UPDATE: On the very day I posted this, Descript announced a workaround so you don't have to read through the whole 30-minute script, but can instead use existing audio of your voice 😂 In any case, I'm going to work on improving the audio quality with more AI training and I'll post a follow-up video with the results. You can try Descript Overdub yourself at www.descript.com/overdub
I know you probably will not read this, but I have noticed that commas create big pauses like periods. I believe that if you skip the commas in places where you are not listing something or a big pause is reasonable, you should just skip the comma. For example the last sentence: "(...) and then you solve for the problems of the day, over and over and over again ..."
If I had it today I would use it today. I have good-voice days and bad, and after training the AI, that'll be no prob. Thanks for a great video! Fascinated to the very end.
I was pretty impressed! It's really only that close if the environments are similar between the trained voice and the audio you're patching, but overall I was really impressed, especially with the intonation!
Great video. A great demo of how close voice cloning has come, and also how far it has to go. If I had to listen to an hour of that AI voice reading a script, there'd be blood on the carpet. It's missing many of the characteristics that make people interesting to listen to. Granted, there are also some extremely bad speakers, and in their case, it might actually improve their speeches.
Much like the content of the demo podcast, the solution is to start small and build organically. Solving the problems along the way. I've seen some mainstream voice recognition software fail due to lack of training by the end user. Did it really fail? I'll never know. People don't want to put forth the effort required to train these systems. Remember, you've had 30 plus years, 24 hours per day to train your ability to speak in thousands of different scenarios. Then asked the AI to replicate it with under an hour of content. If you were to feed it the audio contents of each of your videos, I'm confident it would be able to replicate you well enough to fool your family. Minor adjustments in pitch or cadence could create entirely new 'people' to provide content - eliminating the need for hired talent. Pretty soon, 1000 people will agree that AI voices are best - because one guy created 1000 voices to say so. This will be interesting indeed.
Very helpful! I'm exploring Overdub and Descript at the moment, and waiting for it to digest an audiobook I recorded as my voice sample! Wish me luck. :)
The long text-to-speech segment was quite good, but also obviously AI and wouldn't fool anyone. I think it would be quite annoying to listen to, to the point where a lot of people would find something else to do unless the subject matter was right up their alley.
So I'm a total descript noob and a Mac noob-- I'll warn you now. I really want to explore the overdub feature, but I've walked through all of the steps even uninstalling and reinstalling And when I can "overdub" and "assign the name" I called my od voice, playback is completely silent. I've tried long readings that I had made training other voice rec apps, then I tried there sample reads.. thinking it might train better or something? I still get the same result - it will play the computerized voices, And it will play back me reading me the original audio, but the overdub with a different set of words - silence. The graphic waveform reflects no noise too (But of course I can listen to the files if I just click on the m4a), So I've ruled out everything I know to try. Anyone experience this or got any tips what to do? I keep running out of transcribe time, and quite obviously I'm not handing a company some of my money before it proves that it can do The Thing. THANKS in advance. Jeffy g 🥴
Well it sounds better than those voices from Microsoft. So I would use it Get a random Jamaican Guy pay him for a session get him to read for 3 hours and I have a voice perfect for cartoons, short snippits etc. Then get a heavy italian new York accent, a hillbilly etc and you can make great content like I'm the host and these are my imagenary friends....
lol great idea. Didn't think about it that way, you can essentially have 2-4 co-hosts on your podcasts and you'd just write their scripts for them and go have a couple beers while they record an hour-long podcast for you lol - You could even make them argue back and forth with each other, then break out into a fist fight, ok, ok, maybe I'm overthinking this a little lol
I'm a little speechless. And frightened lol... You're a funny guy too, loved all of your quirky little jokes. Great video and the long form test you performed didn't come out too bad. I could tell at instances it was a robot voice but at some points I was completely fooled into thinking I was listening to your real voice. Pretty awesome and like I said... frightening stuff. Thanks for the great video. Best overdub review I've seen, the first 3 I viewed only test one sentence, 5 -10 word examples, how dumb right?
Theoretically you could have Google Translate translate the transcription, then have the system read it in that language/accent, but the results are going to be pretty sketchy.
Nice review you have here ! Anyway I am curious.. If I upload an existing audio of my voice (speaking French), can Descript create an English audio from English texts? Or if i upload existing audio of my voice (speaking non-fluent English), can Desript create a perfect English slang audio from an English text?
