Bring back classic Overdub / Allow more training for AI Voice
Ryan Loague
I have tried two versions of Overdub: 1 with the classic training model, where I recorded 30 minutes of script; 1 with the new, faster training. The first option is FAR superior. Is there still a way to use the old training method to create a new speaker? We're interested in using Overdub, but this would be a dealbreaker.
M
Makenzie Mizell
Agree! The short script isn't enough for the AI to learn from. Also, the script doesn't include any opportunity for inflection so it creates a very flat voice clone.
Canny AI
Merged in a post:
I cant make a overdub voice?
o
olof
I cant create a new voice, I can add a new ai speaker from my voice with the short little text you provided. But ive heard it needs like 20 minutes of my voice so im trying to record it somewhere but i cant find it. Thankful for all the help i can get
Best, Olof
Canny AI
Merged in a post:
Further Train AI Speakers
J
Jeff large
I would love to further train our AI Speakers beyond the initial short prompt for better tone, style, and accuracy.
For example, I would like to be able to upload existing audio of the person talking so the AI Speaker has a larger sample size to pull from.
Thanks!
A
Andrew Brown
I create cloud certifications and I have to cover a considerable amount of content. I use my generated AI voice to generate out course materials I would not normally have time to personally record.
Here is an example of a course where I use Descript end-to-end.
I generated my voice on the classic overdub, which sound very much like me.
I heard there was an update to the model, which is providing only two minute of training, and the quality of the voice has gone down.
I imagine that this process is simpler and possibly more cost effective for Descript.
The Classic Overdub voice which I still have access to is good-enough and while it receives complaints (because people now expect better) as long as I have it I will be using Descript.
I was hoping the new model would have been better but its sad to see Descript has gone in another direction.
As a customer I want either:
- make an improved voice
- bring back classic so I can train a new model with cleaner audio
A
Austin Duncan
I mainly just use the overdub features. So, with this change - I'm out. Heading over to ElevenLabs to better train an AI voice so I can produce my podcast quicker. Descript used to be awesome.
S
Stefano
Agreed. Removing enhanced training is a huge mistake. The few sentences of the new training data are nowhere near enough to create an AI voice that sounds like the speaker.
If the Descript team doesn't have the resources to optimize AI voices, how about partnering with ElevenLabs and offer their premium service as part of the Descript solution for all paying users?
Canny AI
Merged in a post:
Bring Back Overdub
Zaiden Kristopher
Overdub has recently been removed from access to further train my A.I. voice. Is there any way we can bring that back because the basic script for authentication does not appear to be enough.
Anthony Doyle
Completely agree, the new AI voice trainer simply sucks. It's actually driving me to seek other products as the quality of the new trainer using the super short script is not enough. There is no way the AI voice can be representative of the speaker with this limited amount of training data to build the voice from.
Come on Descript, you can do better on this. you were know for having a great AI voice trainer to build custom voices in the app, that capability has now been removed.
Anthony Doyle
Completely agree, the new AI voice trainer simply sucks. It's actually driving me to seek other products as the quality of the new trainer using the super short script is as you say not enough. There is no way the AI voice can be representative of the speaker with this limited amount of training data to build the voice from.
Come on Descript, you can do better on this. you were know for having a great AI voice trainer to build custom voices in the app, that capability has now been removed.
Eric Arias
Agreed! I am running into the same problem
Load More
→