> ## Documentation Index
> Fetch the complete documentation index at: https://docs.wittify.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Voice Agent Settings
> Step 2 (Hybrid) or Step 1 (Voice-only). Pick a voice, write voice instructions, tune call behavior, add background sounds, and dial in advanced fields like silence detection and call duration limits.
This step is for **Voice and Hybrid** agents. Text-only agents skip it entirely. Voice-only agents start here, so they get the **Agent Name** field on this page (Hybrid agents already named the agent in Step 1).
## What this page looks like
Five cards stacked in the main column, in order:
1. **Voice Setting** , the voice picker, language detection toggle, and interruption toggle
2. **Core Instructions** , the voice version of the system prompt
3. **Creativity** , temperature slider for voice answers
4. **Advanced Voice Settings** , collapsed by default, 7 fine-tuning fields
5. **Background Sounds** , 5 ambient tracks plus a volume slider
On the right (when your screen is wide enough), a live preview of the **Wittify chat widget in voice mode** updates as you edit. The voice call screen is the part of the widget that turns from a chat bubble into a phone-call interface.
## Agent name *(Voice-only variant only)*
Voice-only agents add the **Agent name** field at the top of this page, since they skipped Text Agent Settings. Same rules as the text variant: internal use only, max 100 characters, error **Agent name is required** appears if you blur it empty.
Hybrid agents don't see this field here, they named the agent in Step 1.
## Voice Setting card
### Selected voice
A panel shows the currently selected voice with:
* A **Play** button on the start edge (turns into a **Stop** square while audio is playing)
* The voice's display name (Arabic or English depending on the wizard locale)
* A **Change Voice** button on the end edge (brand-gradient styling)
If no voice is selected yet, the panel shows **No voice selected** instead of a name.
### Voice Picker modal
Click **Change Voice** to open the **Voice Picker** modal. It's 900px wide and works as a flat paginated list, **10 voices per page**.
| Control | Behavior |
| ---------------------- | ------------------------------------------------------------------------------------------------------------------ |
| **Search box** | Searches by voice name, voice ID, or Arabic name |
| **Gender dropdown** | Filters Male / Female / All |
| **Preview text area** | Up to 140 characters. Type your own copy and click any voice's play button to hear it speak that text. |
| **Pagination** | Brand-gradient button on the active page, prev / next arrows. The currently-selected voice always pins to the top. |
| **Use** action per row | Picks that voice and closes the modal. |
All voices in the catalog are **Wittify voices**. Audio previews are generated on the server, you don't need to set up anything to use them.
### Advanced voice tuning (where available)
Some voices expose extra tuning knobs. When such a voice is selected, a chevron expands the row to show:
| Setting | Range | What it does |
| ----------------- | -------- | -------------------------------------------------------------------------------------------------- |
| **Model** | dropdown | Picks the model variant that powers the voice (each variant trades quality for speed differently). |
| **Stability** | 0 to 1 | Lower = more emotion and variation, higher = more consistent reads. |
| **Similarity** | 0 to 1 | How closely the output matches the original voice. |
| **Style** | 0 to 1 | Style exaggeration. Higher pushes the voice's personality further. |
| **Speaker Boost** | toggle | Enhances the speaker's distinctive characteristics. |
If the selected voice doesn't expose these knobs, the chevron simply isn't shown.
### Auto-detect language
A toggle below the voice picker. When **on**, the agent listens for the language the caller speaks and replies in the same language. Helpful for multilingual customer bases.
### Allow interruption
A toggle. When **on**, the caller can talk over the agent and the agent will stop speaking and listen. When **off**, the agent always finishes its sentence before listening. Most natural-feeling calls have this **on**.
## Core Instructions card *(Required)*
The voice version of the system prompt. This is the same idea as in Text Settings, but the prompt is tuned for spoken output.
### Buttons in this card
* **Edit Instructions** , a pencil icon button that opens the **Prompt Editor** (the same editor used in Text Settings, with up to 2,500 characters, **Enhance with AI**, **Keep / Revert**, quick template chips, and the **More templates** dropdown).
* **Copy from Text** , for Hybrid agents, copies whatever you wrote in the Text Settings core instructions into the voice prompt. Disabled until the Text prompt has content. A toast confirms **Copied from Text**, and a tooltip explains the action: *Copy your text agent instructions and tune them for voice.*
When the voice prompt is empty, the card shows the placeholder **No instructions yet**. Once you save text, the placeholder is replaced with the actual content.
### Phrasing for voice
* Shorter sentences than text (a long sentence sounds robotic out loud).
* No markdown (asterisks and headings are read literally).
* Natural pacing , add commas where you want pauses.
* Spell out abbreviations the voice should pronounce (e.g. write *VAT* if you want it spelled out, write *value-added tax* if you want it spoken).
If you leave Instructions empty and click away, the field shows a red border and **Instructions are required**.
