ElevenLabs Voice Cloning: clones from 10 s, 30 s and 2 min

👁 125 reads

AI voice cloning in ElevenLabs means creating a copy of a voice from tens of seconds or a few minutes of audio, then reading any text with that copy. We tested it exclusively on a synthetic voice: first we designed a non-existent narrator with Voice Design, then cloned it from 10 s, 30 s and 2-minute samples. Below you will find the audio, similarity measurements, costs and the rules you should not ignore.

01ElevenLabs Voice Changer: AI Voice Swap Test & ResultsVoice Changer keeps your timing and melody while swapping the voice. We tested it on clean and noisy recordings - here's what survived and what didn't.Read →02ElevenLabs Dubbing v1 vs v2: Test on a 24-Second ClipElevenLabs Dubbing v1 vs v2 tested on a 24-second clip: time, credits, voice similarity, and one failure that still cost credits.Read →03ElevenLabs Agents Voicebot Test: Dental Reception LineWe built a dental reception voicebot with ElevenLabs Agents and tested it with real-time voice calls: response time, speech recognition, cost per minute.Read →04ElevenLabs Scribe: English transcription test with hard namesElevenLabs Scribe v2 transcribed a 59-second English interview with tricky names and 12 numbers in 3.6 s with 2 automatic differences, correctly attributed 6 of 6 turns.Read →05AI Voice Generator: Free & Paid Options for Voice CloningAI voice synthesis has evolved from robotic to natural. We compare free system voices, cloud engines, and commercial cloning services.Read →

First rule: someone else’s voice only with consent

Cloning someone else’s voice requires that person’s consent. ElevenLabs states that it reviews every reported violation of voice rights. Professional Voice Cloning (PVC) can only be done with your own voice: according to ElevenLabs help, even another person’s consent is not enough, because every PVC goes through verification that the voice belongs to the account owner. If someone wants to share their voice with you, they create and verify the clone on their own account, then share it via a private link. That is why we used no real recording in this test. We write more about the risks in the article your voice no longer belongs only to you.

Instant vs Professional Voice Cloning

The documentation warns that IVC handles unusual voices and accents less well, and you cannot change the accent or tone of a clone after creation. You can only replace the samples. Plan prices are without VAT, checked 29 September 2026.

Our test step by step

Step 1: a voice that does not exist. In Voice Design (model eleven_ttv_v3) we described a narrator in English and gave an English preview text of 354 characters. The documentation advises starting the description with language and accent, e.g. in the pattern “Native English. Male, 50s. Studio quality”, so the voice does not drift. We received three variants. We chose the one with no transcription errors and the lowest median pitch (104.1 Hz). Description to copy:

An English man in his fifties with a low, warm, slightly raspy voice.
Calm, unhurried documentary narrator delivery, native British accent,
close-mic studio recording.

Step 2: samples. With that voice we generated a recording of five paragraphs about the Cairngorms using the Eleven v4 model, 120 seconds in total. From that file we extracted three samples: the first 10 s, the first 30 s and the whole thing.

Step 3: three clones. We sent each sample to Instant Voice Cloning without the noise filter (the sample was clean, and the documentation warns that the filter can then worsen quality). Each clone was created in a few seconds.

Step 4: the same text with all voices. The original, three clones and for comparison the stock male voice Brian read the same sentence with date and time. We measured similarity with the WavLM speaker verification model (1 means identical voice) and ran the recordings through Scribe transcription.

Results

All recordings were transcribed word for word. The only difference is the time format: for the original and the 10 s and 30 s clones Scribe wrote “eighteen forty-five”, for the ~2 min clone “18:45”. This is a formatting choice, not an error. The clone also speaks Spanish and keeps its timbre roughly as well as the original (0.79-0.81 vs 0.81).

To be fair: such high results at 10 s are largely due to the fact that the source was synthetic, i.e. without reverb, noise and changes in distance from the microphone. With a phone recording we expect a longer sample to make a bigger difference. This is our opinion. We did not test Professional Voice Cloning because it requires verification of the account owner’s voice, and our test voice has no owner.

How to clone your own voice in ElevenLabs

1. Record 1-2 minutes of speech in a quiet room, with one microphone, without long pauses. The PVC documentation advises peaks from -6 to -3 dB and an average around -18 dB. MP3 format is sufficient, WAV according to the documentation usually does not improve the clone.
2. In the app go to Voices, then Create a voice and Instant Voice Clone, upload the file. Do this only with your own voice or with the written consent of the person whose voice you are cloning.
3. Turn on the noise filter only if the recording actually has noise.
4. Test the clone on a sentence with numbers and proper names. If the accent does not match, replace the samples, because you cannot adjust the clone’s settings.

Via API it is one command (tested 29.09.2026 on a 10 s sample):

curl -X POST https://api.elevenlabs.io/v1/voices/add 
  -H "xi-api-key: $ELEVENLABS_API_KEY" 
  -F "name=My voice" -F remove_background_noise=false 
  -F files=@sample.mp3

In the response you get a voice_id, which you then use in text-to-speech. How to read longer texts with it is described in AI narrator, and how to change the voice on an existing recording in the article about Voice Changer.

How much does it cost

Creating an IVC clone itself used no credits in our test. You pay for each text read by the clone, just as with stock voices, e.g. in Eleven v4 via API 0.08 USD per 1000 characters, and in the promotion until 12 October 2026 0.022 USD. Voice Design cost us approximately 397 credits per call (usage statistics showed 2380 credits for 6 Voice Design calls, 3 previews each). The limit of your own voices depends on the plan: 3 in the free plan, 10 in Starter, 30 in Creator, 160 in Pro. A full comparison of plans is in the article how much ElevenLabs costs, and an overview of all features in the guide ElevenLabs.

Frequently asked questions

How much audio do you need to clone a voice in ElevenLabs?

The documentation recommends 1-2 minutes for an instant clone and 30-180 minutes for a professional one. In our test on a clean, synthetic voice, 10 s already gave 0.992 similarity, and 2 minutes 0.996.

Is voice cloning in ElevenLabs free?

No. Instant Voice Cloning requires the Starter plan at 6 USD per month, and Professional Voice Cloning the Creator plan at 22 USD. Creating an IVC clone itself used no credits in our test.

Can I clone someone else’s voice?

Only with their explicit consent and in accordance with the law. ElevenLabs does not allow you to make a professional clone of someone else’s voice at all, even with consent, because it requires verification of the voice owner.

Does the cloned voice speak multiple languages?

Yes. The clone works with multilingual models, including Eleven v4. Our clone read in English and Spanish, and transcription showed no content errors.

// read moreElevenLabs Use Cases: Voiceovers, Voice Cloning & AudiobooksGuidesElevenLabs Review: Models, Pricing & How to StartAI ToolsAI Voice Generator Comparison: 8 TTS Models Tested in PolishBenchmarks

The link to ElevenLabs is a partner link: if you buy through it, QUANTSLANT earns a commission and you pay the same price. It has no influence on the content, including where the coverage is critical.