would it be easy to train v2 instead?
you trained mmaudio v1. unfortunately that one only outputs 5s fixed. v2 is able to create dynamic lengths. v1 has issues the authors address which is why they came out with v2 right away.
Yeah, at the time I trained it, only v1 had the scripts for retraining available. The author had mentioned that the tools for v2 weren't in the repo yet. I'll check and see if that's changed now.
any update on this? thanks for your hard work
A v2 version would be amazing!
I am also waiting for version 2, as well as information about trigger words! ❤️
I see the demand, and I will continue my work soon
The developer of MMaudio explicitly states that training for v2 is not supported https://github.com/hkchengrex/MMAudio/blob/main/docs/TRAINING.md
This was the case previously, and nothing has changed. I will look for other solutions
I plan to work with ThinkSound
I am compiling a dataset for ~40 hours. With a wide variety. It will definitely be ThinkSound
Could you share your dataset?
Did you get anywhere with ThinkSound ?
Also curious if ThinkSound worked out 👍