I’m a legally blind (super newbie RUclips r and I can’t read scripts well while I’m also trying to monitor my art videos, so I was thinking about a service like this
Hi James, I downloaded the package but when i open it, it says something like my PC does not have internet connection (but clearly it does), please check and retry again. So basically i couldn't install it. Do you have any idea how to fix it? Thank you so much, love you video :)
Hi, Susan. The exactly same thing happened to me. But I somehow found a workaround - when you try to install it, you will see that dialog box telling you there's no internet connection. Just keep pressing cancel until the very last time it appears, telling you it didn't succeed in downloading the installation file from a web address, and you will see that address. You need to make a note of that address, so don't cancel out of it. The only thing you need to do now to get the installation package is to type that address manually in your browser's address bar. So either make a screenshot of that final dialog box, or simply open your browser, type it out in the address bar, and hit Enter. It will automatically start the download. Once you download it, you will need a program called 7-Zip to unzip it, since it's compressed. This program is free, you can find it anywhere. Once you download it and unzip the folder you got from that address, just look for the Descript.exe file (it has a blue icon). Once you click it, it will open the Descript desktop app and connect automatically to your Descript account, the one you had to make in order to get that installation package that didn't work from their site in the first place. And I think that should be all, it should work now :)
I guess it didn’t take off.. it’s not the “greatest tech since digital video”.. 2. It takes longer to write and you can’t express you’re self with writing .. the passion is gone in ai voice cloning.
Congrats on giving away your voice for free. People that make a living off voice over need to read the fine print. 8.2 License to User Content. We claim no ownership rights in your User Content. You hereby grant to us a nonexclusive, royalty-free, sublicensable, worldwide license to access, reproduce, distribute, process, publish, display, perform, adapt, modify, analyze, and otherwise use the User Content to provide, maintain, and improve Descript and the Descript technology, without compensation to you, provided that our use of any Projects you create is subject to the usage limitations and confidentiality obligations set forth in Section 9 below.
@Resemble AI: I am looking for the second feature -- the ability to take a script and create realistic audio out. You might want to have a hard look at your packages and pricing if you want to attract smaller clients, like podcasters and those creating instructional resources.
1) Oooof... poor Arcosanti. You were pretty brutal (though deservedly so.) 2) That extended passage was surprisingly tolerable. Not natural or perfect, but impressive. I'm looking forward to hearing that system with more training.
I really want master-planned communities to work, but they just never seem to! 😅 And I agree, it wasn't *quite* there yet in my opinion, but I'm hopeful that additional training will help smooth out the rough edges. It'll never be perfect, but if I can get it to, say, 80% of perfect, that's still pretty dang good!
UPDATE: On the very day I posted this, Descript announced a workaround so you don't have to read through the whole 30-minute script, but can instead use existing audio of your voice 😂 In any case, I'm going to work on improving the audio quality with more AI training and I'll post a follow-up video with the results.
You can try Descript Overdub yourself at www.descript.com/overdub
That workaround mentioned above seems to be very interesting. May this work in other languages, too (German, for example)?
I know you probably will not read this, but I have noticed that commas create big pauses like periods. I believe that if you skip the commas in places where you are not listing something or a big pause is reasonable, you should just skip the comma.
For example the last sentence: "(...) and then you solve for the problems of the day, over and over and over again ..."
If I had it today I would use it today. I have good-voice days and bad, and after training the AI, that'll be no prob.
Thanks for a great video! Fascinated to the very end.
U can use it today, buy a standard plan and you are done bro
Bro, this is a cutting-edge review. I want to be like you when I grow up. great content. keep it up.
Thank you, Sam! I appreciate your comment!
Your opening lines are my life! haha That's why this app is so interesting to me. Thanks for your review.
This is an amazing piece of software! That part with the Resemble AI voice over tool had me dying lmao!!
That is remarkable when used to patch up words. I seriously wouldn't know.
I was pretty impressed! It's really only that close if the environments are similar between the trained voice and the audio you're patching, but overall I was really impressed, especially with the intonation!
Great video. A great demo of how close voice cloning has come, and also how far it has to go. If I had to listen to an hour of that AI voice reading a script, there'd be blood on the carpet. It's missing many of the characteristics that make people interesting to listen to. Granted, there are also some extremely bad speakers, and in their case, it might actually improve their speeches.
