# How do I improve the SNR of my signal given a noise profile (additive noise)?

From an iPhone microphone, I have a background recording of 5s(fs = 44.1 Khz), and immediately after that, I have a voice recording of a person saying 'aaa...' for 10s(fs = 44.1 Khz). How can I use the background recording (How do I characterize the noise from it?) to improve the SNR in the voice recording? The noise can be colored too, also I am modeling it as a additive noise.

Can someone please point me to any work related / in a similar direction (theory, implementation etc.,)

• Well, you say "improve SNR", so you have a definition of what your S and your N are – simply fill these in. We don't know which SNR definition you use, so it's pretty much impossible to help you. – Marcus Müller Oct 24 '17 at 20:50
• Agreed. I feel that my question is a bit misleading. I want to remove the signal with properties (from the 5s recording) from the 10s recording signal. i,e, the N is modeled from the 5s recording. y = S+N, is the 10s recording. I do not have a clear analytical model for S (depends on person), but have a frequency range for it(the formant ranges for 'a' phoneme). Is it possible to extract S ? If not what constructs is my problem missing ? – itismeghasyam Oct 24 '17 at 20:57
• well, that is what I'd call a signal model, and I'd think if you tried to actually write down how the formant range for that phoneme is represented in literature, you'd get an excellent approach for an estimator :) – Marcus Müller Oct 24 '17 at 21:00
• IEEE literature is full of speech / voice enhancement papers...? – Fat32 Oct 24 '17 at 21:07
• Fat32, I am not that familiar with those, could you please give me a starting point in that literature. – itismeghasyam Oct 24 '17 at 21:21