Scout9917
Results
2
comments of
Scout9917
> 我把wishiper抽取语音特征，以及最后口型图片贴回脸都用gpu改写了。能到50fps替换audio_processor.feature_extractor(pcm_array, Sample_rate=16000, return_tensors="pt").input_features import torch import torchaudio import torchaudio.transforms as T import numpy as np class FastWhisperFeatureExtractor: """ 完全模仿OpenAI Whisper输出为 [1, 80, T]，T torch.Tensor: B = face_tensor.shape[0] output =...
> 不太正常，我4090优化了一下可以跑120-160fps，你应该是CPU或者是CPU-GPU带宽有瓶颈 what？？？120+fps？没有搞错吗 求指点