Almanac
Microsoft/teamsMicrosoft 365

Consultant-focused KB for Microsoft Teams: implementation notes, gotchas and configuration decisions beyond the official docs — across chat and collaboration, meetings, webinars and town halls, Teams Phone, PSTN connectivity (Direct Routing, Operator Connect, Calling Plans, Teams Phone Mobile), auto attendants, call queues and contact centre integration, Teams Rooms and devices, apps and extensibility, Copilot in Teams, administration and governance, security and compliance, network and call quality, and licensing — plus cross-cutting decision guides.

feature-teams-copilot-interpreter-agent.mdv1 · history
CurrentApplies to Microsoft 365 CopilotUpdated 3 weeks agoSource Microsoft Learn ↗

What it does

Real-time speech-to-speech translation in a Teams meeting or 1:1 Teams call. Each licensed participant hears the meeting in their own language, either in a simulation of the speaker's own voice or in a preset synthetic voice. It's an audio interpreter, not captions.

Key facts

  • Included with the Copilot licence (Learn labels it the Microsoft Copilot licence). Learn doesn't flag it as preview.
  • 20 hours of interpretation per user per month are included. Beyond that, access is "subject to available capacity" — not a hard stop, not a guarantee.
  • Nine languages for speaking and listening: Chinese (Mandarin), English, French, German, Italian, Japanese, Korean, Portuguese, Spanish.
  • Voice options: simulate the user's own voice, or preset voices Ava (female, the default automated voice), Andrew (male) and Fable Turbo (neutral). Voice samples and biometric data aren't stored.
  • Admin control is PowerShell only. -AIInterpreter on Set-CsTeamsMeetingPolicy (meetings) and Set-CsTeamsCallingPolicy (1:1 VoIP calls), values Enabled / Disabled, enabled by default.
  • -VoiceSimulationInInterpreter on the meeting policy sets the default for "Choose your voice when interpreted": Enabled (default) means "Simulate my voice", Disabled means automated voice. Users can still change it. Microsoft recommends leaving it enabled.
  • Supported: scheduled meetings, channel meetings, 1:1 Teams-to-Teams VoIP calls, desktop (Windows and Mac), mobile, web (Chrome, Edge, Safari, Firefox), Teams Rooms on Windows and VDI.
  • Not supported: 1:1 PSTN calls, Teams events, Teams Rooms on Android, personal peripherals used with Teams Rooms, and Teams free.
  • Processing runs through Azure speech services: speech recognition to English text, machine translation, then text-to-speech. A bot carries the meeting audio for cloud processing.

When to use / skip

This is the feature that makes multinational clients sit up. A UK team running a weekly call with colleagues in France, Germany and Japan has been relying on goodwill and everyone's second language. Interpreter gives each person the meeting in their own language with roughly their own voice. For internal cross-border meetings it's genuinely useful, and the 20-hour monthly allowance covers most people's working pattern.

Don't position it as a replacement for professional interpretation in formal settings — board meetings, legal proceedings, works council negotiations, anything where a mistranslation becomes a dispute. It's AI translation in real time and it will get nuance wrong. It also doesn't work in Teams events, so town halls and webinars still need the existing language interpretation feature or live translated captions.

The voice simulation question is the one to put in front of HR and the DPO early. Voices aren't stored, but "the AI speaks in my voice" causes far more discomfort than "the AI translates me". Some organisations will want the automated-voice default.

Configuration decisions

  • Leave -AIInterpreter enabled everywhere (the default), or disable it in policies for groups where AI translation isn't acceptable.
  • Voice simulation default: Enabled for the best experience, or Disabled so users opt in to having their voice simulated.
  • Whether 1:1 VoIP calls get Interpreter too — that's the calling policy, managed separately.
  • Which meeting types to steer away from Interpreter (events, formal or contractual meetings) and what to use instead.
  • Language coverage check against the client's actual languages — nine isn't many if they work in Polish, Dutch or Arabic.
  • How to monitor usage against the 20-hour allowance for heavy users.

Gotchas

  • No Teams admin center toggle. If you go looking for it in meeting policies in the portal, you won't find it — it's PowerShell.
  • Teams events aren't supported, even though the Learn page header lists events in its "applies to" line. Trust the supported list.
  • Teams Rooms on Android isn't supported. A mixed Windows and Android rooms estate will get inconsistent behaviour.
  • PSTN callers can't use it. A customer dialling in by phone gets no interpretation.
  • The 20 hours is per user per month, and beyond it you're on available capacity. Heavy daily users can hit it.
  • Speech recognition converts to English text before translating. Language pairs that don't involve English still go through that path.

Consultant notes

  • Policy check: Get-CsTeamsMeetingPolicy | Select Identity, AIInterpreter, VoiceSimulationInInterpreter and the calling-policy equivalent for AIInterpreter.
  • Pilot with a real multilingual team, not English speakers pretending. You learn more about quality in one genuine meeting than in a week of demos.
  • Write a short "when not to use Interpreter" line into the user guidance — formal and contractual meetings, external customer calls over PSTN.
  • If the client already runs human interpreters through the meeting language interpretation feature, keep that for events and formal sessions and let Interpreter cover everyday internal meetings.
  • Check the rooms estate before promising in-room interpretation: Windows yes, Android no.

Worth revisiting when Learn adds languages, events support or Teams Rooms on Android to the supported list

Was this accurate?