Soniox releases STT v6, a speech-to-text model that reads a voice agent's text to filter its echo

Soniox has released Soniox STT v6, the latest generation of its speech-to-text API. According to MarkTechPost, the model reads what a voice agent says while it listens to the caller, and uses that text to recognize short replies and to drop the agent's own voice.

What shipped

• Model: Soniox STT v6, a speech-to-text model aimed at voice agents.
• Agent text input: the model takes the text the voice agent is speaking as context while it hears the caller.
• Echo filtering: it uses that text to remove the agent's own voice from what it transcribes.
• Short replies: the same text is used to help recognize brief caller responses.
• Model identifiers: stt-rt-v6 for streaming and stt-async-v6.

The report comes from MarkTechPost, published November 3, 2026.

What this means

Voice agents have to separate what the caller says from the agent's own speech playing back into the microphone. Soniox's approach with v6 is to give the recognizer the agent's outgoing text so it can use that text to filter the agent's voice and to interpret short replies. That moves part of the echo problem into the speech-to-text layer, which is relevant to teams building on the stt-rt-v6 streaming model.