More details about the clip separation for training

YuanxunLu / LiveSpeechPortraits

Live Speech Portraits: Real-Time Photorealistic Talking-Head Animation (SIGGRAPH Asia 2021)

MIT License

1.16k stars 198 forks source link

More details about the clip separation for training #51

Open DreamtaleCore opened 2 years ago

DreamtaleCore commented 2 years ago

Hi, Recently I'm trying to repeat the training part of audio2feature, could u please share more details about the dataset construction, e.g., the duration of each clip, and the overlap of two clips?

Thanks and looking for your reply😊.

Yunfei

YuanxunLu commented 2 years ago

There're no overlaps between the two clips. I just use one clip for training. Check issue #27.

DreamtaleCore commented 2 years ago

Thanks! This helps me a lot.