Your HeyGen avatar looks convincing, but the voice still sounds like off-the-shelf stock audio? Many creators know this problem. HeyGen's standard voices are usable, but they often seem flat and impersonal. If you want your avatar to speak with a voice that actually sounds like you, you have two options: HeyGen's built-in audio function or external services such as ElevenLabs and MiniMax.
The video is only loaded from YouTube when you play it – data (including your IP address) is then transmitted to YouTube/Google. More in the privacy policy
In this tutorial, I'll show you how to integrate your voice into HeyGen, what results the built-in options deliver, and where the quality of external providers helps significantly. You'll also learn how much audio you need at a minimum and what to look out for in the settings.
Key Takeaways
- HeyGen offers a built-in audio function that lets you upload your own recordings or record directly. The lip movements adjust automatically.
- In the test, the internal voice doesn't sound entirely like the original. For a credible voice, an external provider is therefore worthwhile.
- With Instant Voice Clone, ElevenLabs offers a quick option that works with just about 10 seconds of audio. For high quality, the Professional Voice Clone is the better choice, with at least 30 minutes of material, ideally around 2 hours.
- You download the finished audio file and then upload it as a new audio track in HeyGen. After that, you can have the gestures adapted to match the voice.
Step 1: Prepare the Avatar and Text
Open your avatar in HeyGen and select the desired format, in this example landscape. Then enter the text your avatar should speak. In the video, I use an advertising text for this, but you can use any text you like. Make sure the text sounds natural and the sentences aren't too long, otherwise the emphasis later on will seem restless.
Next, you can choose a gesture that fits the content. This makes the avatar in the video more lively, even if the voice isn't perfect yet.
Step 2: Select the Voice
In the upper area of the interface, you'll find the voice selection. Here you can use voices that have already been created. In the example, I prepared a test version of my cloned voice as well as a few other voices. When you select a voice, you can listen to it directly before generating the text.
While listening, pay attention to whether the emphasis, tempo, and timbre suit you. With cloned voices in particular, it quickly becomes apparent whether the result is usable.
Step 3: Use Your Own Audio Recording Internally
In the Audio section, HeyGen offers the option to upload your own file or record directly in the browser. For the test, I used an older voice recording of about 15 seconds. To do this, upload your file or start the recording, speak the text, and confirm the selection.
After uploading, HeyGen recognizes the emotions in your voice and automatically adjusts the playback. As a result, the avatar no longer seems so monotone. You can activate the option if you want the expression of your voice to be carried over.
Step 4: Transcription and Overwriting the Text
As soon as you confirm your recording, HeyGen automatically transcribes the audio. The previous text in the input field is overwritten and no longer matters afterward. This is important to know, because you lose your original text. Copy it beforehand if you still need it.
After the transcription, check whether the recognized text is correct. Technical terms and names are sometimes spelled incorrectly. A quick check saves you rework later.
Step 5: Check the Lip Movements
Now you can render the video and test the playback. In my test, the lip movements adapted well to the recorded speech. Play the scene several times and watch for places where the mouth and sound don't match up cleanly.
With your own voice, the movements usually look significantly more natural than with standard voices. The result is therefore especially worthwhile when the avatar is meant to embody both your face and your voice at the same time.
Step 6: Use External Providers
The internal solution has its limits. If you don't want to make long recordings yourself, it's worth taking a look at external providers. One example is MiniMax (minimax.io). There, in the Audio section, you'll find the Voice Cloning function. You upload your voice as an MP3 file or record it directly.
In addition, MiniMax offers settings with which you can remove background noise or refine an accent. Then you choose which voice fits your project best. Afterwards, you enter the text and have the voice output generated.
Step 7: Set Up ElevenLabs as the Market Leader
For the highest sound quality, I tested ElevenLabs. The provider is considered the market leader in speech synthesis. In the video, I use the Creator plan. A cheaper entry-level tier called Starter is also available; in the video it is given as about 5 US dollars. Check the current prices yourself before signing up, as they may change.
After signing up, you open the Voices section and select the option Create or clone a voice. Here you decide which variant you want to use.
Step 8: Choose the Right Cloning Variant
ElevenLabs offers two variants. The Instant Voice Clone is the quick solution and gets by with just a few seconds of audio. In the video, around 10 seconds of material is sufficient for this. However, the result doesn't sound very close to the original yet.
The Professional Voice Clone delivers significantly better quality. For this, you need at least 30 minutes of material, or even better, about 2 hours. In the video, I worked with one hour of audio. The processing took about an hour. So plan enough time.
When recording, make sure you're in a quiet environment, ideally without reverb and background noise. A clean recording is the most important foundation for a convincing clone.
Step 9: Generate Text and Adjust Settings
Once your voice model is ready, you select it and insert your text. Then you click Generate Speech. The processing is usually quick, and you immediately get an audio file back.
In the settings, you can adjust further parameters. In the video, I changed most of the values but left the speed at its original value. Try out different combinations and listen to each one. Make sure the emphasis doesn't become too artificial.
The video also compares the older, more stable model with a new alpha version of Model 3. The newer model sounds promising in some passages, but is still in the testing phase. Decide by your own ear which model suits your project better.
Step 10: Try Alternative Voices
If you don't necessarily want to use your own voice, ElevenLabs offers many other voices. In the video, it's best to filter by German language. I liked the voice Felix. Here too: have the same text generated multiple times to hear the differences directly.
Step 11: Download the Audio File and Integrate It into HeyGen
When you're satisfied with the result, you download the file via the Download button. Then you go back to HeyGen and add a new audio track. To do so, select the option Upload Audio and choose your downloaded file. In the video, the file had already been uploaded beforehand, so I could use it directly.
Next, you enable the option so that the gestures are also adapted to the voice. Then you click on Add Audio. HeyGen creates a transcript in the background while you continue working.
Step 12: Check the Final Result
Finally, you start playback and evaluate the result. The voice in the video sounds significantly better than the internal test recordings, but it is still not completely identical to my real voice. This is a realistic picture of the current state of the technology.
Evaluate your result honestly. If the voice is not convincing, adjust the recording quality or the settings and regenerate the file.
Summary ??? Cloning an AI Voice with HeyGen: Guide for Your Own Avatar
You can integrate your own voice into HeyGen either by using the internal audio function or by using an external provider such as ElevenLabs or MiniMax. The internal solution is fast, but does not always deliver the desired sound fidelity. With the Professional Voice Clone, ElevenLabs offers better quality, but it requires significantly more audio material and processing time. Then upload the finished file to HeyGen, check the lip movements, and adjust the gestures. This way you avoid the typical robotic sound and get an avatar that sounds like you.
