Hiring
An interview with a Chinese employer where neither side speaks the other’s language. First deployment — on a hiring platform for foreign professionals.
外籍人才.cnWAIJIRENCAI.CN
聊无界Communication without borders
Speak freely. Language is no longer a barrier.
A video room where two people speak different languages — and each hears the other in their own. A job interview, a negotiation with a partner, or a call with a friend in China. No human interpreter. About a second and a half from the end of a phrase to voice on the other side.
上海 · 中文
Krasnoyarsk · Russian
Transcript · job interview
delay 0.0 s
≈1.5 s
from the end of a phrase to playback on the other side
278 ms
first sound of streaming synthesis, server in Hong Kong
6
translation directions: Russian · Chinese · English
$5
per hour of conversation — paid by the room creator
用Who it’s for
Three conversations that couldn’t happen before
One engine. Only the subject changes.
Business
Negotiations with a supplier or a partner: price, terms, deadlines — and a transcript both sides return to after the call.
suppliersuniversity admissionsvisastelemedicine
Friends
Call a friend, a classmate, or your wife’s family in China and talk — instead of texting through a translator app.
friendsfamilytravel
层How it works
Three independent layers. If one fails, the conversation goes on.
Video flows directly between browsers, translation runs on a separate channel, and the transcript gathers everything into a single record.
Video
WebRTC · browser ↔ browser
The connection is established directly. The server neither relays nor records the video. This is where the face, the emotion and the trust live — the reason for a video call at all.
Not recorded. The value of the conversation is the transcript, not a video file.
Translation
recognition → translation → synthesis
Each side’s microphone feeds streaming speech recognition, then translation, then streaming speech synthesis played to the other side — in a male or female voice chosen by the speaker.
Direction is set by the participant’s declared language, never guessed from audio. That removes errors on mixed speech and names.
Transcript
original · translation · who · when
Every utterance lands in a shared feed. Typed chat messages go through the same pipeline and end up in the same place. The result is a single record of the conversation you can reread and save.
One document for both sides. What people come back to after the call.
规Room rules
The decisions everything rests on
Live translation breaks not in the models but in the details: echo, overlapping voices, a queue of phrases. Each rule below closes one such gap.
- Half-duplex
- While playback is running on one side, that side’s microphone is off. Otherwise the synthesized voice feeds back into recognition and the conversation loops.
- Remote audio is muted
- With the interpreter on, the other side’s original voice is not played: it isn’t understood anyway, and two voices at once are unintelligible.
- Utterance limit — 10 seconds
- A phrase is force-closed so the translation queue doesn’t pile up during a monologue.
- Graceful degradation
- If synthesis is unavailable, the room doesn’t fall over: the conversation continues with subtitles and text.
- Video is not recorded
- Cheaper, and it removes a class of personal-data concerns.
价Pricing
$5/ hour
Per-minute billing — every started minute of the call counts; packages are sold in hours. The room creator pays; the other party joins by link — free, no registration. An hour with a human interpreter costs dozens of times more, and has to be booked in advance.
- First sound of synthesis, streaming278 ms
- First sound of synthesis, non-streaming2,886 ms
- End-to-end run RU → ZHworks
- Directions RU · ZH · EN6 of 6
- Extra trafficrelay only
私Privacy
What is stored — and what is not
- Video is never stored. Neither on the server nor on either side.
- The transcript with masked contacts is stored permanently. Raw speech recognition text is kept for 90 days, then erased.
- Contacts spoken aloud are masked the same way as typed ones.
- Both parties see a transcription notice before entering the room.
The personal-data operator is Aziya Global Limited (Hong Kong). Speech is sent for recognition and translation to Alibaba Cloud (Singapore), and the service runs on a server in Hong Kong: this is a cross-border transfer, and we ask for separate consent to it at sign-up. See the Privacy Policy.
限Limits of the current version
Honest about what’s missing
- Exactly two participants. Bringing a colleague or your own interpreter isn’t possible yet — it changes the connection model.
- Headphones required. Without them the microphone picks up the playback.
- Take turns. The system doesn’t yet allow interrupting mid-phrase.
往Roadmap
From first conversations to an engine under your own brand
- Next
Personal account and the first live conversations
Registration, room creation by link, connectivity checks from mainland China, first interviews and negotiations.
- Then
A call summary after the conversation
Agreements extracted from the transcript: for hiring — position, salary, visa, dates; for a deal — price, volume, terms. What people come back to after the call.
- Long term
The engine as a standalone product
Embeddable into other services where the two sides share no language.
Try a conversation without an interpreter
Registration takes a minute. Create a room, pick the languages and send your counterpart the link — they don’t need an account. Works in the browser, nothing to install.
Questions — viktor.pavliukov@aziyaglobal.com, phone / WhatsApp / WeChat +86 152 1006 1536. Hiring on waijirencai.cn — an employer pilot, write to us.