Build your bots

Voices, audio takes and the Simulator

Voice Studio is where an account with AI Voice picks and installs the voices its AI Voice Agent speaks in. Pre-recorded bots play the recordings you upload, with optional takes, voices and state recordings, and the Simulator runs a script in your browser before it takes calls. The Simulator is a dry run: it places no call and never contacts your VICIdial® server, but it can write a real row to a Google Sheet. It does not test transfers, closer routing or the AI disclosure line.

Applies to:Every account

Last checked against the product: October 18, 2026. Found a mistake? Tell us.

Which voices does VoiceOI use, and where do I manage them?

Three separate voice lists feed different parts of the product.

List	What it holds	Where you use it
Voice Studio	The 215 ready-made synthetic voices, in the families VoiceOI Pulse and VoiceOI Kindred.	The Voice list and Rotate between voices in the AI Voice Agent editor.
Voices in the Script Builder	The roster of speakers (“actors”) that the VoiceOI team adds.	The Voices box under Audio Takes on a pre-recorded script. See audio takes and voice rotation.
Script Studio voices	A third list, used only if the VoiceOI team has turned Script Studio on for your account. The team adds voices to it.	The Voices choice on the Script Studio form. See ways to build a script.

How do I use Voice Studio?

Open Voice Studio in the sidebar. The search box counts the library, for example Search 215 voices.

Listen
You
Click a tile to play a short sample and click again to pause. You can hear any voice, including a family your plan does not include.
Narrow the list
You
Search by name, filter with Any family, Any accent, Any gender and Any quality, or switch between All, Ready and Not added.
Download the voice
You
Press the download button on the tile. It shows the pack size in MB, and its hover text says how many other voices the same pack installs. A pack is the download behind one or more voices. It goes onto every bot node in your organization (a bot node is a server that runs your bots).
Wait for Ready
Both
The tile shows Queued, a progress bar, then Ready. A voice is offered in scripts only after every node reports it ready. With no node online, a pop-up says so, and the download starts when a node checks in.
Use the voice
You
In your AI Voice Agent script, open Persona and pick it under Voice.

What a tile shows
On the tile	What it means
A line such as `Pulse · US · Medium`	Family, then gender (only where the voice has one), then accent, then quality (Low, Medium or High).
Default	A starting voice every account has. It has no remove button.
No audio	The sample file is missing, so clicking the tile does nothing. If a sample exists but will not load, a message starts “No audio for”.
Locked with a padlock	Your plan does not include this voice family.
Ready with a bin icon	Installed on every node and available in your scripts. The bin takes the voice off your list, so it is no longer Ready. The pack stays on your nodes while another voice from it is still Ready. Once the pack is gone, a script that still names the voice speaks in the default voice. For an AI Voice Agent, install the voice again rather than picking another, or saving the script saves the replacement.

While a pack installs, or after it fails, an Installing on your nodes panel shows each node with a tick, dots or a cross. “Not reporting” means the node is online but its installer has not checked in. Retry appears when an install failed.

The 215 voices and their families

The library has 215 voices in two families. VoiceOI Pulse has 161 voices and is included by default. VoiceOI Kindred has 54 and shows Locked unless the VoiceOI team has turned it on for your account. Only Kindred voices have a gender, so Any gender filters only those. Most voices speak English with a US, GB or UK accent. The 26 Kindred voices in other languages (Spanish, French, Hindi, Italian, Japanese, Portuguese and Chinese) show no accent, and Voice Studio has no language filter. Callers are understood in English only today, so choose US, GB or UK under Any accent and play the sample before you install a voice.

How do audio takes and voice rotation work?

A take is one recording of a line. A Play Audio, Listen, AM Detect or Age Extraction node has a main recording and can have more takes, added with Add Rotation Variation under Audio Takes. Turn the ROTATION switch beside that heading on only for a node that has added takes. What a call plays depends on whether the script has Voices.

