Soniox STT v6 Adds Agent Text Input and Echo Filtering for Voice Agents
Soniox has released Soniox STT v6, the latest generation of its speech-to-text API. Soniox STT v6 is a speech-to-text model that reads what a voice agent says while hearing the caller. It uses that text to recognize short replies and to drop the agent's own voice. The release ships as stt-rt-v6 for streaming and stt-async-v6 for files, across 60+ languages.
TL;DR
- Size: Soniox STT v6 is a proprietary hosted model, no parameter count published.
- Runs on: Soniox cloud API only (WebSocket streaming and async REST). No self-hosted or on-device option disclosed.
- Performance: Soniox calls it "significantly" more accurate than v5, but publishes no v6 numbers yet.
- Best: Price. $0.12/hour streaming and $0.10/hour async, with diarization and language ID bundled.
- Worst: Zero independent v6 results. Its v5 predecessor scored 1.34% semantic WER on Pipecat, behind v4's 1.25%.
What Problem Does Soniox STT v6 Solve?
In a cascaded voice agent, STT usually hears only the caller. It never sees the question being answered. Soniox STT v6 changes that by accepting the agent's own text as an input, which lets the model recognize short replies more reliably and filter out the agent's voice from the audio stream.
This addresses two common failure modes in voice agents: misrecognition of brief caller responses such as "yes" or "no," and echo contamination when the agent's own speech is picked up by the microphone. By feeding the agent's text into the STT pipeline, Soniox STT v6 gives the model context that a caller-only ASR system lacks.
Availability and Pricing
Soniox STT v6 is available through the Soniox cloud API in two variants:
stt-rt-v6for real-time WebSocket streamingstt-async-v6for asynchronous file transcription via REST
Both variants support 60+ languages, with speaker diarization and language identification included. Pricing is $0.12 per hour for streaming and $0.10 per hour for async transcription.
Evaluation Gap
Soniox describes v6 as "significantly" more accurate than v5, but has not yet published benchmark numbers for the new model. Independent results for v6 are currently unavailable. For reference, the v5 predecessor scored 1.34% semantic WER on Pipecat, behind v4's 1.25%.
Teams evaluating Soniox STT v6 for production voice agents should validate accuracy on their own audio, particularly in scenarios involving short caller replies and agent-voice echo, where the new capabilities are designed to help.
via MarkTechPost