How to Create an AI Voice That Sounds Like You With ElevenLabs

Generative AI and deepfakes have collided with the development of AI voice tools. The idea is simple: you take a voice and manipulate it to speak the words you give it.

Leading the pack in this area is ElevenLabs, which offers a free-to-use AI voice tool.

What Is ElevenLabs?

Founded by an ex-Google machine learning engineer and an ex-Palintir deployment strategist, ElevenLabs is a voice technology research company. AI speech software is a key element of its strategy, but the final aim is to create a tool that “instantly convert[s] spoken audio between languages.”

MAKEUSEOF VIDEO OF THE DAYSCROLL TO CONTINUE WITH CONTENT

ElevenLabs has developed new text-to-speech models that can create a realistic-sounding human voice. Its website states: “Our mission is to make on-demand multilingual audio support a reality across education, streaming, audiobooks, gaming, movies, and even real-time conversation.”

Google Translate and its alternatives are one thing, but can you imagine a tool that instantly translates what you’re hearing? Cloning the voice of the speaker so that you hear the speech as they would say it is an important stepping stone towards that.

What Is AI Voice Generation?

Described simply, AI voice generation lets you take a voice and make it say whatever you want to hear. Simply choose a voice, provide dialogue, and the tool does the rest.

You might think “well, Microsoft Sam was doing that back in the 1990s” and you would be quite right. But Microsoft Sam and similar tools sounded like robots. ElevenLabs’ tool, meanwhile, sounds far closer to humans.

ElevenLabs offers three speech AI options: its completely free “premade” voices, the voice generator (allowing you to select sex, age, and accent) and the subscription-only “cloned” voices that you can upload.

Here’s an example:

Use of AI for creative purposes comes with some moral and ethical responsibilities and creating voices with ElevenLabs’ speech AI tool is no different.

In short, don’t use someone’s voice without their permission. While it’s not illegal, they might be upset about it.

Before you proceed, remember that at the time of writing, ElevenLabs’ speech AI tool is in beta. This means that it is not the finished product.

Generating a Basic AI Dialogue

The simplest way to use ElevenLabs is with the free speech AI tool.

To use this, go to beta.elevenlabs.io and create an account (you can use your own email, a Google account, or Facebook).

Click Speech Synthesis
Select one of the premade voices in Settings (male and female voices are available)
Expand Voice Settings to set Stability and Clarity + Similarity Enhancement (high stability is monotonal, high clarity closer to the intended voice) sliders
Select Eleven Monolingual (standard English)
Input the text you wish to convert to speech
Click Generate
Once the process completes, it should autoplay; if not, click Play

You can also Download the generated sample.

Generate a Voice With ElevenLabs

If you prefer to create a new voice, you can use the Add Voice button to visit the VoiceLab screen. To generate a new voice based on ElevenLabs’ presets:

Click Add Voice > Voice Design
Set the Gender, Age, and Accent fields
Adjust the Accent Strength slider as required
Input the text you wish to convert
Click Generate
When it’s done, have a listen

In testing, I found that both the Female/Young/Australian and the Male/Old/Australian accents were distinctly “American.” This is an issue that will probably be ironed out as the technology develops.

Creating Your Own Voice in AI

While the premade and configurable options are interesting, the really exciting element of ElevenLabs’ technology is the Instant Voice Cloning tool.

Unlike the other options Instant Voice Cloning requires a subscription. Several options are available, the cheapest being $5 a month. At the time of writing, this comes with an 80% discount for the first month, making it just $1.

Other options cost $22, $99, and $330 a month, with the possibility of generating up to 40 hours of audio per month.

To use Instant Voice Cloning, not only do you need some dialogue, you also need a sample of your voice. Anything will do, as long as it is clear, and in MP3 format. The longer the sample, the better, up to 5 minutes.

From the VoiceLab screen:

Click Add Voice > Instant Voice Cloning
In the resulting window, set a name
Click or drag a suitable file to upload a sample (up to 25 samples can be added for improved accuracy)
Click Labels and specify a key + value (e.g. Accent/British)–do this up to 5 times
Input a brief description of the voice
Check the consent confirmation check box then Add Voice

With the voice added, you can adjust it in the Speech Synthesis screen as above.

What Can You Do With an AI Voice?

AI speech with premade and cloned voices has numerous possibilities. As noted, ElevenLabs’ final aim is for live translation, but they’ve noted various other uses.

Audiobooks are mentioned (perhaps read by a long-dead movie star) along with video games (using AI speech would save on voice actors). But it has uses beyond this, from music to satire to self-help, and probably beyond.

You can even create a podcast using AI speech, although the results could sound flat and boring.

The introduction to an episode of our Really Useful Podcast was produced using ElevenLabs:

While the results weren’t quite what we’d hoped, it’s good enough to use, and the technology can only get better.

Meanwhile, ElevenLabs is planning a generated “voice conversation” feature to be introduced at a later date.

Use Your Voice in a New Way With ElevenLabs’ Speech AI

Artificial intelligence has brought us some amazing new tools over the past few years. Chat-GPT can be used to create text, answer questions, outline reports, and more. Midjourney is an astonishing tool that generates art based on prompts.

Now, the speech AI tool from ElevenLabs makes it easy to manipulate a voice. It’s like an impersonation, but with a clone of the original voice.

While there are ethical arguments against using voices without consent, this is a powerful tool with some interesting possibilities. Best of all, it’s surprisingly easy to use and delivers impressive results.

Source link