Haha this was awesome. Overall the ending script was pretty impressive. Gotta check this out more!
Much like the content of the demo podcast, the solution is to start small and build organically. Solving the problems along the way. I've seen some mainstream voice recognition software fail due to lack of training by the end user. Did it really fail? I'll never know. People don't want to put forth the effort required to train these systems. Remember, you've had 30 plus years, 24 hours per day to train your ability to speak in thousands of different scenarios. Then asked the AI to replicate it with under an hour of content. If you were to feed it the audio contents of each of your videos, I'm confident it would be able to replicate you well enough to fool your family. Minor adjustments in pitch or cadence could create entirely new 'people' to provide content - eliminating the need for hired talent. Pretty soon, 1000 people will agree that AI voices are best - because one guy created 1000 voices to say so. This will be interesting indeed.
I feel like if it was a tad slower, and that strange robotic background noise wouldnt be there, the podcast would be extremely realistic o.o
Very helpful! I'm exploring Overdub and Descript at the moment, and waiting for it to digest an audiobook I recorded as my voice sample! Wish me luck. :)
The long text-to-speech segment was quite good, but also obviously AI and wouldn't fool anyone.
I think it would be quite annoying to listen to, to the point where a lot of people would find something else to do unless the subject matter was right up their alley.
The AI did a good job.
So I'm a total descript noob and a Mac noob-- I'll warn you now. I really want to explore the overdub feature, but I've walked through all of the steps even uninstalling and reinstalling And when I can "overdub" and "assign the name" I called my od voice, playback is completely silent. I've tried long readings that I had made training other voice rec apps, then I tried there sample reads.. thinking it might train better or something? I still get the same result - it will play the computerized voices, And it will play back me reading me the original audio, but the overdub with a different set of words - silence. The graphic waveform reflects no noise too (But of course I can listen to the files if I just click on the m4a), So I've ruled out everything I know to try. Anyone experience this or got any tips what to do? I keep running out of transcribe time, and quite obviously I'm not handing a company some of my money before it proves that it can do The Thing. THANKS in advance. Jeffy g 🥴
Some very good sections but here and there it sounded like you were gargling, as if frames were skipped or fast-forwarded.
Yeah, I agree, there's a certain amount of scratchiness to it. I'm hoping that'll go away with more training!
Well it sounds better than those voices from Microsoft. So I would use it Get a random Jamaican Guy pay him for a session get him to read for 3 hours and I have a voice perfect for cartoons, short snippits etc. Then get a heavy italian new York accent, a hillbilly etc and you can make great content like I'm the host and these are my imagenary friends....
lol great idea. Didn't think about it that way, you can essentially have 2-4 co-hosts on your podcasts and you'd just write their scripts for them and go have a couple beers while they record an hour-long podcast for you lol - You could even make them argue back and forth with each other, then break out into a fist fight, ok, ok, maybe I'm overthinking this a little lol
I need one that sounds like hell of a fun!
This is going to be so useful in small efficient ways. Great video as always
Thanks, Timothy! 🙌
Dude... this is wild. And creepy as hell. Every content creator's dream - and worst identity hacking nightmare!
They dub your voice , and you're confessing to major crimes , to a radio show . people recognize your voice ? Yeah that sounds like a plan .
hank you very much for this.
Great video and example. Thanks for putting this together!
I'm a little speechless. And frightened lol... You're a funny guy too, loved all of your quirky little jokes. Great video and the long form test you performed didn't come out too bad. I could tell at instances it was a robot voice but at some points I was completely fooled into thinking I was listening to your real voice. Pretty awesome and like I said... frightening stuff. Thanks for the great video. Best overdub review I've seen, the first 3 I viewed only test one sentence, 5 -10 word examples, how dumb right?
Nice video, James. I like how you ended it.
I need a program that can take my videos and translate them in spanish over my english, wish it could do that, but I'm guessing not right?
Theoretically you could have Google Translate translate the transcription, then have the system read it in that language/accent, but the results are going to be pretty sketchy.
Excellent explanation James . Coming from your newest subscriber
Can I use an overdub voice for my youtube videos? will it monetize?