Script setup	What each call plays
No Voices defined	At each node, one take picked at random from the main recording and the added takes.
Voices defined	Each call is assigned one voice and keeps it, spread across the eligible voices but not in strict turns. Every node plays a take tagged with that voice, picked at random.

A voice must be complete. It is eligible only if it has a take on every step that plays audio. The badge beside it, such as 3/5, counts the covered steps. Amber means “Missing takes on N step(s). This voice is skipped until it is complete.”
Tag every take. Each take has a Voice box, and so does the main recording. “Any voice (counts toward none)” counts toward no voice, and such a take is not played on a call that has an assigned voice.
Add voices from the roster. In Voices > Manage, pick a speaker under “Add a voice from the roster…”. For one who is not listed, ask the VoiceOI team.
Switch a voice off. Each voice has a switch (“Use this voice on live calls”). The bin removes the voice and turns its takes into “Any voice”.

AI Voice Agent: rotate between voices

In the AI Voice Agent editor, Persona has a Voice block with Voice, Speaking rate (0.80x to 1.30x) and Rotate between voices with Choose voices. With rotation on, each call is assigned one of the chosen voices and keeps it, and nothing needs recording per voice. The list shows only voices that are Ready, and a voice marked “(not in your plan)” or “(engine offline)” cannot be chosen. An empty list says “No downloaded voices yet. Install one in Voice Studio first.”

On the VICIdial (SIP) path, the agent introduces itself by the name of the voice the call was assigned, not by the Agent name from the script. Each voice in the list has a name box: type a name there to choose what that voice calls itself, or leave it empty and the voice uses its own library name, shown as the grey hint. Write {{agent_name}} in your wording so the right name is used (see Use placeholders instead of typed names). With no voices chosen (rotation off), the agent keeps the Agent name you typed.

How do I add audio files?

Open the upload window
You
On a Play Audio, Listen, AM Detect or Age Extraction node, press + next to the audio picker. The Add Audio window opens.
Add the file or record it
You
On Upload File, click or drag WAV, MP3 or M4A files. Several files are saved separately unless you switch on Merge into one recording. On Record Audio, press the microphone and speak.
Name it and write what it says
You
Type a Display Name. In Bot Transcription (Optional), type exactly what the recording says. The Simulator’s voice mode uses those words to tell the bot’s own voice, coming back through your speakers, from you.
Choose the language and save
You
Language starts as English (US). Press Save Audio.

A file can be up to 50 MB.
VoiceOI converts every upload to 8 kHz mono telephone audio, so play the saved recording to hear what callers hear.
Your plan can limit the number of audio files. Each take, each state recording and each recording in an imported bundle is a file.
Change a recording. The pencil (Edit Audio) next to the audio picker, and next to each take, renames a recording, trims it and changes its Bot Transcription. It changes the recording itself, so every node that uses it changes. The X (Clear Audio) takes the recording off the node and does not delete it.
Delete a recording. The Audio Library page is for VoiceOI staff only, so your login does not show it. Ask the team to delete a recording from the library.

How do state recordings work?

State recordings let one node play a different recording for callers in different US states. They appear only when the VoiceOI team turns them on for your account. The state comes from the area code of the number called. Calls in states you did not record, and calls where no state is found, hear the main recording.

Add the main recording first
You
Add state recording stays disabled until the node has one. The screen says: “Add a main recording first. It plays for every state you haven’t covered.”
Add the state recordings
You
In the node panel open State recordings. Press Add state recording and pick the recording and its states, or press Upload per-state files, which reads the state from each file name. Check it or pick one, because files without a state are skipped.
Tag the voice and save
You
If the script has Voices, choose the voice each recording is in. Then press Save Flow. A node holds up to 600 state recordings.

To test, pick Caller’s state in the Simulator before you start. A line under the bot’s message says which recording played, and Call Logs uses the same words for live calls.

