RookieJunChen
Control the strength of Enhancement
Can we control the strength of enhancement of Fullsubnet by config this?
Sorry, I don't quite understand your question. Generally speaking, the length of speech that needs to be enhanced is generally the same as the input. If you want to adjust it maybe you can adjust the input or output length?
Sorry, I don't quite understand your question. Generally speaking, the length of speech that needs to be enhanced is generally the same as the input. If you want to adjust it maybe you can adjust the input or output length?
Excuse me, I mean Can we control the effect of enhancement, in some cases I need there is remain a little bit of noise so that the speech is more naturalness.
Oh, I see. Our model is end-to-end, so the degree of noise reduction depends on the data seen during training. If you want to reduce the degree of speech suppression, you can try training with data with a higher SNR; or train the model with asymmetric loss [1]. [1] Quan Wang et al. “VoiceFilter-Lite: Streaming Targeted Voice Separation for On-Device Speech Recognition..” conference of the international speech communication association (2020): n. pag.