Just getting started on audio--removing a squeal with Audacity

[solarblast]

I have an avi file that has a segment with a high squeal imposed on someone talking. I used VDub to remove the segment to operate on it with Audacity under Win7. I've attached a jpg of the wave, and have played a little with the generators. Just by looking at the wave, I'm not sure what would help. I would think knocking out high frequencies, and using noise removal. That didn't quite do it. Any suggestions? What's the difference in the wave between the light and dark blue of the wave?

[poisondeathray]

you will probably get better suggestions if you posted an audio sample

[solarblast]

Forgot file.

[solarblast]

Here comes an example. Close to 15 meg. I'm talking with a fellow inside a domed building and plenty of electronics are around.

[jagabo]

Use the little pulldown near the top left of the waveform view and change it to Spectrum. That will show you the frequency distribution of the noise. Then use a notch filter (Equalization...) to eliminate those frequencies. Since the frequency of the noise changes with time you'll have to work in sections. You can also try a low pass filter at ~3 KHz (most of the noise is over 4 KHz) but that will leave the audio sounding dull.

[solarblast]

That's quite a graph. Red, blue and white. What exactly that means I'm not sure. Perhaps the colors denote volume. Blue = high? To bad there's no manual. Online help has no search ability. Ah, I used 1.2 a bit under XP and found a manual pdf. Not a lot of description on how to use the spectrum. Maybe it's in another doc? Perhaps there's a tutorial on audio methods somewhere?

Freq along the y-axis. Time on x. I just went straight to equalization with not much of any improvement. I've tried low pass at 1k, 2k and now 4k. Squeal is still there.

[jagabo]

Lowpass at 3K with a 48 dB/octave falloff (the default is only 6 dB/octave).

The spectrum view is time on the horizontal axis, frequency on the vertical axis, and color represents volume, blue for the lowest volume, through reds, then whites for the loudest frequencies. In the example below I've used the Equalization filter to apply two notches to a short segment of the sound, one at about 4 KHz, the other about 6KHz. You can see the white peaks are gone. And when you listen to the segment the squeal is much attenuated.



You can also mark a small section and use Analyze -> Plot Spectrum to see where the peaks are.

[solarblast]

Thanks very much. I'll try to duplicate your work on a broader segment. There's about 5 min. BTW, did you understand the Spanish speaker that I was talking to. My voice was much closer to the microphone. He had a mild Spanish accent, so his voice might have "improved" with your change.

What allows me to mark the peaks?

I just had an interesting thing happen. My HP suddenly died. I called their tech support and the OS had corrected the fault. OS or something failed. However, Audacity had failed too. It was showing an error,but had nothing to do with the PC failure. The tech knew of the program and uses it. I asked him if he knew where one could learn more about how to use it. YouTube. A search for "audio editing tutorial". A good coincidence.

[SLK001]

This was done with Adobe Audition's noise reduction function. This is just a "quick and dirty" test.

[jagabo]

A short segment, before and after:



Remember you have to adjust the notches as the frequency of the noise changes.

[solarblast]

This is all well and good, but I don't have the slightest idea how to make a notch filter. I puttered around with equalization and found no way to "draw" a curve. All that moved was the horiz line.

[jagabo]

Each time you left click on the blue line it makes a new node. You can then drag the node around by left clicking on it.

By the way, enabling the Linear Frequency Scale option will make it easier to draw the curves in the high frequencies.

[solarblast]

I didn't notice your notch.wav file, but just tried it. I can clearly understand the person I'm interviewing. That's very good. The lead in was pretty squeaky; however, you may not have isolated that portion of the clip.

I can at least operate the curve, but there is much more to explain about the operation of the eq. I noticed there is an audicity forum and posted there to see if there is a full operational description.

[jagabo]

That clip included both before and after processing of the same ~17 seconds.