Descript: Text to Speech Feature

This isn't actually me speaking. I am an AI recreation. Descript has just made the Text to Speech feature better than ever. Descript's Text to Speech feature is no longer called Overdub, it's now called AI Speaker. It's faster to create and set up, and it works amazingly well. Here's how to do it. To create your AI speaker in your workspace, go to the AI Speakers section right here. And then you'll just hit New speaker. You'll then get this pop up asking you to name your AI speaker. Presumably this is you, but maybe it's your client or your family member or friend. But go ahead and give it a name and this is how you will assign it. Later in your project, after you've labeled your AI speaker, you'll be asked to read this script. Or if you're creating a voice for someone else, you will need them to read the script. And if someone else you're creating an AI speaker for isn't physically with you, they can record a voice memo on their own, send you that file to upload using the Other Ways to Authorize button. To get the best result with this recording, read the script at a natural pace. Don't try to read fast or slow, just your normal natural pace. However, do try to add lots of range and tone to your voice. Read it in an animated and a bit exaggerated way so that you can train the AI on your range of intonation. And lastly, make this recording in the best environment possible. If you don't have a soundproof room, a closet full of clothes does wonders for sound deadening and use the best microphone possible. If you have an external mic, use that over your computer's built in mic. Then simply hit record, read the script and when you're done, hit stop. Here's me recording mine at fast pace. That took me about 39 seconds to record, but obviously I sped it up in post so as not to bore you. And after I stop the recording, you get a box that says by submitting, I authorized the script to enable speech generation for this speaker. No one will be able to use this speaker to create speech without my consent. Then there is a submit button which I'm going to go ahead and click and we get a pop up in the bottom right telling us that it is authorizing the speaker. You can see the timer in the middle of your screen. I am not speeding this up at all so that I can show you how long this actually takes. And it's done. That was 20 seconds to create an AI version of my voice off of a 39 second sample. If that's not amazing to you, I don't know what is. Our AI voice is now trained. So let's give it a test. I've opened up a new project entered write mode using this pencil at the top here. You can toggle it on and off by clicking that. And I wrote out a sample script that I'm going to have the AI read now to apply the AI speaker to this text. There's one last step and that is to label the speaker of this text. So right here above the text, we have an add speaker button. If we click it, we can now see my name. The speaker that I trained. We can also choose a stock speaker. These are built in speakers that Descript has and their names like Don, Emily. Life is like a camera. Just focus on what's important like that so you can get that really deep narrator voice. But for the sake of this, I'm going to use the AI speaker that we just trained. That is me. So it's labeled Ross. I click it and you can see that the words are pulsing. It's like a dark blue to a light blue. It fades in and out. That's just letting us know that it's processing. And if we look down on our timeline, there's no sound waves. If I were to play this right now, there's nothing it says sound will be present once text to speech is done generating. We also have the circle loading icon. And it's just taken a moment to process it. All right, that took about 20 to 30 seconds to process. And you can see it's done because it's dark blue and it's got the sound waves down here as well. Keep in mind it uses the cloud to process all this. So the longer your file, the longer that's going to take. This was only 5 seconds, so it was very fast. But keep in mind your results may vary. Okay, let's go ahead and play this and see how it sounds. This is a test of the new AI speaker feature. How does it compare to overdub? Pretty good. Pretty good. It even put a space. It automatically did that space between the end of feature and the start of the next sentence. That's something that overdub was pretty bad at. It was just very mono spaced, mono speed. And it's got some intonation. It sounded pretty natural. A lot more natural than it did before. And the astounding thing is, once again, this took me less than a minute to record and less than 30 seconds to process. So my friends. The future is here. Welcome to the era of text to speech. Enjoy.