Live speech translation · model creator
Lexven is a core AI/ML product: live speech translation models and a developer API covering fifty-plus languages. Apps stream audio in, receive translated speech and text with timestamps, and keep turn-taking intact on a call. Lexven is not building AI agents — the gallery stays a room of people.
AI features
Lexven is in the business of building a core AI/ML product. Named jobs: speech-to-speech, speech-to-text with timestamps, turn-taking, and in-house translation models. This is a Model Creator — not an agent company.
AI feature · Speech-to-speech
For rooms that cannot wait for a transcript
Apps stream audio in and receive translated speech with the original cadence held close enough for a meeting. SDKs for WebRTC, SIP, and a virtual mic.
PCM / Opus in → POST /v1/streams → translated audio
AI feature · Speech-to-text with clocks
For UIs that draw a waveform and a transcript together
Live captions in the listener's language, with word timings. Every audio packet and caption share a stream clock so turn-taking stays intact.
Stream clock → captions + audio packets
AI feature · Turn-taking
For people who talk over each other the usual way
Lexven keeps a short lookahead so a listener can cut in. The output stream ducks instead of finishing a stale sentence. Fifty-plus languages on Room.
Lookahead → duck → keep the floor
AI feature · In-house translation models
For teams that need glossaries and their own stack
In-house translation stacks, with fine-tunes for medical and legal glossaries. Planned families also include AWS Nova and Llama 3 8B. Access: Bedrock, Hugging Face, Groq. Lexven is not building AI agents.
Glossary + model pack → stream
IVA / speech. Core AI/ML product. Model Creator. Not building AI agents. MVP stage — no invented customers, funding, or uptime claims. Existing site figure of 240 ms median first audio on a quiet line remains on the product story, not a new claim.
Method
Send PCM or Opus. Name source and target languages.
Lexven returns audio and text with matching clocks.
Barge-in still works. People talk over each other the usual way.
Stack
Translated audio with the original cadence held close enough for a meeting.
Live captions in the listener's language, with word timings.
WebRTC, SIP, and a virtual mic for apps that already have a call path.
In-house translation stacks. Fine-tunes for medical and legal glossaries.
Start
Write team@lexven.club. You get a test token, a sample project, and a 30-minute walkthrough on the product you actually want to run.
Write team@lexven.club. This form stores the request on the page so you can copy it into that mail.
Thanks. We saved this request on the page. Mail team@lexven.club if you want a live reply.