KNet Voices
Speech in and speech out for KNet agents — transcription, synthesis and real-time voice sessions.
Placeholder. The overview, links and concepts below are accurate; the detailed guides are still being written. Anything not yet covered is listed under What is not documented yet.
KNet Voices handles the audio side of an agent: turning speech into text for the agent to act on, and turning the agent's replies back into speech. Use it for phone-style interactions, voice assistants, and any flow where typing is not practical.
Open it: flow.knetai.com/voices
Where it fits
- Voices converts between audio and text.
- Workflow Builder decides what to do with the text.
- KNet Traces records the run, so you can see whether a bad answer came from a mis-transcription or from the agent's reasoning.
That last point matters more in voice than in text. When a voice agent answers the wrong question, the cause is usually transcription rather than the model, and the trace is where you tell the difference.
Concepts
Transcription — speech to text, for the agent to read.
Synthesis — text to speech, for the user to hear.
Session — a continuous voice interaction. Sessions are real-time and stateful, unlike a one-shot transcription request.
What is not documented yet
- Choosing and configuring voices
- Real-time session setup and latency tuning
- Language and accent support
- Handling interruptions and barge-in
- API and MCP endpoints
- Audio format requirements and limits
Getting help
Something here wrong or missing? Get in touch — a person reads every support request, and gaps in these pages are worth reporting.