Xiaobin-Rong / gtcrn

The official implementation of GTCRN, an ultra-lite speech enhancement model.
MIT License
202 stars 34 forks source link

人声过度抑制问题 #43

Open hulucky1102 opened 5 days ago

hulucky1102 commented 5 days ago

作者你好,当人声中混有瞬态噪数(例如 严重的敲击声),这时模型会过度抑制人声来达到去除敲击声。请问这类情况可以怎样从数据或者损失函数得到改善么?