Blog › Feature deep-dive
Multilingual AI receptionist: WarmLane answers calls in 29 languages, end-to-end.
Every WarmLane subscription can configure its voice agent to greet, transcribe, and reply end-to-end in any one of 29 languages — English (US, UK, AU, India), Spanish (Spain, Mexico, US), French (France, Canada), German, Italian, Portuguese (Brazil, Portugal), Dutch, Danish, Swedish, Norwegian, Polish, Russian, Turkish, Hindi, Arabic, Mandarin Chinese, Japanese, Korean and more. One agent runs one language per call; multi-language businesses run one agent per language, each with its own inbound phone number.
No extra cost. Every voice we ship is Amazon Polly Standard, bundled into Twilio's per-minute call charge — same $79/month whether your agent runs in English or Japanese. Premium neural voices (Polly Neural, Google Chirp3-HD, ElevenLabs) exist and cost more per character; we don't hide them but we don't push them either.
The full pipeline runs in the picked language: text-to-speech (Polly voice), speech-to-text (Twilio Gather with the matching STT model), and the AI response (a directive in the system prompt forces the model to reply only in the target language for the whole call). You pick the language in the editor; there is nothing else to configure.
What "multilingual" actually means on a live call.
Most vendors that claim multi-language support mean one of three narrower things: (1) they can read a translated script in a foreign accent (fine for a greeting; useless for a real conversation), (2) they support Spanish and nothing else, or (3) they surcharge you per character for each non-English word the agent speaks. We wanted the fourth thing: the whole call, in the language of the caller, at flat cost. Here is what that actually requires.
- The voice speaks the language natively. An English voice with a Hindi accent reading Hindi words is a bad experience; you need an actual Hindi voice. WarmLane uses Amazon Polly Standard voices — 60 voices across 29 languages — chosen per language in the editor. Aditi speaks Hindi. Zeina speaks Arabic. Marlene speaks German. Nothing pretends to be something else.
- Speech-to-text runs in the same language. If a caller speaks Arabic and Twilio's transcriber is set to English, the LLM receives garbled text and answers nonsense. WarmLane sets the
<Gather language="ar-SA">attribute on the transcriber to match the picked TTS voice on every call — so Arabic speech gets an Arabic transcript before the model ever sees it. - The LLM is told to respond in that language, and only that language. Left alone, GPT will happily switch to English mid-conversation if the caller mixes languages. WarmLane appends a strict directive to the agent's system prompt whenever a non-English language is picked: "You MUST respond ONLY in Hindi for the entire call. Do NOT switch to English regardless of what the caller says." Same rule for every supported language.
- Every turn uses the same language. A voice call is a many-turn conversation; each turn re-emits the TwiML. WarmLane propagates the language selection through every subsequent turn's
<Say>and<Gather>, so the agent never "forgets" partway through and defaults back to English.
Who benefits — a few concrete cases.
Latino-heavy patient base
Twenty percent of new-patient calls are in Spanish. Front desk speaks English. Before WarmLane: those callers reach voicemail, hear an English greeting, hang up, and try the next clinic. After: agent picks up in Spanish, books the cleaning on the calendar, sends the confirmation SMS in Spanish. Zero patients lost to language.
International guest bookings
Guests from France, Germany, Japan, Brazil. Twenty-four hour front desk isn't in the budget. Runs a single WarmLane agent per language and routes each inbound number to the right agent (say, one for French/German inbound, one for Japanese). Every caller hears their own language on the greeting; bookings land on the same shared calendar.
Bilingual dispatch
Half the emergency calls in the summer are in Spanish. The dispatcher speaks English. WarmLane's Spanish agent qualifies the call, decides whether it's an emergency, and books it — or warm-transfers to the dispatcher with a Spanish-to-English summary of the problem. No triage decisions lost in translation.
Reservation line in 3 languages
English, Spanish, Portuguese (Brazil). One receptionist can't cover all three fluently. WarmLane runs one agent per language, each with its own phone number. Every reservation lands in OpenTable, regardless of which line the caller dialed.
How WarmLane's multi-language pricing compares.
Language support is a genuine business capability — but the industry is inconsistent about pricing it. Some vendors treat it as a standard feature; others charge a per-character surcharge on non-English speech (which quietly turns your Hindi agent into a more expensive product than your English one).
| Vendor | Languages | Extra cost for non-English? |
|---|---|---|
| WarmLane | 29 languages, end-to-end (TTS + STT + AI) | No · bundled in $79/mo, no per-character fee |
| Rosie | "Multilingual support" listed as a feature, not specified count | Included in tier |
| Goodcall | Language selection not surfaced on public pricing / feature pages we could find | Not documented publicly |
| Smith.ai | English & Spanish (human receptionists) | Same per-call rate — but the tier is $300+/mo before per-call fees |
| Ruby Receptionists | English — bilingual coverage not documented publicly | Not documented publicly |
| AnswerConnect | English & Spanish (bilingual receptionists) | Same per-minute rate |
Competitor language claims read from each vendor's public feature page or pricing page, August 2026. Vendors don't always publish an exhaustive language list; where we couldn't find one we say so rather than guess. WarmLane's list is the exact set of Amazon Polly Standard voices currently wired into our editor's language picker — verifiable directly in the app at signup.
The 60-second setup, per language.
- Create the agent in Voice Agent → Agents → New agent (or edit an existing one).
- Pick the language in Advanced → Voice → Language dropdown. WarmLane auto-picks the first voice for that language; hear it read a sample by picking a different voice from the Voice dropdown.
- Translate the greeting + instructions. The editor shows a "Translate to <language>" button next to each field when the picked language isn't English. One click uses your workspace's OpenAI key to rewrite the text in the target language. You review, edit, save.
- Optional: translate your AI Brain content. If your workspace-wide business knowledge (services, hours, pricing, FAQs) is in English, the agent will translate on the fly for foreign-language callers. For lowest latency and highest fidelity, translate the AI Brain sections your multilingual agent grounds on ahead of time — the "Auto-translate" button in the AI Brain editor works the same way.
- Assign a phone number to the agent (Voice Agent → Numbers). One number can only serve one language at a time — but the number can be swapped between agents whenever you want, and you can buy or bring as many numbers as you need.
The honest limits.
We would rather set expectations correctly than sell you a feature that surprises you on the third call.
- One language per agent per number. A single agent is set to one language at a time; a call answered by that agent runs entirely in that language. If your callers are a mix of English and Spanish and you want the agent to detect language on the fly and switch, that's a different feature (call-language detection) and it's not what we ship today. The practical workaround is a separate agent + separate number per language.
- Basic voices for uncommon languages. Twilio's "Alice" (built-in polyglot voice) covers 27+ languages beyond the 29 with dedicated Polly voices, but the audio quality is lower — it works for guaranteed-uptime backup, not for a premium receptionist experience.
- Premium voices exist. If you specifically want Amazon Polly Neural (higher-quality synthesis, ~$0.016 per 1,000 characters extra), Google Chirp3-HD (via Twilio ConversationRelay, always-on infrastructure add-on) or ElevenLabs (custom voices, priced per character), those are all supported by Twilio and available on request — but they cost more, and the vast majority of our customers do not need them.
How to hear it.
Call our live demo number at +1 (218) 683-6315 — it currently answers in English. If you want to hear a specific language, email us the language and we'll spin up a test number in that language you can call the same day. It takes about a minute at our end.
Try it on your own callers.
14-day free trial, no card. Free Done-For-You setup — we build and translate the agent for you. Free ongoing support.