2:51 sounds like one of my internal dialogs; coherent at first but I quickly get bored by my own thoughts so I drift off :)
Surprisingly good!
Is it possible to clone my voice in Spanish?
Nice review you have here ! Anyway I am curious..
If I upload an existing audio of my voice (speaking French), can Descript create an English audio from English texts?
Or if i upload existing audio of my voice (speaking non-fluent English), can Desript create a perfect English slang audio from an English text?
Are you allowed to import the file and use a better program to record your voice on? This might be why it sounds off?
I’m a legally blind (super newbie RUclips r and I can’t read scripts well while I’m also trying to monitor my art videos, so I was thinking about a service like this
Super helpful! Thank you!
Cool James! did you do the extended overdub voice training review?
Descript Speaking 6:11 longer segment 7:47
We only speak a few words, how can the app speak the rest of the words, where did the original voice come from so amazing
Is it possible to use a voice from a game if i have enough of the audio joined together to create a clone of the voice from the game?
Aloha- but do you get demonetized by using this?
I'm new to all this , will descript run on laptop? Or do I need a pc
I wonder if you use morph cut in your videos hehehe.
can u send your voice without reading their script
morph cut by adobe premiere is more scary
Hi James, I downloaded the package but when i open it, it says something like my PC does not have internet connection (but clearly it does), please check and retry again. So basically i couldn't install it. Do you have any idea how to fix it? Thank you so much, love you video :)
Hey, Susan, I don't know that one. Sorry I can't be more help!
Hi, Susan. The exactly same thing happened to me. But I somehow found a workaround - when you try to install it, you will see that dialog box telling you there's no internet connection. Just keep pressing cancel until the very last time it appears, telling you it didn't succeed in downloading the installation file from a web address, and you will see that address. You need to make a note of that address, so don't cancel out of it.
The only thing you need to do now to get the installation package is to type that address manually in your browser's address bar. So either make a screenshot of that final dialog box, or simply open your browser, type it out in the address bar, and hit Enter. It will automatically start the download. Once you download it, you will need a program called 7-Zip to unzip it, since it's compressed. This program is free, you can find it anywhere. Once you download it and unzip the folder you got from that address, just look for the Descript.exe file (it has a blue icon). Once you click it, it will open the Descript desktop app and connect automatically to your Descript account, the one you had to make in order to get that installation package that didn't work from their site in the first place.
And I think that should be all, it should work now :)
where is the script on the app to read please?
That’s funny funny very funny hilarious
I guess it didn’t take off.. it’s not the “greatest tech since digital video”..
2. It takes longer to write and you can’t express you’re self with writing .. the passion is gone in ai voice cloning.
This software useful for Urdu or Punjabi language....?
bro plz increase the the sound of the videos plz
Nice content
I wish they have app on android
How many voices can you make? Thanks for the review btw this was amazing
I think it depends on your plan
Congrats on giving away your voice for free. People that make a living off voice over need to read the fine print.
8.2 License to User Content. We claim no ownership rights in your User Content. You hereby grant to us a nonexclusive, royalty-free, sublicensable, worldwide license to access, reproduce, distribute, process, publish, display, perform, adapt, modify, analyze, and otherwise use the User Content to provide, maintain, and improve Descript and the Descript technology, without compensation to you, provided that our use of any Projects you create is subject to the usage limitations and confidentiality obligations set forth in Section 9 below.
Hey James, ready for v2 with Resemble? (check your email 😉)
@Resemble AI: I am looking for the second feature -- the ability to take a script and create realistic audio out. You might want to have a hard look at your packages and pricing if you want to attract smaller clients, like podcasters and those creating instructional resources.
don't work in italian!!! :-((((
Not good enough to even consider - still a long way to go...
I dont get the hype around this, you only get crappy text-to-speech and ugly jump cuts for 30 bucks per month?
1) Oooof... poor Arcosanti. You were pretty brutal (though deservedly so.) 2) That extended passage was surprisingly tolerable. Not natural or perfect, but impressive. I'm looking forward to hearing that system with more training.
I really want master-planned communities to work, but they just never seem to! 😅
And I agree, it wasn't *quite* there yet in my opinion, but I'm hopeful that additional training will help smooth out the rough edges. It'll never be perfect, but if I can get it to, say, 80% of perfect, that's still pretty dang good!