Languages and how they are chosen

Akume does not ask you to pick a language to translate from and a language to translate into. You give it a single list: the languages that will be spoken in this meeting. Each time someone talks, Akume recognizes which of those languages they used and translates that turn into the others. The current list of supported languages is the picker in the app: it is the one place that is always up to date, which is why this page does not reprint it.

One list, chosen before the call

• From 2 to 10 languages per session. Below two there is nothing to translate between; ten is the current ceiling.
• Regional variants of one language count as one. English (US) together with English (UK) is still a single language as far as the session is concerned, so a list made only of those two will not start. Pair them with a genuinely different language.
• The list is fixed once the session starts. To add or drop a language, stop the session and start a new one, which also means a new listener link.
• Every language except the spoken one costs minutes. The language being spoken is not counted, so a minute of the call uses one minute fewer than the number of languages selected: two languages cost one minute per minute, five languages four. Free demo and paid intro has the arithmetic.

Nobody hears their own language

Spoken translation is produced for the other languages in the list, never for the one being spoken. On the listener link, someone following the call in Spanish hears nothing while Spanish is being spoken: their audio comes to life as soon as somebody speaks one of the other languages. The captions are not affected: every utterance is captioned, and the Spanish ones simply appear as the original text. Silence at the start of a call is usually this, not a fault.

What listeners choose

Every meeting gets its own listener link. Whoever opens it in a browser chooses one of the session’s languages; if the status line asks them to press play, one tap on Play starts the audio, and the same button pauses. From there they follow the call with captions and translated audio: no account, nothing to install, up to 500 listener connections per session. One meeting can therefore be followed in several languages at once, each person on their own.

Getting the best accuracy

• List only the languages that will really be spoken. Extra entries cost minutes and give the detection more candidates to confuse.
• Where a language offers regional variants, pick the one closest to the speakers: it matches pronunciation and everyday vocabulary better than a generic choice.
• Use a headset. Most translation mistakes start life as audio that was never clean enough to recognize: laptop speakers picked up by a laptop microphone are the usual culprit.
• One speaker at a time. Heavy crosstalk is hard for people too.
• Introduce product names, company names, and acronyms slowly the first time they come up in a call.

Need a language you cannot find in the picker? Tell us which one at support@akume.ai.