search

open-mmlab / Amphion

Amphion (/æmˈfaɪən/) is a toolkit for Audio, Music, and Speech Generation. Its purpose is to support reproducible research and help junior researchers and engineers get started in the field of audio, music, and speech generation research and development.
https://openhlt.github.io/amphion/
MIT License
4.41k stars 373 forks source link

[Help]: Is there any loss that linearly correlate to performance of TTA autoencoder? #184

Open Jiang-Stan opened 4 months ago

Jiang-Stan commented 4 months ago

Problem Overview

(Briefly and clearly describe the issue you're facing and seeking help with.) I tried to reproduce the TTA recipe, but it seems that it is hard to assess performance through any loss.

Thanks in advance for your sicerely help!

Jiang-Stan commented 4 months ago

Can you share training detail of TTA autoencoderKL? I tried to reproduce the result with the origin repo with your preprocessed mel data and training configs, but I failed to get result similar to opensource ckpt. Or is there any way determining performance of ckpts of different iter?

HeCheng0625 commented 4 months ago

Hi, the detail can be found in https://github.com/open-mmlab/Amphion/blob/main/egs/tta/autoencoderkl/exp_config_latent_4_10_78.json

fpicetti commented 3 months ago

So the rest of the parameters are inherited from config/autoencoderkl.json, right?