed · We don’t sell leads, take placement fees, or let anyone buy their ranking
What this site does
- 1
What is actually out there
Every conversational voice AI provider we can verify, on one page, compared on the same four questions rather than on whichever features each one chose to advertise.
- 2
Which one fits your business, and why
A shortlist built from your languages, call volume, channels and appetite for engineering — with the arithmetic shown, so the ranking is something you can argue with rather than trust.
- 3
How the thing you are buying works
The stack and the architecture, layer by layer: what each one does, what each one bills for, what breaks first, and what a vendor's claim about it leaves out.
Why this exists
Everyone in this category sells speed. Somebody has to publish the timeline.
The claims are not lies — you really can have a bot answering a phone number this afternoon. But “live” in a pitch means a bot that answers, and “live” in your head means a bot in front of customers. Everything on this site exists to close that gap before it costs you a launch date.
“Live in 5 minutes”
15–65 working days to production
Five phases sit between the demo and your customers. Exactly one of them is the phase a platform can shorten on its own; two are yours whatever you buy.
See the five phases →“One simple per-minute rate”
6 billable layers, plus number rental
A quote usually prices one of them. The rest — transcription, the model, speech, orchestration, storage — arrive later, and telephony is priced by the country you dial.
Work out the real bill →“Supports 100+ languages”
Coverage is not proficiency, and neither is code-switching
Handling Hindi and English is a different capability from following a caller who moves between them mid-sentence — which is the normal case, not the edge case.
Why it breaks →10 claims translated in full, none of them attributed to a named company — these are the sentences the category uses, granted where they are true and questioned where they are ambiguous. Read them all.
I want to hire someone
Compare providers on what decides the project.
Search the directory, or describe the work and get a ranked shortlist with the arithmetic shown, everyone who was ruled out and why, and the questions to ask on the first call.
Build my shortlist →I want to build it myself
Understand the six layers before you buy any of them.
What telephony, transcription, the model, speech synthesis, orchestration and storage each do, what a developer actually has to build at each one, and what the whole thing costs per minute.
See how the stack works →Results
Tick two. See them side by side.
Ordered by how much each provider has actually published and evidenced — not by how good we think they are, and never by what they paid. A yellow cell is a question, not a mark against anyone.
| Provider | Languages | Live in | Built by | Pilot | Profile | |
|---|---|---|---|---|---|---|
| 7 · switching | 7 days | Guided | Yes | 99% | ||
| 4 · switching | 3 days | Self-serve | Yes | 93% | ||
| 7 | 14 days | Guided | Yes | 90% | ||
| 4 · switching | 21 days | Vendor-built | Yes | 88% | ||
| 3 · switching | 30 days | Vendor-built | No | 75% | ||
| 8 | not published | Guided | not published | 40%compiled |
Up to 4 at a time.
“Not published” means exactly that. It is not a penalty — it becomes a question we hand you for the first call, and a fast published lead time gets a question too.
What it costs
A per-minute quote covers one layer of six.
Voice AI is a stack, and almost every quote you receive prices one part of it. Here is the whole bill — telephony, transcription, the model, speech, orchestration and storage — with the arithmetic shown and every rate editable.
Per minute
$0.0448
Per conversation
$0.1344
Per month
$2,688
20,000 conversations a month of 3 minutes, calling India, at indicative mid-band rates. Shown in USD. Change any of it on the calculator.
- Telephony$0.0120/min
27% of the bill — Small per minute, but the one layer with regulatory exposure attached.
- Speech to text (STT/ASR)$0.0080/min
18% of the bill — Runs for the whole call, including silence, unless you gate it.
- Language model$0.0100/min
22% of the bill — The layer whose cost you control most directly, by controlling context length.
- Text to speech (TTS)$0.0150/min
15% of the bill — Often the single largest line item once voice quality matters.
- Orchestration and turn-taking$0.0060/min
13% of the bill — Cheap to buy, expensive to build. This is the real build-versus-buy decision.