## Creativity card
Same as the text creativity card, but it controls **voice temperature** specifically. The voice agent can be tuned to a different creativity level than the text agent (helpful when, for example, you want voice replies to be more formal than chat replies).
| Slider value | Label | Use for |
| ------------ | --------------------- | ----------------------------------------------- |
| 0.0 to 0.3 | **Precise** (green) | Policy bots, IVR menus, formal customer service |
| 0.4 to 0.6 | (middle) | Balanced. Most general voice agents. |
| 0.7 to 1.0 | **Creative** (purple) | Storytelling, character voices, marketing |
A help icon (**?**) next to the **Creativity** title shows a tooltip explaining the trade-off.
## Advanced Voice Settings card (collapsible)
Closed by default. Click to expand. Seven fields in a 2-column grid, each with a help icon and tooltip.
| Setting | Range | Default | What it does |
| ------------------------ | ------------------ | ------- | -------------------------------------------------------------------------------------------------------------------------------------------- |
| **Speaking speed** | 0.5 to 2.0 | 1.0 | How fast the agent speaks. Slower for narration, faster for ads. |
| **Min silence duration** | 0.5 to 5.0 seconds | 1.0 | How long the agent waits after the caller stops talking before treating it as the agent's turn. Lower = more responsive (but may interrupt). |
| **End call on silence** | seconds | 60 | Auto-hangup if neither party has spoken for this long. Set to 0 to disable. |
| **Max call duration** | 1 to 180 minutes | (unset) | Hard ceiling on how long any call can go. The agent ends the call when the limit is hit. |
| **Min speech duration** | seconds | 0.35 | Filters out very short sounds (coughs, "um", lip-smacks). Anything shorter than this isn't treated as speech. |
| **Widget popup delay** | seconds | 0.1 | How long after page load the voice widget appears on your website. |
| **Ask for feedback** | toggle | on | Whether the agent asks for a 1-to-5 rating after the call ends. |
**Most voice agents work great at the defaults.** The two fields most worth tuning early are **Min silence duration** (lower for more responsive agents, higher to avoid interrupting slow speakers) and **End call on silence** (lower for short transactional calls, higher for sales calls).
## Background Sounds card
A grid of five ambient tracks. Pick one, or pick **None** to disable. A volume slider sets the level from 0 to 100.
| Sound | Best for |
| ------------------ | ------------------------------------------------------------------------------------- |
| **None** | Sterile, minimal, headset-only feel |
| **Contact Center** | Ambient call-center hum (other operators in the background, occasional ringing) |
| **Keyboard** | Light typing in the background , makes the agent feel like a real person taking notes |
| **Hold Music** | Soft instrumental music, useful for hold loops |
| **Cafe** | Subtle cafe noise, casual feel |
Background sounds add a subtle realism to voice calls. Volume defaults to a low level so the agent's voice stays primary.
## The bottom action bar
Same as every other wizard step:
| Button | Behavior |
| ----------------------- | --------------------------------------------------------------------------------- |
| **Cancel / Exit** | Cancel before first save, Exit (with unsaved-changes guard) after |
| **Back** | Returns to the previous step (Text Settings for Hybrid, dashboard for Voice-only) |
| **Step dots** | Jump to any unlocked step |
| **Save draft** | Saves what you have without moving forward |
| **Next , *Appearance*** | Saves this step and moves to Step 3 (or Step 2 for Voice-only) |
## The wizard sidebar
The current step is highlighted in the sidebar's step list. The agent's name shows in the **Agent name** card at the top after your first save.
For a full walk-through of the wizard sidebar, topbar, tutorial, and bottom nav, see [Text Agent Settings](/en/Systems/AI-Agents/Wizard/Text-Settings).
## Variant differences
This is **Step 2**. The agent name was set in Step 1, so this card doesn't show on Hybrid. The next step is **Appearance**. Hybrid agents have 7 steps total.
This is **Step 1**. The **Agent name** field is added to the top of this page. The next step is **Appearance**. Voice agents have 6 steps total.
This step **does not exist** for text-only agents.
## Common questions
Open the Voice Picker, type some sample text in the **Preview text area** (up to 140 characters), and click the play icon next to any voice. The audio plays in your browser. Click **Use** when you've found one you like.
Voice cloning is part of the Wittify Faseeh product. See [Text to Speech , Clone Wizard](/en/Systems/Text-to-Speech/Clone-Wizard). Once cloned, the voice appears in the Voice Picker.
Lower **Min silence duration** further. The agent treats this much consecutive silence as "your turn".
Turn **Allow interruption** **on**. By default, voice agents finish their sentence before listening.
Raise **End call on silence** (default 60 seconds). For sales calls, 90 to 120 is common.
The Creativity slider on this page controls voice only. The chat creativity is on the Text Agent Settings page. They can be different.
The Stability / Similarity / Style / Speaker Boost knobs are only available on certain Wittify voices. Voices that don't expose those knobs simply hide the chevron, the voice still works exactly as well, you just don't tune the extra parameters.
Yes. Once a voice is cloned in Faseeh, it shows up in this picker for both web and telephony deployments. Telephony quality is also tuned by the **Speaking speed** and **Stability** values.
## Next
Brand the chat widget, set per-language text, configure the voice call screen.