hust-cxl

ask for help

1

Recently, I have read your paper about MTSAL, and I am trying to re-implement your algorithm. But I cannot understand your cost function exactly. Could you give me your email...

In your function feature_compute.py/return_results(), I saw a critical parameter and it was like as follow: if max_score < 0.78: continue else: ab_normal = cur_cat I was confused how you obtained...

你好，看到你的代码，觉得会对自己的工作有帮助，想问下有没有对应的paper，我想去读一读

嗨，yxlu， 我用您训练好的g_best_vb模型做推理，如果输入的是整个wav文件，会报内存不足的错误，看了之前的issue，您回复说可以在推理时把wav分段推理，但不保证效果。于是我尝试把一段长1分20多秒的wav，切分成2s的小片段，逐步推理，具体在您的inference.py上做了一些简单修改，修改的代码段如下： `def inference(a): print(h) model = MPNet(h).to(device) state_dict = load_checkpoint(a.checkpoint_file, device) model.load_state_dict(state_dict['generator']) test_indexes = os.listdir(a.input_noisy_wavs_dir) os.makedirs(a.output_dir, exist_ok=True) model.eval() with torch.no_grad(): for index in track(test_indexes): noisy_wav, _ = librosa.load(os.path.join(a.input_noisy_wavs_dir, index),...