Speak naturally.
Be understood everywhere.
Parla listens, understands and speaks back in another language as the conversation happens — no waiting, no push-to-record, no language barriers.
Works instantly in demo mode. Add an API key for full translation coverage.
Built for conversations that cannot wait
Every layer of the pipeline is optimised for latency — from microphone capture to spoken audio.
Real-time by design
Partial transcripts stream in as you speak, so translation starts before you finish your sentence.
20+ languages
From English and Turkish to Japanese and Arabic. Manual or automatic language detection.
Natural voice
Interpretations are spoken back with a natural voice, streamed and interruptible when you talk over it.
Two-way ready
Auto-detect mode is built for back-and-forth conversation between two speakers.
Provider agnostic
Speech, translation and voice are pluggable. Swap providers with a single environment variable.
Private by default
Keys stay on the server. Sessions are isolated and nothing sensitive is written to logs.
From your voice to theirs
You speak
Your voice is captured and analysed locally in the browser.
We understand
Speech is transcribed, the language is detected and partial results appear instantly.
We translate
The meaning is translated into the target language with tone preserved.
It speaks back
A natural voice delivers the interpretation — and stops the moment you speak.
Speak in your language
A growing catalogue of languages, managed from a single shared configuration.
Your voice stays yours
Audio is streamed over an authenticated WebSocket to a dedicated session and processed on the server. Provider keys never reach the browser, sessions are isolated, and transcript content is excluded from production logs.
- No API keys in client bundles
- Per-session isolation and rate limiting
- WebSocket origin validation
- Configurable transcript retention
Break the language barrier in seconds
Open the interpreter, press start, and just talk.
Start interpreting