Note under the bot’s line	What it means
“Texas recording”	That state’s recording played.
“Main recording: no caller’s state picked”	Caller’s state is Not set. A live call shows “Main recording: no state for this number”.
“Main recording: no Texas recording”	No recording covers that state.
“Main recording: Texas is only recorded in another voice”	The state’s recording is tagged to a different voice than the one this call was assigned.
“Main recording: the Texas recording isn’t on this bot server yet”	The bot server has not received it. Press Force Bot Sync in the sidebar footer and test again.

How do I run a Simulator test?

Get a campaign
VoiceOI team
The Simulator tests the script linked to a campaign, and the VoiceOI team creates campaigns. If Select Campaign to Test is empty, ask for one.
Link your script
You
In the Script Builder press the gear icon (Script Settings), choose the campaign under Assigned Campaign and press Save Changes. Use a test campaign (see the warning below) and keep one script on it: with more than one, the Simulator may run a different script from the one live calls run, and its screen does not say which.
Pick a mode
You
Open Simulator in the sidebar, which every role can use. Choose the campaign, then Text Simulator (Dry Run) to type the caller’s answers, or Browser Mic (WebSpeech API - Dry Run) to speak them.
Start and talk
You
Press Start Simulation. The chip at the top right changes from Offline to Live Session and the bot’s recording plays. Press Restart Session to begin again or End to stop. A pre-recorded script picks a voice and takes anew at each start, so run it several times. An AI Voice Agent test always uses the voice chosen under Voice.

Tags under the bot’s message show what happened: Smart Listen or Smart Reply decisions (see Smart Listen and Smart Reply), the state recording played, the Exit an AI Voice Agent took, and the Notes for the closer card at a transfer.

Browser Mic mode

It needs Chrome or Edge. Other browsers show “Voice mode needs Chrome or Edge. This browser has no speech recognition, so use the Text Simulator here.”
Silence Delay: Auto (as live) ends your turn the way a live call would at each node. Custom ends it after the pause you set, from 300 ms, on every node.
Using headphones: with it off, the microphone also hears the bot, so the bot’s own words are filtered out of what you said. Headphones give the most realistic test.
If the browser blocks sound (“Check Auto-Play permissions”), allow sound for the site and press Click to play.
Record Conversation works in both modes and asks your browser for microphone permission in either. When you press Stop Recording, your microphone and the bot’s audio download to your computer as a .webm file, which contains your own voice. The Simulator does not upload it.

What does the Simulator not test?

A passing test shows that the wording, branches and recordings behave, not that a live call will. Test on a real call to a phone you control before a campaign starts.

Part of the call	In the Simulator
Phone line, carrier and VICIdial	None is used. The Simulator never contacts VICIdial, so no status is written to a VICIdial lead, and the Vicidial Disposition set on nodes is not recorded.
Transfers, ZIP Resolver and Matrix Router	Nothing is transferred and the test ends. The Notes for the closer card shows what a live call would write into the lead’s Comments. ZIP Resolver and Matrix Router are not run. See Closer Hub and the routing steps.
The AI disclosure line	Not played, so the opening you hear is not the opening a caller hears. See AI disclosure.
AI Voice Agent voice	Always the voice chosen under Voice, while live calls rotate across the voices you picked. If the voice engine is slow, voice mode reads the reply in the browser’s own voice and marks the line “Spoken by the browser: the voice engine did not answer in time”.
Machine detection and the engine	The AM Detect node and the recognition engine are not tested, but phrases on Immediate Hangups are (see Phrase rules). Text mode uses what you type, and voice mode uses your browser’s speech recognition. Check AM Detect, engine choice and pauses on a real call.
Check Condition and the Interruption line	Both run in the Simulator. The VICIdial (SIP) engine ends a call at Check Condition and does not follow the Interruption line, while the built-in (Asterisk) engine runs both. See how the engines differ.
Max Retries	Not counted. A Listen node with no line on Any Other Input asks again every time.
Phrase rules	A phrase that matches a rule on Immediate Hangups (titled Automatic Call Rejection) ends the test with “Machine Detected. Hanging up...”, for a voicemail phrase or a do-not-call phrase alike.
Your numbers	The Dashboard leaves tests out, but Dialer Dispositions counts them. Each test is saved as a call against a test lead named Simulator User, so it appears in Call Logs and counts toward a call-log limit.

