Meet Rachel; she’s not real. She’s one of the voiceover AIs that I use to create advertisements.
Up until two months ago, I used professional, human voiceover talent for all the business advertisements I made during the last few decades.
There’s still a place for professional talent if a client seeks a specific well-known voice actor. But for most cases, speed and financial considerations have been and continue to be the driving factors.
Two months ago, the technology I had been experimenting with for the last few years took a giant leap forward. Now for $22 per month, I can quickly create multiple versions of voiceovers indistinguishable from human ones.
Top voiceover talent can turn around a fully edited and mastered audio track in 24 hours, sometimes less, after I’ve sent them the final script.
Technically, a 60-second voiceover from an AI can be completed in less than 60 seconds. But AI still needs improvement, so additional steps are required to get a high-quality, edited, and mastered audio track for your next advertisement.
On average, there are five characters in an English word. For a 60-second advertisement, you’ll want 150 words at most, preferably less.
With the $22 per month subscription from ElevenLabs, I can create 130 advertisements. This equates to $0.17 per ad, assuming that each ad I make is perfect with the first rendering.
AI rendering is currently imperfect, so additional steps are needed to get a high-quality, edited, and mastered audio track for your next advertisement.
- Expect to run at least five iterations of a single advertisement through the voiceover AI. This is similar to music, where most singers will record at least five takes that are then comped together into a single recording.
- Note: “Comp” or “comped” means editing together the best parts of numerous voiceover takes into a single, cohesive audio track. For the highest level of performer, editing comes down to syllable-by-syllable. Depending on your advertising client, this may also be the case.
- To “train” the AI system, the current best-practice process is to include a paragraph of lyrics or poetry that has the same feel of how you want the AI to generate the voiceover from your script.
- Then repeat your script three times after the lyrics or poetry.
- Render and repeat five times.
- Comp your voiceover together in your audio or audio/video editor.
- Master your voiceover using equalization and compression (4:1 compression is a good starting point). The goal is to get the average loudness of your edited track as high as possible without clipping. Numerous studies have shown that with all else being equal, a louder audio track sounds better to humans.
Voiceover Time and Financials revisited
Let’s assume that your script contains 150 words which equates to 750 characters. Let’s also assume that the training lyrics or poetry includes 750 characters giving you 3,000 characters per render. With five renders, you’ll have used 15,000 characters of your available 100,000 allotment available with the $22 monthly subscription from ElevenLabs.
This equates to $3.30 per advertisement from the AI and 30 minutes to generate the audio.
Now add in the cost of your time to write the script, generate the voiceover tracks, edit, and master, and you’ll find that you can substantially reduce your client’s cost over a human voice. You’ll also reduce the time needed to deliver the voiceover to your client. And you’ll be able to produce more advertisements per day.
Here’s the beginning part of an advertisement for Big Land Cheap House. I used the Voiceover System listed in this article to create this voiceover.
While I wrote this article myself, I used Grammarly to make suggestions for improving it.
I created the images within the video by training the MidJourney Image AI with photographs I took and a prompt that I wrote.
Also, the background music was generated using the SoundRaw AI.
Note: I don’t receive any compensation for mentioning the company’s in this article.
We’ll review how to train and use a text AI, like ChatGPT, to help you write a script.