- Recording, storage and analytics$0.0020/min
4% of the bill — Rounding error on the bill, load-bearing for the decision.
Telephony is priced by country.
It is the only layer geography changes, and it changes it hard. A minute to India runs about $0.0120; the same minute to the Philippines runs about $0.1100 — roughly 9× the price, on the line item most estimates treat as a constant. Pick your market in the calculator and the rate follows.
What we compare
Four questions, on every provider.
These are the same fields the matching engine scores against, so the copy on this page and the logic behind a shortlist cannot drift apart. See the whole scoring model.
01
Time to value
Days to a bot you can dial, weeks to production, and hours before anyone even replies to your enquiry.
02
Implementation method
Who builds it — you, you with help, or them — and what training data they need from you before it performs.
03
Pilot terms
Whether a pilot exists, how long, what it costs, and whether a contract has to be signed first.
04
Language reality
Per language, at what level, and whether it follows a caller switching mid-sentence.
Start from where you are
“How long does it take?” depends on who does the work.
Who configures it
- Self-serveYou build it. Sign up and start today; nobody is waiting on a vendor calendar.Browse →
- GuidedYou build it with their solutions engineer alongside. Usually a shared Slack channel.Browse →
- Vendor-builtThey build it and hand it over. Least effort from you, longest wait, hardest to change later.Browse →
What they need from you before it works
- Prompt only
- Write instructions in plain language. Changes take minutes. No data assembly.
- Documents
- Point it at your existing PDFs and help centre. Days, if the documents already exist.
- Call transcripts
- You must supply real call transcripts before it performs. Budget weeks for collection, redaction and consent review — this is usually the longest pole, and it is your effort, not theirs.
- Custom model
- A model is fine-tuned for you. Highest ceiling, but retraining is a scheduled project rather than an edit, so every later change is slow too.
Knowledge base
The things nobody tells you until week three.
Written here, not collected from vendor blogs. No claim about a named company without provenance, and no number without its denominator.
9 min read
The Hinglish problem: why your voice bot will fail on real Indian callers
Supporting Hindi and English is not the same as following a caller who switches between them mid-sentence. What breaks, why, and how to test for it before you buy.
Read →
11 min read
DLT, DND and outbound voice bots in India: what the obligations actually are
A plain-language primer on the regulatory layer under automated outbound calling in India — who carries the obligation, what it means for a pilot, and the questions to settle in writing.
Read →
10 min read
Per-lead pricing: the only model where your vendor wants what you want
Per-lead pricing aligns incentives better than per-minute — and creates a definition problem that decides whether the deal works. How to write the definition before you sign.
Read →
Built in India
6 companies, one wall.
Indian teams are building the hardest version of this problem — real time, multilingual, code-switching, over networks that drop.
Amazon Lex
Bland AI
Cartesia
Cognigy
Deepgram
ElevenLabs
Exotel
Gnani.ai
Google Cloud CCAI
Haptik
Kore.ai
LiveKit
Observe.AI
Ozonetel
Retell AI
Sarvam AI
Vapi
Yellow.ai
- +6 more →
All company names, logos and trademarks are the property of their respective owners. Marks are shown to identify each company, resolved from that company’s own website, and every one links back to it. Their appearance here does not imply any affiliation with, endorsement by, or sponsorship of Voice AI Bots. Voice AI Bots is an independent consolidation platform. Full notice, and how to have yours removed.
Not sure what to search for?
Describe the work instead.
Three minutes on your languages, volume, deadline and what the bot has to accomplish. You get a ranked shortlist with the arithmetic exposed, everyone who was ruled out and why, and the exact questions to ask each one.
Where this directory currently stands
18 of the 24 listed profiles are compiled — assembled from public sources, unclaimed and unconfirmed by the companies themselves. 5 publish a time-to-prototype figure, because we are still asking. A further 6 marked Demo are invented sample data, shown so the matching and comparison features can be seen working end to end.
That is why compiled scores look low. A score reflects how much is known and evidenced, not how good a provider is. Read it as our to-do list — and as your question list for the first call.