Open Andong-Li-speech opened 4 years ago
Hi, thanks for your release of the code w.r.t. component loss, it is an interesting work that separates noise suppression and speech reservation respectively. I have a question about the range of output mask. Actually, in the recently published paper "Using separate losses for speech and noise in mask-based speech enhancement" (ICASSP2020), the CNN topology estimates only the real-valued mask M_{l}^{k} \in [0, 1] to enhance ...., however, in the released code, I find no sigmoid function is utilized to constrain the range of mask to [0, 1]. So I am wondering whether the sigmoid function is used as the output activation function in this study.
Sorry for replying late. Yes, the activation function for the output layer is "Sigmoid". I correct it in the code. Thank you for your comment.
Hi, thanks for your release of the code w.r.t. component loss, it is an interesting work that separates noise suppression and speech reservation respectively. I have a question about the range of output mask. Actually, in the recently published paper "Using separate losses for speech and noise in mask-based speech enhancement" (ICASSP2020), the CNN topology estimates only the real-valued mask M_{l}^{k} \in [0, 1] to enhance ...., however, in the released code, I find no sigmoid function is utilized to constrain the range of mask to [0, 1]. So I am wondering whether the sigmoid function is used as the output activation function in this study.