Cloning your voice for AI calls: how to record, consent and set it up
Voice cloning needs a clean 1–2 minute recording of natural, varied speech, explicit consent from whoever's voice it is, and a few minutes in the Voice Library to create and name the clone before assigning it to an agent. It works best for an owner's or brand voice on reminder and receptionist calls — always test the cloned voice on a real call before using it on customers.
- You need explicit, informed consent from the person whose voice is being cloned — never clone a voice without it.
- A quiet room, natural pace and varied sentences in the sample produce a noticeably better clone.
- Cloning suits an owner's or brand voice on receptionist, reminder or follow-up calls.
- Always test the cloned voice on a real call before it ever reaches a customer.
- Cloning is part of your BYOK voice provider account — Convo360 adds no separate markup.
Prerequisites and consent
Before recording anything, get clear, informed consent from the person whose voice will be cloned: usually the business owner, a senior staff member, or a professional voice actor engaged for the purpose. Never clone a customer's, employee's or public figure's voice without their explicit agreement; this is both an ethical line and, in most jurisdictions, a legal one. Keep a record that consent was given.
You'll need a voice provider key already added under Integrations → Provider keys, and a quiet space to record.
A simple written note (who agreed, for what purpose, on what date) is usually enough for a small business; larger teams may want a short signed form. Either way, keep it somewhere retrievable, not just remembered.
Recording a good 1–2 minute sample
- Record somewhere quiet, with no fans, traffic or other voices in the background.
- Speak at a natural pace, the way you'd actually talk to a customer, not a slow deliberate 'recording voice'.
- Include varied sentences: statements, questions, and a few numbers or names, since real calls will need all of these.
- Aim for 1–2 minutes total, enough for the model to learn the voice's texture without needing a long studio session.
- Avoid reading the same sentence twice; repetition doesn't add much and wastes the sample.
- Turn off background music, fans and air conditioning where possible; even a quiet hum can be picked up and baked into the cloned voice's texture.
- If the voice will answer calls in more than one language (say Hindi and English), include a little of each in the sample so the clone handles both naturally.
The exact click path
- Open the Voice Library under your voice provider settings.
- Upload or record your 1–2 minute sample.
- Name the cloned voice clearly (e.g. 'Owner, Hindi/English') so it's easy to find later.
- Once processed, open the voice agent you want to use it on and select the new cloned voice from the voice picker.
- Save and make a test call.
Cloning usually takes a few minutes to process after upload. There's no need to keep the tab open, just check back and confirm the voice shows as ready before assigning it anywhere.
Worked example: an owner's voice for reminder calls
A salon owner wants appointment reminder calls to sound like her, not a generic voice. She records a 90-second sample in her back office after closing (quiet, no music playing), naming it 'Priya, reminders'. She assigns it to the outbound reminder agent, keeping the knowledge-base-driven booking agent on a standard voice. Customers hear a personal touch on reminders while the busier FAQ line stays on a standard, always-available voice, and the whole setup takes less than half an hour from recording to a working test call.
This split also limits risk: if the cloned voice ever needs to be retired, say if the owner leaves the business or wants a refresh, only the reminder agent needs updating, not every voice touchpoint the business runs.
Troubleshooting
| Symptom | Likely fix |
|---|---|
| Cloned voice sounds robotic or muffled | Re-record in a quieter space with a better microphone; background noise is the most common cause. |
| Voice doesn't appear in the agent's picker | Processing can take a few minutes after upload; refresh and check the Voice Library shows it as ready. |
| Mispronounces certain words or names | Try switching languages/accent settings, or add a short pronunciation note in the agent's system prompt. |
| Someone requests their own voice be cloned without proper consent process | Decline until you have their explicit written consent on file. This is a firm line, not a formatting detail. |
| Clone sounds fine alone but odd mid-conversation | Test with the full script, not just the sample text, since tone can shift on longer sentences or emotional cues. |
If more than one of these shows up together, start over with a fresh recording rather than trying to patch a poor sample. Cloning is quick enough that re-recording is usually faster than troubleshooting a marginal result.
What to test before going live, and cost
- Make a real test call and listen on a phone, not just a laptop speaker, since call audio compresses differently.
- Test at least one number, one name and one longer sentence from your actual script.
- Confirm the cloned voice is only assigned to agents where you have consent for its use.
- Have someone who knows the real person listen to a call and confirm it still sounds natural, not uncanny, on an actual phone line.
Voice cloning runs through your own voice provider account (Sarvam, Smallest.ai, ElevenLabs or OpenAI, depending on what your key supports). It's billed at that provider's own rate, with Convo360 adding no markup. You still need a Pro plan (₹2,999/month, excluding GST) or higher to use AI voice calls at all.
A good habit for growing teams: keep a short internal log of whose voice was cloned, when consent was given, and which agents use it. It takes two minutes to maintain and saves an awkward conversation later if someone asks where a call's voice came from.
Step by step
- Get consentConfirm the person being cloned agrees, and keep a record of that consent before recording anything.
- Record a clean sampleRecord 1-2 minutes of natural speech in a quiet room, with varied sentences rather than a flat script.
- Upload to the Voice LibraryOpen Voice Library, choose Clone a voice, and upload the recording.
- Assign it to an agentOnce processed, select the cloned voice for a voice agent like any standard voice option.
- Test call and refineCall the agent yourself, check pronunciation and tone, and re-record if quality needs improvement.
Example conversation
Explore the Voice Library
Start free in minutes — no code, no credit card. Bring your own AI keys and pay no per-message markup.
Frequently asked questions
How long a recording do I need to clone a voice?
About 1 to 2 minutes of clean, natural speech is enough for the Voice Library to create a usable clone. Record in a quiet room at a natural pace, include a mix of statements, questions and a few numbers or names, and avoid repeating the same sentence, since variety trains the model better than length does.
Do I need consent to clone someone's voice?
Yes, always. Only clone a voice with the explicit, informed consent of the person it belongs to, and keep a record that consent was given.
Which voice providers support cloning?
Cloning is managed through your connected voice provider account in the Voice Library; Convo360 itself adds no extra cost or markup on top of your provider's rate.
Can I use a cloned voice for only some calls?
Yes. Assign the cloned voice to specific agents, for example reminder calls or a receptionist line, while keeping busier or knowledge-base-driven agents on a standard voice. This also limits risk, since retiring the cloned voice later only means updating the agents that use it.
Why does my cloned voice sound different on a real call than in testing?
Phone call audio compresses differently from a laptop speaker or headphones — always do a real test call, not just a sample playback, before going live.
Related reading
Available across India — see WhatsApp & AI calling automation by state and city.