A Google Sheets step writes a real row and ends the test as a sale

When a test reaches a Google Sheets step with no line leaving it, the Simulator writes a row to the spreadsheet you chose and ends the test as SALE. The row is real, and lead columns show the test lead. It signs in with your latest Google connection, not the Integration Account set on the step. Point the step at a test sheet while you test.

It fills only columns that have a mapping. If a line leaves the step, as it does in an AI Voice Agent hand-off to a sheet, the Simulator follows the line, writes no row and does not end the test as SALE. On a live hand-off, the AI builds the row from the transcript using the sheet’s column headings, and an empty column or a failed write does not show on the call, so check the sheet after your first test calls.

The test says “Data synced to Google Sheets. Simulation ended.” even when no spreadsheet is set or the write failed, so check the sheet. See the Google Sheets step.

How do I export or import a script with its recordings?

In the Script Builder header, Export downloads a .voiceoi bundle and Import reads one back. Use them to keep a copy before a big change or to move a tested script.

What the bundle holds. The script as it is on screen now, with node settings such as transfer destinations and spreadsheet IDs, every recording it uses (including takes and state recordings), its Voices list with the tags on the takes, and the name of its reply library. Share a bundle only with people who may see those settings. A deleted recording is left out, and the download message says how many.
Import asks first. “Importing a .voiceoi bundle will clear the current canvas. Continue?” Recordings your account already has are reused, and each new one counts toward your audio file limit. Review the script, then press Save Flow.
One failure stops it. If a recording cannot upload, for example at the file limit, the import stops, tries to remove what it uploaded and imports nothing.
A `.json` script carries no recordings. It asks “Importing a script will clear the current canvas. Continue?” first, then keeps only the recordings already in your library and removes the rest, with a message saying how many.

Can I clone a voice?

Voice cloning is a service the VoiceOI team turns on for an organization. There is no cloning screen in the dashboard, and Voice Studio does not clone, so ask the VoiceOI team.

What do I do when something goes wrong?

A voice in Voice Studio shows Locked

Likely cause: Your plan does not include that voice family.
How to check: Hover over Locked.
How to fix it: Ask support to turn the family on, or choose a Pulse voice.
Still stuck: Your organization name and the voice name.

A voice stays Queued or Failed and never becomes Ready

Likely cause: A voice is Ready only after every bot node has installed its pack. An offline node, a node that has not checked in, or a failed download holds it back.
How to check: Read the Installing on your nodes panel beside each node.
How to fix it: Press Retry on a failed pack. If a node stays “not reporting” or nothing changes, contact support.
Still stuck: The voice name, what the panel shows beside the node, and any red message.

The Simulator says “This voice is not installed on this node”

Likely cause: The voice the step names is not installed, so the test used the default voice. A script that names a voice you removed can do the same once its pack has left the nodes.
How to check: See whether the voice is Ready in Voice Studio.
How to fix it: Install it, wait for Ready, then restart the session.
Still stuck: The script name and the voice name.

An audio upload fails

Likely cause

“This file isn’t a recording” or “This file couldn’t be converted to call audio” means the file is not readable audio.

“Recordings can be up to 50 MB.” means the file is too large.

“You have reached your audio file limit of N files” means your plan’s limit is full.

How to check

Read the message in the red notice.

How to fix it

Save the audio again as WAV or MP3, trim it, or delete files you no longer use. For the file limit, contact support.

Still stuck

The exact message and the file name.

The Simulator says “No script found for this campaign”, or the campaign list is empty

Likely cause: No script is linked to the campaign, or the VoiceOI team has not created a campaign.
How to check: On Scripts, see whether Campaign Assignment says Unassigned.
How to fix it: Link the script under Assigned Campaign in Script Settings, or ask the team for a campaign. “Script has no steps” means add steps first.
Still stuck: The script name and the campaign name.

