> For the complete documentation index, see [llms.txt](https://oodlz-ai-studio.gitbook.io/docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://oodlz-ai-studio.gitbook.io/docs/inbound-agent/every-channel/voice.md).

# Voice

A parallel voice pipeline — same intelligence, restructured for ears, with a clickable transcript on the side.

> **The agent's voice — clean prose to listen to, full markdown to read.**

Bolting voice onto a text AI just plays the text back with markdown asterisks, hashtag headers, and URLs spoken as words. The Oodlz inbound agent runs a dedicated voice pipeline in parallel with the text path so the audio sounds like a person talking, not a robot reading HTML.

## How It Works

When a request arrives marked as voice, the agent's response splits into two paths. The audio path runs through a voice-cleaning layer that strips all markdown, removes URLs and deeplinks, replaces them with a short spoken handoff sentence directing the user to the chat for clickable links, and restructures brand listings into natural spoken sentences — name, description, and cashback rate formatted for ears, not eyes. The cleaned text is sent to [text-to-speech](/docs/inbound-agent/every-channel/voice/text-to-speech.md) and returns as audio binary.

The same agent response is also written to Redis under a query ID with a 1-hour TTL. A voice client can call back with that query ID to retrieve the full markdown version — formatted brand cards, real deeplinks, intact links — and display it on screen alongside the audio.

Voice input runs in the other direction. Audio binary is captured by the client, transcribed via [speech-to-text](/docs/inbound-agent/every-channel/voice/speech-to-text.md), and the transcript is fed into the main agent pipeline with the same session reference as text turns. Conversation memory is shared across voice and text turns of the same session.

## What You Get

* Audio responses that sound like a person, not a TTS engine reading markdown.
* Brand listings restructured for the ear — natural sentences with pauses between brands.
* A clickable transcript on the side — full markdown with real links — retrieved separately by the voice client.
* Voice and text turns sharing one conversation thread by session reference.