简体中文In one sentence: native calls are covered by Samsung, Apple and Google on some models; Zoom and Teams cover only their own meetings and require a licence; in-app calls such as WhatsApp and WeChat are out of reach for all three, because the operating system will not hand another app's audio to a translation module. On a PC, a virtual audio cable that takes over the system audio is the only stable route for those calls today.
Coverage by call type
| Call type | What works today | What it costs |
|---|---|---|
| Face to face | Samsung Interpreter, Apple's Live Translation with AirPods, Google Translate conversation mode | Free or bundled with the device; both people must be in the same room |
| Native phone call | Samsung's call translation, Apple's Live Translation in Phone and FaceTime | Supported models only; the language list is shorter than the marketing suggests |
| WhatsApp / WeChat call | Windows desktop plus a virtual audio cable and an interpreter app | Driver install, manual device routing, 2–5 second delay, synthetic voice |
| Zoom / Teams / Tencent Meeting | Platform voice translation, translated captions, human interpretation channels | Add-on licence or Copilot seat; both sides usually on the same platform |
| Conference and events | Wordly, Kudo, Interprefy, or rented headsets plus interpreters | Bought by the hour; Wordly lists about US$1,500 for a 10-hour package |
What phone-level call translation really translates
Samsung ships two features that get conflated. Interpreter is the split-screen face-to-face mode; call translation is the one that speaks inside a phone call, covering 20-plus languages including Simplified and Traditional Chinese, running on-device where it can. Availability varies by model and region, and the vendor states that AI output is not guaranteed to be accurate.
Apple 's Live Translation arrives with iOS 26, covering Messages, Phone and FaceTime, processed on-device with Apple Intelligence, with call languages including Simplified and Traditional Chinese. General availability in the China mainland is slated for December 2026 with iOS 27.3, so as of November 2026 it does not count as available there yet.
Google Translate's Live mode covers 70-plus languages for conversation, listening and split-screen use, keeps translating in the background on Android, and limits offline Live translation to some Pixel models. None of the three will translate a WhatsApp call, because the translation happens inside the phone's own telephony stack.
So the quickest test is to ask one question: is this call going through the native dialer? If yes, the built-in feature has a chance; if not (WhatsApp, WeChat, Telegram and other in-app calls), go straight to the desktop route.
Why meetings are easier than calls
Meeting platforms own the whole audio path, so they can insert translation at any point. The trouble is that every platform has several features that all get called interpretation:
- Zoom sells three things: human interpretation channels (paid plan required, and an auto-generated meeting ID), AI translated captions (generally Business Plus or an add-on), and AI voice translation (Live Translation add-on or ZoomMate, desktop client 7.0.0 or later). Zoom's documentation states that long stretches of speech are played after the speaker pauses.
- Microsoft Teams bundles Interpreter into the Copilot licence, 20 hours per user per month; guests and anonymous joiners cannot use it, and recordings keep the original audio only.
- Tencent Meeting offers AI interpretation free during its stated free period, but only between Mandarin and English for now.
- DeepL Voice plugs into Teams, Zoom and Google Meet for speech-to-speech and captions, but the vendor states it is not officially supported in the China mainland.
One more practical question: who needs to hear the translation. Most products serve only you by default, and the conditions for letting the other party hear it differ — a translation button in a menu does not mean the other party hears your voice.
The desktop audio path is the real barrier
On Windows the trick is a pair of virtual audio endpoints: the calling app sends its output to the cable, the interpreter app reads it, translates, and speaks the result into a second cable that the calling app uses as its microphone. The other party hears ordinary speech and installs nothing.
Three costs come with it: configuration is manual, and picking the wrong device on either side means the other party hears nothing, or the tool translates your own voice; latency usually runs two to five seconds; ; the synthetic voice is recognisably synthetic, and that gets amplified on calls where tone carries meaning.
The four things that break first in a real call
- Numbers, prices and spellings: fifteen versus fifty, part numbers, email addresses. Anything headed for a contract should be confirmed in writing.
- Interruptions: turn-taking is the first casualty; in practice people fall into shorter turns and repeat themselves often.
- Accents and terminology: the same sentence can be fine in one direction and weak in the other, so test both.
- Long monologues: the longer the other party talks without pausing, the more the translation bunches up at the end.
Before you pay, spend ten minutes testing it yourself
- Use a real recording of your own call, not a vendor demo: include the names, numbers and accents you actually deal with.
- Time the gap with a stopwatch, from the end of their sentence to the first word of your translation.
- Test both directions. Many tools are fine from a foreign language into Chinese and clearly weaker the other way.
- Count how many manual steps it takes before a call, and whether the devices are still selected after a reboot.
- Read the data terms: whether audio is stored, for how long, whether a voiceprint is kept, and what happens to the transcript.
Frequently asked questions
Can WhatsApp or WeChat calls be translated in real time?
The built-in phone features cannot. They cover native calls and FaceTime only. On Windows, a virtual audio cable plus an interpreter app is the only stable two-way method.
Does the other party need to install anything?
Meeting features need the other party on the same platform with a licence; interpreter earbuds generally need a unit on each side; with the desktop audio route they install nothing.
What latency is acceptable?
Two to five seconds is usable. Above eight seconds it only works for one-way comprehension, not for negotiating.
Can it use a cloned voice?
Some products offer it, but it hands your voiceprint to a model. Tell the other party first and check the consent-to-record and privacy terms.
Can AI call translation replace a human interpreter?
For routine calls yes; for negotiations, hearings, medical and legal matters no. Vendors price human and AI separately for exactly this reason.
