Haohe (Leo) Liu / 刘濠赫

@snufas Apologize for the inconvinence. I tried but cannot reproduce this issue on my windows laptop. Could you provide more context if you like? I guess the main issue is...

Sorry for the late update. I'm trying to look into it recently. Will inform this thread once it's fixed.

Hi Ivan, thanks for your interest and suggestions. I'll try to add one example when I have time. I'll let you know when it's ready. And right, the current restore_inmem...

Great! Looking forward to it!

@pandysong Sorry for the late reply. The model is non-autoregressive and mainly CNN-based. So if running on a GPU with some optimizations, it could be functioning in real-time.

@pandysong Porting to an SoC, that's cool! Actually, the pre-trained model is available [here](https://github.com/haoheliu/voicefixer). You can play with it if you are interested.

Hi @mosem Thanks for pointing out. This part of configuration is a little bit redundancy, which means it's not actually used by code. So changing this part will not affect...

Hi @LqNoob, I'm not sure if you still need the answer or not. Many apologize for the late reply. These are good questions. 1. The implementation of TFGAN is confidential...

Modeling 16k data is usually easier than 44.1k data because of the smaller dimension. Personally, I think the 16k model would be easier to train and get a better result.

感谢关注，训练数据合成相关的逻辑已经包含在这个仓库中了。 大部分distortion是在这一行加入的： https://github.com/haoheliu/voicefixer_main/blob/main/dataloaders/dataloader/FixLengthAugRandomDataLoader.py#L104 Lowpass是在这里加入的： https://github.com/haoheliu/voicefixer_main/blob/main/dataloaders/data_module.py#L18 噪声是在这里加入的： https://github.com/haoheliu/voicefixer_main/blob/main/models/gsr_voicefixer.py#L217