Soniox STT v6 Adds Agent Text Input and Echo Filtering for Voice Agents
What changed
Soniox has released the sixth version of its speech-to-text (STT) API, named Soniox STT v6. This update is specifically designed for voice agent use cases, where the system processes audio from both the agent and the caller simultaneously. It introduces two key features: agent text input and echo filtering. The model listens to the conversation, transcribes the caller’s speech, and also reads the text input representing the agent’s responses. By doing this, it can recognize short replies from the agent more accurately and filter out the agent’s own voice from the transcript to avoid confusion or duplication.
Why builders should care
Voice agent interactions have unique challenges because the system must distinguish between the agent’s voice and the caller’s speech. Without proper echo cancellation, the agent’s voice can bleed into the transcription of the caller, reducing transcript quality and making intent detection or downstream automation less reliable. Soniox STT v6 addresses this by combining live speech recognition with agent text input, effectively telling the model what the agent just said. This reduces transcription errors caused by overlapping speech and improves response detection. For developers building voice automation or call center integrations, this simplifies the filtering task and improves transcription accuracy.
The practical takeaway
Operators running voice agents or call centers can expect clearer transcripts that separate agent and caller speech, cutting down on manual cleanup or complex post-processing. Builders integrating conversational AI will get faster, more reliable recognition of short agent replies, enabling smoother automation triggers or real-time analytics. The solution ships in two variants: stt-rt-v6 for streaming transcription and stt-async-v6 for asynchronous jobs. This makes it flexible for different deployment needs, whether live customer support or batch processing of calls.
What to watch next
The next step is monitoring how developers adopt the agent text input feature in real-world scenarios and whether it significantly reduces errors in mixed speech environments. Also, watch if this approach pressures other STT providers to improve echo cancellation and agent-side integration for voice automation. How Soniox handles accents, noisy lines, and different agent workflows will influence broader STT use in service automation. Finally, tracking API pricing and latency changes will reveal practical operator costs for deploying this new model at scale.
AI Quick Briefs Editorial Desk