The problem
Customers call in the languages they are comfortable in, and few teams can cover all of them at every hour. The usual answers are a phone menu, a callback in another language, or a caller who gives up.
Separate setups per language also drift apart: a change to a policy has to be made several times, and sooner or later one version is out of date.
Who it is for
Hotels and travel: Guests calling from abroad, at any hour.
Companies across borders: One support line for customers in several countries.
Local businesses with mixed communities: Practices, offices and shops whose customers do not all share one language.
Contact centres: Covering languages that are hard to staff on every shift.
What it handles
One language per flow
A flow's primary language sets transcription, voice and the language the model answers in. More than 30 languages are supported.
Several languages in one call
Set a flow to a list of languages, or to detect the language, and the call can be held in more than one.
Shared logic
One flow can back several agents, so the same instructions run in English and German with only the voice and language changed.
Natural speech
Text is formatted for speech in the flow's language, so numbers, dates and abbreviations are read the way a native speaker would say them.
Pronunciation
Dictionaries tell the voice how to say brand names, product codes and unusual surnames.
How it works
Choose how to split it
One flow with several languages, or one agent per language on the same flow.
Set the languages
Pick a primary language, a list of languages, or detection.
Choose voices
Pick a voice that suits each language, or add a custom one to your workspace.
Set the disclosure
AI disclosure presets exist for every supported language, or you can write your own sentence.
Test with real speakers
Have native speakers call it, including people who change language mid-sentence.
What you need
- The languages your callers actually use
- Instructions, and knowledge in the languages you need it answered from
- A voice for each language
- People who speak those languages, for testing
What it connects to
Voices: ElevenLabs, PlayHT and Fish Audio, among others, plus custom voices.
Telephony: Twilio, Telnyx and SIP trunks.
Flows: A call can move to another flow, for example a language-specific one, and how much history carries over is configurable.
What to expect
- Callers can be served in their language at hours when no speaker of it is on shift.
- A change to your instructions reaches every language that shares the flow.
- Calls in every language leave the same kind of record and can be analysed together.
Limits
- Not every combination works within one call. By default, English, German, French, Spanish, Italian, Dutch, Portuguese, Russian, Hindi and Japanese can be combined.
- Other languages, Turkish among them, run as the single language of a flow. Mixing them into one call needs the flow's alternative multilingual transcriber, so test that setup with real speakers first.
- Quality differs by language and voice. Do not assume a language works because it is on the list.
- Names and addresses are hard in any language. Have the agent confirm them.
How to evaluate it
Start with two languages: Pick your two most common ones and get those right first.
Test the switch: Change language mid-call, and test a language you did not configure.
Check the details: Dates, numbers and names are where a language setup usually shows its weak spots.
Compare in analytics: Look at outcomes per language before adding the next one.
Questions
How many languages are supported?
More than 30 as the language of a flow. The set that can be combined within a single call is smaller.
Can an agent switch language during a call?
Yes, when the flow is set to several languages or to detection, within the languages that can be combined.
Can one agent speak German, English and Turkish in the same call?
German and English combine by default. Turkish works reliably as a flow's single language, so the simplest setup is a separate Turkish agent or flow. Mixing all three in one call needs the alternative multilingual transcriber and should be tested first.
Do we have to rebuild the agent for each language?
No. One flow can back several agents that differ only in voice and language.
Is the AI disclosure translated?
Presets exist for every supported language, and you can write your own wording.