Realtime voice
Realtime voice is the most natural way to run a phone agent in Hania. The agent listens and speaks through a single live AI connection, so turn-taking, interruptions, and emotional tone are handled inside the model as it talks. You turn it on by picking a realtime engine and a voice on the bot.
Turning it on
Open the bot's Voice settings and choose a realtime engine. Then pick a voice from the gallery. That is the whole switch. The bot keeps all of its tools, knowledge, memory, transfers, and call features, so you are only changing how the agent hears and speaks, not what it can do.
Engines and voices
Hania gives you four realtime engines to choose from. Each one comes with its own voice catalog, and you audition every voice with a play button in the gallery before you commit. Pick the engine whose voices you like and whose controls fit how much you want to tune the conversation.
| Engine | Voices | Good to know |
|---|---|---|
| Grok voice | 26 | The widest set of live-conversation controls, including a reasoning speed mode, playback speed, transcription hints, and an automatic re-engagement timer. See Voice tuning. |
| Gemini voice | 30 | The largest voice catalog. Turn-taking is tuned with start and end sensitivity settings. |
| OpenAI voice | 10 | A choice of two models: a standard one and a lighter, faster one. |
| Azure voice | Curated | Includes high-definition voices and dedicated Urdu voices. You can also pair a voice with a menu of underlying models, so natural speech and the reasoning engine are chosen separately. |
What every realtime call gets
Some things are handled for you on every realtime call, whatever engine you chose, so you do not have to switch them on:
- Noise suppression. Hania cleans up background sound on the caller's audio by default. You can turn it off per bot if your callers are in a studio-quality room and you want fully unprocessed audio.
- Accurate interruptions. When a caller talks over the agent, the agent stops and keeps only what the caller actually said, so its next answer stays on track.
- Session continuity. If the connection hiccups mid-call, Hania keeps the conversation going with its context intact. There is more on this in Call reliability.
Fine-tuning the conversation
Every engine exposes turn-taking controls: how sensitive the agent is to speech, how long a pause counts as the end of a turn, and how much lead-in audio it keeps. Grok adds several more, and each engine shows only the settings that apply to it. The full list, with ranges and examples, is on the Voice tuning page.
When to use the classic pipeline instead
Realtime voice is the right default for a conversational phone agent. Reach for the classic pipeline when you want a specific synthesized voice, a specific recognition engine for a particular language, or separate control over each stage of the call.