How to make text to speech sound human!
Using AI to improve your latest project is an affordable way to add voiceover to projects. If it doesn’t give you what you need, at least it often gets you close.
In 2023 I started playing around with AI voiceover options and tried a few different tools including Murf and WellSaid Labs. Each one had options I liked – slowing down the rate of speech, manipulating intonation, adding pauses, etc. – but none had everything I wanted and I wasn’t completely blown away by the results.
AI voiceover significantly reduces costs by eliminating the need for costly recording sessions. – Harnessing the Power of AI Voiceover in eLearning
Then I tried Synthesys and while honestly, it really didn’t have these options at all, I felt that the end result was much more realistic. I was surprisingly impressed. Synthesys has recently gone through an update and they are adding some additional features which I haven’t fully investigated. But even without those, it has been my go-to text to speech voiceover tool. You should definitely check it out if you haven’t already.
My 5 Tips for Making Text to Speech Sound More Natural
AI Voiceover is getting better and better every day. But for now, here are 5 tips for making AI voices sound more natural.
Add contractions to your text
When we write, particularly for our work, we tend to write in a more formal way then we speak. And when we create AI voiceover from that writing, it can sound robotic, which is exactly what we are NOT going for! Make sure when you add your text to the voiceover tool that you change “do not” to “don’t” and “I will” to “I’ll”, etc.
If your project has text on screen (for example, in an e-learning module) you can even just leave the onscreen text as is and just make the voiceover sound more relaxed and natural by adding contractions where you can.
I love the ai voiceover tips from Daisy Rogozinsky to “Type acronyms as they’re pronounced. For example: ‘Innov-X’ becomes ‘In-oh-vex’.” Ironically, making your script less human readable is exactly how you will make AI text to speech sound natural.
Edit the pauses between sentences so that they vary in duration
In natural speech, the pauses between sentences are not all the same length. Unfortunately, I’ve found that a lot of AI tools create voiceover that does have the same pause length between each sentence, thereby adding to the robotic nature of the speech.
My favorite tool, Synthesys, did not always have the ability to vary pause length (which they now do as of this month!), so I just use a quick workaround. After creating and downloading the VO file from the AI tool, go ahead and use whatever sound editing tool you have and make a few quick edits.
I use Descript. While not specifically a tool designed for editing ai voiceover, it serves this purpose very well. I firmly believe that you will have an easier time making text to speech sound more natural if you use the tools that are familiar to you.
Vary the length of the pauses between each sentence, or even within a sentence. Play around with it and it really will add to the speech sounding a lot more natural.
International AI voices aren’t just for other languages
In the world we live in, no matter the language we speak, people have a wide variety of accents. For a recent project we were working on I used one of the international voices available, but entered English text into the tool. The result was a unique voice speaking English with a beautiful accent, and I think it added to making the group conversation in our project much more realistic.
Generate text-to-speech early in your project
In the early days of our project work in Articulate Storyline and Articulate Rise, we built elearning projects with the basic text-to-speech generation included in the tools. It sounded ok as a placeholder but wasn’t up to the level of quality that our customers expect from custom elearning design and development.
We recently had a client who wasn’t 100% able to envision where the project was going, so we ditched our robot friend and polished up the early version of the project with one of these new AI voices. It was a subtle, but successful, way to show them what the end product could be. This helped keep the focus on the details that needed to be addressed at that stage, until we inserted the final human voiceover in place in the end.
Don’t give up on AI voices
Don’t think when you enter text into an AI voiceover tool and create the file, that you are going to get the best outcome the first time around. I won’t claim to be an expert on AI voices, but I do know that you can render the clip over and over and get different results. Sometimes it will sound the same the first few times you do it. But often small nuances in the language will change and sometimes this is all you need for it to sound more natural.
I created a clip on Synthesys and it just wasn’t what I wanted.
I tried a few more time and after rendering it about five times, the AI voiceover artist literally chuckled.
It was so realistic that I jumped out of my seat! So, be persistent in trying to get what you need out of AI voiceover. It might not end up being what you’re hoping for, but chances are you get a surprise.
How can you make AI voices sound more human?
Integrating AI voiceover into your eLearning development workflow offers undeniable speed and cost benefits, allowing you to scale training content faster than ever before. However, the raw output isn’t magic; it often requires patience and a “human in the loop” to truly shine.
The difference between a robotic monologue and an engaging guide lies in the details. By mastering specific editing techniques—such as using phonetic spellings for tricky proper nouns, adding contractions to smooth out stiff phrasing, and strategically adjusting pause lengths—you solve the puzzle of how to make text to speech sound more natural.
It is not enough to simply copy and paste a script. Ultimately, taking the time to refine pitch, tone, and pacing is the secret to how to make AI voice sound more human. When you layer these technical adjustments with a well-written script, you ensure your learners remain engaged by narration that feels authentic, warm, and effective.
Disclosure: Endurance Learning is an affiliate of Synthesys, Descript and WellSaid Labs. Our team earns a small commission if you click through and make a purchase.
