Researcher in need of some help with noise reduction


I’m an Canadian anthropology PhD student doing interviews in China for my thesis. I use an Olympus LS-10 to record my interviews, then later, after returning to the city, transcribe the interview to a written transcript.

Yesterday I came across an important audio interview that has really messed up sound. The batteries in my recorder, an Olympus LS-10 (quite good) must have been dying and recorded a nasty buzz that changes pitches through the recording as the batteries begin to fade.

I tried to clean out the noise, but I’m certainly not very technical, and the result was terrible. I only need to be able to hear intelligible voices. It doesn’t matter if the speech sounds unnatural or not. The main thing is I need to be able to hear what is being said.

Since I am very pressed for time- and have only very rudimentary audio-related knowledge, I would like to ask if someone here might have a look at this for me and see if they can strip out enough of the buzz but leave the voices intelligible. I don’t know if the file is savable or not, but I sure hope so.

Or perhaps someone might be able to provide me with step-by-step instructions on how to remove the noise myself using Audacity or some other application.

Thank you for your time!

see … Audacity noob needs help with filtering noise - #3 by Trebor

Select about five seconds of the show and Export as FLAC. Attach the FLAC file to the forum with the “Upload Attachment” tools under the text entry window.

The magic phrase is ‘a buzz that changes pitch’. That’s pretty serious.

That one problem kills a lot of useful tools.


Thanks for your useful suggestions.
I know that the varying pitch of the hum sound will make this a tough job.

I just don’t want to torture my poor assistant who must spend 9 hours typing a transcript of this conversation. If theres’ a way to reduce that annoying buzzing-hum, it will be great.
Will breaking the recording into different sections each of which is characterized more or less by the pitch of the noise?

I have no idea. I’ll await suggestions

It is possible to notch out the whistle in the bit you’ve attached …

You can do this by selecting the audio you want to remove the whistle from,
then post the code below in something called “Nyquist prompt” which is in the “Effects” drop down menu

(notch2 (notch2 (notch2 (notch2 s 1820 4) 3680 5) 7276 5) 100 5)

select the track, copy and paste the 'notch' code into the 'Nyquist Prompt' then press OK (I've coloured green).png
The bad news is if the whistle changes pitch in other parts of the recording this code won’t remove it.

Thank you very much!
Ok … the difference is very nice indeed. Is there a way to remove the ‘white noise’ sound left after using the notch filter?

But is there some way to analyze the entire recording and find the various frequencies and remove them?
I looked here and will try this tomorrow and see if I can’t clean it up.

Any advice as I proceed with this?

You have the worst of all possible noise problems. A buzz or whine that moves over the course of the show, plus white noise. We will not be able to remove the hiss from the actual spoken words, but we can remove the hiss from between the words simulating a semi-quiet performance.

We will never get back to a good theatrical show.

The short version of noise reduction is select some performance with noise only and apply that to the “profile.” Then select some of the actual show and apply the filter. It will try to suck noise similar to the profile out of the show.

The amount of suck is the suppression of the noise and the other two settings have to do with protected voices – to keep people from sounding like Space Aliens. Delay is the time before and after each word and Smoothing is the vocal characteristics. 18dB reduction sounds approximately like half-way.

The early noise reduction didn’t have all these fancy-pants settings and it was almost universally terrible.


You can’t magically inspect the whole show and derive all the special settings. The tools are good, but they’re not sophisticated enough to do the noise reduction in groups or segments.

We count ourselves lucky if we can deal with one single noise that’s not too bad.

You’ll have to do each discrete pitch change separately. New notches, new profile, new reduction.

How close is retirement?


Because the noise is similar to “white noise” : which equally present across the entire sound spectrum, then noise reduction techniques like Audacity’s won’t help a lot : they are only effective when the noise only affects parts of the spectrum and reduces those parts.

One final tweak to reduce hiss in a speech recording would be to cut back as much of the high end, (above about 6000Hz), that you can get away with before it affects intelligibility. The overall reduction in hiss by doing this will noticeable but minor.