Rares - 2014-06-24

Hello,
I would like to request help regarding the following problem. On some systems there is a possibility to set the microphone gain of the input audio samples. I've read that recognition accuracy can be affected by this. What I couldn't accurately find was the methodology to compute the gain in order to minimize the degradation of performance. I am interested in some general approaches. From what I have read there are many types of gains that could be applied (at the feature level, at the spectral spectrum) but I am interested in one that operate in the time domain and applies the gain to audio samples. I have read that the gain is computed based on a desired "target level" or "desired amplitude", also "gain curve" has come up a few times. What are the steps to find those values? I realize my knowledge about this is incomplete but I want to ask for some directions on this topic like a document that treats this subject top to bottom. If it can be explained in simple steps or if some examples are available a link would be much appreciated. The only thing that I found was in sphinxbase in ad_oss.c . However I couldn't find an explanation as to how that input gain value was determined. If more details are required please ask. Thank you in advance.