Start Simulation fails with “Call log limit reached”

Likely cause: Each test is saved as a call and counts toward the call-log limit on your plan.
How to check: The notice says only “Call log limit reached”. It does not show the number.
How to fix it: Ask support to clear old calls or raise the limit. VoiceOI staff delete call logs.
Still stuck: Your organization name and the time of the test with time zone.

The Simulator shows “Failed to send message” or stays on “Thinking”

Likely cause: Each turn has a time limit. If the AI or the voice engine is slow, the Simulator answers with the fallback line or with text and no audio. “The simulator is still finishing the last turn. Try again.” means the last turn had not ended.
How to check: Wait a few seconds and send again.
How to fix it: Press Restart Session. If an AI Voice Agent test keeps failing, look at In effect now on the LLM provider tile of Integrations, and ask support before you save a new key or model, because that changes every AI script in your organization. See the AI model behind the agent.
Still stuck: The time of the test with time zone, the campaign and the message shown.

Browser Mic mode says “The bot took more than 10 s to answer, so the simulator stopped waiting”

Likely cause: The AI or the voice engine did not answer in about 10 seconds.
How to check: See whether it happens on one turn or on every turn.
How to fix it: Speak again. If it repeats, press Restart Session, then contact support.
Still stuck: The time of the test with time zone, the campaign and the step name.

Record Conversation says “Could not access microphone for conversation recording.”

Likely cause: The browser blocked the microphone or none is connected. Record Conversation needs it in both modes.
How to check: Look for the microphone icon in the address bar.
How to fix it: Allow the microphone for the site, then press Record Conversation again.
Still stuck: The browser and version.

My Google Sheet has no new row after a Simulator test

Likely cause: The step has no spreadsheet or column mapping, the write failed, or a line leaves the step, as in an AI Voice Agent hand-off, which writes no row. See A Google Sheets step writes a real row.
How to check: Open the step and check the spreadsheet, tab and column mappings, and that Integrations shows the connected Google account for Google Sheets.
How to fix it: Set the spreadsheet, map the columns, remove any line that leaves the step, and sign in to Google again if needed.
Still stuck: The script name, the sheet tab and the time of the test with time zone.

Common questions

How many voices does VoiceOI have, and can I listen before I choose?

The library has 215 ready-made voices: 161 in the VoiceOI Pulse family and 54 in VoiceOI Kindred. Voice Studio plays a sample of any voice, including a family your plan does not include.

Does the Simulator call my VICIdial server or place a phone call?

No. The Simulator runs the script in your browser with no phone line, carrier or dialer, and it writes no status to a VICIdial lead. It does not test transfers, closer routing, the AM Detect node or the AI disclosure line.

Why did my Simulator test add a row to my Google Sheet?

When a test reaches a Google Sheets step that has no line leaving it, the Simulator writes a real row to the spreadsheet you chose and ends the test as a sale. Point the step at a test sheet while you test.

Why does my pre-recorded bot sound like different people on one call?

With no Voices, each node picks a take at random, so different speakers can play on one call. With Voices, a call keeps one voice, but a voice is skipped until it has a take on every audible step. Check the badges, such as 3/5, and tag every take with its voice.

Can I clone my own voice or a voice actor’s voice?

Voice cloning is a service the VoiceOI team turns on for an organization, so ask the team, as there is no cloning screen in the dashboard. A cloned voice needs the speaker’s name and your relationship to the speaker, which VoiceOI does not verify, so the speaker’s authorization stays with you.

VoiceOI provides tools. The customer is the caller and is responsible for consent, do-not-call lists, calling hours, caller ID, recording and monitoring consent, and disclosures. This page is not legal advice.

VICIdial is a registered trademark of the Vicidial Group. VoiceOI is an independent product and is not affiliated with or endorsed by the Vicidial Group. Asterisk, Twilio and other product names are trademarks of their respective owners.

Questions about your setup? Talk to the VoiceOI team.