Soniox STT v6 listens to what your voice agent says, so it understands replies and ignores the bot's own echo

Soniox released STT v6, a speech-to-text model for voice agents, dictation, meetings and live translation. The new trick is "agent text input": your voice agent streams the words it's speaking to Soniox along with the caller's audio. The model then knows what the caller is answering ("River Roast", not "River Road") and filters any leftover echo of the agent's voice out of the transcript, even when the caller interrupts. It costs about $0.12 an hour for real-time transcription and $0.10 for files. Soniox claims "significantly higher accuracy" in 60+ languages (its docs list 60) but hasn't published v6 numbers yet.

Why it mattersIf you run a phone bot, mishearing short answers and bots interrupting themselves are two of the most common failures, and this targets both. But "world's first" is a stretch: AssemblyAI's streaming model already takes the agent's text as context, and Speechmatics' SDK could already ignore an assistant's voice. You still need normal echo cancellation, and you have to send the agent's words in time with playback. v5 users are moved to v6 automatically on 15 Dec 2026, and v6 writes prices as words, so check any code that reads amounts. Our check rates the claim as partly fact. We haven't tested it.

Curated by Sara Simeone as part of our human editorial coverage. The original sources are linked below; curation is distinct from a Hype or Fact? assessment.

SHARE THISPost on X ↗Share on LinkedIn ↗Copy link

How we knowCompany figures are labelled as their own claims; anything we couldn't verify is labelled as a report or a claim.