Hi
I am doing some study on speech signal processing, specifically on background noise reduction
my advisor told me that one of the ways speech can be enhanced in a noisy environment is by taking a sample of the environmental sounds when no speech is present
Then use it to create a noise spectrum
What exactly is a noise spectrum or a noise profile and how can it help me determine whether a signal present is noise or speech? (I know it’s a vast topic so please give some general ideas to get a starting point for my work)
Start off with a read of this tutorial from the Audacity Wiki: http://wiki.audacityteam.org/wiki/Noise_Removal
This page in the manual will be useful too: http://manual.audacityteam.org/o/man/noise_removal.html
WC
The two sentence version: Select a portion of Noise Only and tell Audacity that’s the “Profile,” character or flavor of the noise. Then tell Audacity to try and subtract that character from the show.
If the primary interference is an electric motor whining at 3000Hz (fingernails on blackboard), then Audacity will try to subtract all tones at 3000Hz from the show. If you happen to be singing at 3000Hz, you lose on Sports Call. Audacity will try to remove your voice. It’s straight arithmetic, it’s not magic, and it works far less often than people would like.
The real trick is doing noise reduction when you can’t stop for a Noise Profile capture-- like on a cellphone.
“Can you stop talking for a minute while my cellphone captures the noise profile?”
What.
Then you do tricks like take six seconds of whatever is there, average it out for the profile and hope the human voice survives, which remarkably, it usually does.
Koz
Cellphones are remarkably good at this. I looked in on an Engineering meeting in Los Angeles where we were talking to one of the Systems people in Taiwan. He was calling from the Data Center which is loud enough to be a health hazard with thousands of machines screaming and throbbing at the same time. Amazingly enough, even though there was an abundance of noise, we could still understand him speaking.
The noise reduction system went as far as it could and then threw up its little hands and said, “This sucks, but it’s as good as I can do. Have a happy day, OK?”
And we did.
Koz
So does that mean any range of frequencies that have been pointed out as noise will be removed?
It’s more like a sophisticated form of “gating”.
There is a brief description of how it works here: http://wiki.audacityteam.org/wiki/Noise_Removal#How_does_it_work.3F
For a more technical description see: http://wiki.audacityteam.org/wiki/How_Noise_Removal_Works
So does that mean any range of frequencies that have been pointed out as noise will be removed?
Yes. If you have the removal part of the effect set up like that. That process usually results in terrible voices, so an additional layer was added that shuts down noise removal in the presence of voices – or significant sounds. That’s the gate. That’s about what the Smoothing adjustment does. But it interacts with the other sliders. You can tell immediately when somebody has done this to a particularly difficult news interview. The spaces between words are relatively quiet, but the words themselves are hissy or noisy and sometimes the words have little wishy tails on.
None of these tools is free. For the most part, you can’t ever rescue a bad recording. If you have led an impure life, all you will do is trade one damage for a different damage and still be no further along except you’ve wasted your time.
Koz