How to Use ElevenLabs to Clone a Voice for a Podcast Intro
You recorded a dozen podcast episodes before realizing the intro still sounds like a generic text-to-speech robot. The good news is that ElevenLabs can clone your actual voice from a short audio sample and let you regenerate that intro any time you update the script, without booking studio time or re-recording anything.
The voice clone is not magic. The quality of what comes out depends almost entirely on what goes in. If you feed it a noisy Zoom recording, you get a noisy-sounding clone. Feed it clean, breathy, natural speech and the result is genuinely hard to tell apart from the real you. That gap between mediocre and convincing is what most tutorials skip over, so let's start there.
Instant Clone vs. Professional Clone
ElevenLabs offers two paths: Instant Voice Cloning (IVC) and Professional Voice Cloning (PVC). They solve different problems.
| Feature | Instant Clone | Professional Clone |
|---|---|---|
| Training audio needed | 1 minute minimum | 30 minutes (ideal: 3 hours) |
| Training time | Under 30 seconds | Up to 4 hours |
| Output quality | Good for most use cases | Near-indistinguishable from source |
| Plan required | Free tier (limited) | Creator plan or above |
| Best for | Quick prototypes, podcast intros | Audiobooks, ongoing production |
For a podcast intro that runs 20 to 40 seconds, Instant Clone is almost always enough. I tested both on the same voice and the PVC version was noticeably more expressive on long sentences, but for a short punchy intro the IVC held up fine. Start with IVC, upgrade if you hear something that bothers you after the first few months.
Recording a Clean Sample
This is the single most important step. You want 2 to 5 minutes of yourself speaking naturally, not reading robotically.
Record in a quiet room, close to your microphone (4 to 6 inches for a cardioid condenser, 1 to 2 inches for a dynamic). Turn off fans, AC, and anything that hums. No music in the background.
Read the actual content you plan to generate later, or something with similar rhythm. If your intro uses short punchy sentences, record short punchy sentences. If it has a warm conversational tone, record warmly. The model learns the prosody of whatever you feed it.
Export as WAV at 44.1 kHz or MP3 at 192 kbps minimum. ElevenLabs accepts both. Avoid heavy compression or noise reduction in post before uploading because those artifacts bake into the clone in ways that are hard to predict.
Creating the Clone in ElevenLabs
Once you're logged in, go to Voices, then click Add Voice, and choose Instant Voice Clone.
- Give the voice a clear name (something like "YourName Podcast Voice" so you don't confuse it with other voices later).
- Upload your audio file. You can upload multiple clips; if you have several clean recordings from different sessions, adding two or three helps the model average out any room inconsistencies.
- Check the consent checkbox. ElevenLabs requires you to confirm that you have the right to clone the voice you're uploading. More on this below.
- Click Add Voice and wait. Instant cloning really is instant.
After the clone is created, run a short test generation from the Voice Lab before you use it in production. Type something you did not record and listen for any odd artifacts or vowel shifts. If you hear something weird, try uploading a different clip.
Stability and Similarity Sliders
Every voice in ElevenLabs has two main sliders that most people ignore and then wonder why the output sounds flat or erratic.
Stability controls how consistent the delivery is. High stability (around 0.75 to 0.85) means the voice reads calmly and steadily, which works well for intros where you want a confident, measured tone. Low stability (below 0.50) introduces more variation and emotion, but it can also produce random mispronunciations on shorter text.
Similarity Enhancement controls how closely the output matches your uploaded sample. Cranking this above 0.85 can make the voice sound slightly processed or over-sharpened. The sweet spot for most podcast intros is 0.70 to 0.80.
My recommended starting point for a 30-second podcast intro: Stability at 0.78, Similarity at 0.74. Generate the script once, listen, then nudge one slider at a time until it sounds right to you. Do not change both at once or you lose track of which change made the difference.
Ethics and Consent
This matters a lot, both morally and practically. ElevenLabs' terms are clear: you can only clone a voice you own or have explicit written permission to clone. Cloning someone else's voice without consent violates their terms of service and, depending on your jurisdiction, may violate laws covering voice likeness.
If you're cloning your own voice for your own podcast, you're fine. If you're a producer and want to clone a host who's going on parental leave, get written permission from the host, keep a copy of that consent, and limit usage to what they agreed to. The technology is useful; misusing it is both wrong and risky from a legal standpoint.
ElevenLabs also has a voice verification feature on Professional Clones that records a brief sample at cloning time to confirm the person being cloned is present. It's worth using even when not required, because it creates a clear audit trail.
Writing the Intro Script
A podcast intro is not the place for long complex sentences. Aim for 25 to 50 words, two or three punchy lines.
Bad example: "Welcome to our podcast where we discuss the complex and ever-evolving landscape of modern entrepreneurship."
Better example: "This is The Build Podcast. Every week we talk to founders who did it without funding, without luck, just work. I'm [name]. Let's get into it."
Short sentences with natural pauses work better with voice synthesis because the model handles punctuation cues well. A comma produces a short breath. A period produces a longer one. Use that to your advantage.
Avoid words with unusual spellings that the model mispronounces. If ElevenLabs consistently mispronounces your company name, use the Pronunciation Library in your account settings to add a phonetic entry (for example, map "Acme" to "ACK-mee" if it keeps saying "AC-may").
Generating and Exporting the Intro
Once you're happy with the script and settings, generate the audio from the Text to Speech tab. Select your cloned voice, paste the script, and hit generate.
Listen to the full output before downloading. ElevenLabs sometimes stumbles on the last word if it trails off. If that happens, add a period or a short silent beat at the end of your script text.
Export as MP3 (high quality) for podcast platforms, or WAV if you plan to do any further audio processing. Most podcast hosts accept MP3 at 128 kbps or higher. If your intro needs a music bed underneath it, export the voice track clean and layer the music in Audacity, Descript, or your usual DAW.
For a professional intro, match the output loudness to your episode standard. A common target is -16 LUFS integrated for podcast audio. You can check this in Audacity via Analyze, Measure RMS, or use a free plugin like Youlean Loudness Meter.
When to Re-Clone
Your voice changes over time, especially if you're recording regularly. I'd suggest re-cloning every six to twelve months if you notice a drift between how your live episodes sound and how the intro sounds. The process takes about ten minutes total, so it's not a big lift.
If you significantly change your microphone setup or the room you record in, that's also a good trigger for a fresh clone. The goal is for a listener to hear the intro and not be able to tell it's synthetic. When your gear changes, the gap can widen.
ElevenLabs keeps updating its underlying models, so an older clone created on a previous model generation sometimes sounds noticeably worse than a fresh one on the current model. Check your voice settings occasionally; there may be a "Recreate" option that ports your existing sample to the new model without re-uploading.
A clean voice clone and a well-written 35-word script will give you a podcast intro that sounds like you recorded it in a booth. The key is spending the time on the input: record clean, pick the right sliders, and write short crisp sentences. The tool handles the rest.