> ## Documentation Index
> Fetch the complete documentation index at: https://docs.callivox.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Voices and tuning
> Preview and pick the voice your agent speaks with, and understand the advanced settings behind the recommended defaults.
Two things decide how an agent sounds on the phone: the voice you give it, and a set of timing
values that govern when it speaks, when it waits, and when it gives up waiting. The voice is worth
your time. The timing values already have sensible defaults.
## Picking a voice
The **Voice** field sits in the **Language & Voice** card on the **General** tab.
The voice list is filtered by language. Change the language later and you'll need to pick a
voice again.
Type to narrow the list. Each option shows the voice's name.
Press the play button next to an option to hear a sample before you commit.
Preview at least three candidates, and listen on a phone speaker rather than good headphones —
that's how your customers will hear it. A voice that sounds warm in headphones can sound thin
down a phone line.
The voice list is loaded from your connected voice service. If it can't be reached you'll see a
"couldn't load options, retry" message — wait a moment and reload.
## Advanced settings
The **Advanced settings** card on the **General** tab starts collapsed, and that's deliberate.
Every value in it already holds the setting our voice advisor recommends, chosen from what goes
wrong on real calls when they're set otherwise.
Treat these as recommended defaults. Change one when you have a specific problem to fix and
something to measure afterwards — not on a hunch. A wrong value here doesn't produce an error
message; it produces calls that feel subtly broken.
### The defaults that matter most
| Setting | Recommended | Why |
| -------------------------------------------- | ----------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Max silence duration (seconds)** | 60 | How long the agent tolerates silence before ending the call. Shorter values like 40 seconds cut calls off during natural pauses — while someone is fetching an order number or thinking. 60 seconds is the floor. |
| **Re-engagement interval (seconds)** | 45 | How long the agent waits before checking whether the caller is still there. At 30 seconds it re-engages so eagerly that it reads as impatient. |
| **Enable noise cancellation** | Off | It filters the caller's audio aggressively enough that the agent misses speech. Turn it on only when your callers are genuinely in noisy places. |
| **Turn detection threshold** | 0.5 | How much sound counts as the caller starting to speak. Move toward 0.3 for fast-paced conversations where people jump in, and toward 0.7 for noisy environments where background sound triggers false starts. |
| **LLM temperature** (on the **General** tab) | Set for you | How much the agent varies its wording. Speech-to-speech voice agents go flat and repetitive when this is set low, so a voice agent starts around 0.8 — much higher than you'd set a text agent. |
The **Re-engagement prompt** is the instruction the agent follows when the caller goes quiet. It
starts as:
> The customer has gone quiet. Politely check whether they are still there, then pick the
> conversation back up where it left off.
On an agent that's already live, the re-engagement prompt is left exactly as you wrote it and is
never quietly refreshed — it's spoken content, and changing it mid-flight would change what your
agent says to real callers.
### Everything else in the card
| Setting | What it does |
| ------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------- |
| **Allow interruptions** | Whether the caller can cut across the agent mid-sentence. On by default — a voice you can't interrupt feels like a robot. |
| **Max duration (seconds)** | The hard ceiling on call length. 10 minutes by default, and capped between 20 seconds and 20 minutes. |
| **Max silence duration (seconds)** | See above. |
| **Max initial silence duration (seconds)** | How long to wait for the caller to say anything at all at the start. |
| **Ringing time (seconds)** | How long an outbound call rings before giving up. 30 seconds by default. |
| **Turn detection threshold** | See above. |
| Setting | What it does |
| ----------------------- | --------------------------------------------------------------------------------------------------------------------- |
| **Voice stability** | How consistent the delivery is. Higher is steadier and flatter; lower is more expressive and less predictable. |
| **Voice similarity** | How closely the output tracks the original voice sample. |
| **Speech speed** | Between 0.7× (slower) and 1.2× (faster). Slow it down for older callers or technical detail; leave it at 1 otherwise. |
| **TTS emotion enabled** | Lets the voice carry emotional colour rather than reading everything at one pitch. On by default. |
Endpoint detection is how the agent decides the caller has *finished* speaking, rather than
merely paused.
| Setting | What it does |
| ------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Endpoint type** | **VAD** decides from the sound itself — fast and cheap. **AI** decides from what was said, which handles trailing-off sentences better but adds a little delay. |
| **Endpoint sensitivity** | 0 to 5. Higher means the agent jumps in sooner. Lower it if the agent keeps talking over people who pause mid-sentence. |
| **Interrupt sensitivity** | 0 to 5. How readily a caller's noise stops the agent talking. Lower it if a cough or a "mm-hm" keeps cutting the agent off. |
| **Min interrupt words** | 0 to 10. How many words the caller must say before it counts as an interruption. Raise it to ignore short acknowledgements. |
| Setting | What it does |
| ----------------------------- | -------------------------------------------------------------------------------------------------------------------------- |
| **Enable noise cancellation** | See above. Off by default. |
| **Fillers (verbal ums/ahs)** | Adds hesitation sounds. They make the agent sound more human to some listeners and evasive to others. Off by default. |
| **Record calls** | Whether the audio is kept. Off by default — turn it on to get a player in the [call detail pane](/voice-agents/call-logs). |
| Setting | What it does |
| ------------------------------------ | ------------------------------------------------------------- |
| **Re-engagement interval (seconds)** | See above. 45 seconds. |
| **Re-engagement prompt** | What the agent does about the silence. See the default above. |
| Setting | What it does |
| ----------------------------------- | -------------------------------------------------------------------------------------- |
| **End call on voicemail detection** | Hang up rather than talk to an answering machine. On by default. |
| **Voicemail message** | Leave this filled in and turn the switch off if you'd rather the agent left a message. |
Before turning on **Record calls**, check what you're required to tell callers about recording in
the places you operate. Many jurisdictions require an announcement or consent.
## Next steps
Listen to what these settings produced.
The rest of the editor.