unnatural hum

I ran into something new tonight. Maybe someone can explain it and even offer a solution. The original source of this was tape but I didn't do the transfer to computer so I don't know where the problem originated. The recording is a couple people speaking, no music.

There is noticeable hum. The frequence analysis graph shows peaks at 60Hz, 180Hz, and 300Hz. I've used CE's Notch Filter to remove that sort of things from both LP and tape sources a fair number of times. Results have generally been quite satisfactory to me. On this recording it produces a distinct and unpleasant artifact. Speech sounds like it is taking place inside a barrel.

Some experimenting showed me that if I used the filter only for 60Hz, things seem ok; at least I don't get that unpleasant hollow echo effect. While filtering out the 60Hz has a significant beneficial effect according to the Frequency Analysis graph, Analyze/Statistics, and the general appearance in Spectral View, it does not really reduce the hum for my hearing. It actually makes is sound a bit more harsh, or in some way even less pleasant than with the 60Hz left in.

I've moved forward by applying the notch filter on 60Hz only, then managed a noticeable reduction in the remainder through NR. It's not perfect but it is better than the starting condition.

Have I provided enough information to suggest WHY the notch filter is producing this unpleasant, and uncharacteristic, result? Any suggestions for better hum reduction by doing ...

Any suggestions for better hum reduction by doing ...

Try taking a sample of the hum and using it as a noise reduction profile. This often works quite well when there are significant signal components near any of the hum harmonics - which is probably what's causing this.

This a strange and difficult problem. As I mentioned in the first post, I got better (but not perfect) results with NR than with filtering for the 180Hz and 300Hz. I had a vague idea floating around in my head about doing some more specific NR for 180Hz, but only a fuzzy picture about somehow chopping out a frequency specific region to make the noise profile. It wasn't until after reading SteveG's post that a (possibly) reasonable way to do that came to mind.

I used the notch filter set to ‘super narrow' and adjusted the attenuation of each of the three frequencies to get a pretty flat looking Frequency Analysis Graph from the filtered noise sample. Then I mix-pasted-inverted the filtered noise sample back into the unfiltered noise sample to (supposedly) get only those three hum peaks. The graph of the final result looked a bit strange to me, but I made a noise profile, and NR did a very good job on that recording.

There are a number of ‘reels' in this recording, however. They seem much the same to me, but the process did not work very well on the next cassette. Two separate problems showed up that were not apparent in the first file.

The first of these is that NR results seemed quite uneven, depending on the immediate content. The pauses between speech, at least those over a few hundred milliseconds long, seemed very clean, but the hum was still definitely audible in the midst of the speech. I have no idea why this is so on the second file but not on the first. Here the filters seem to give more consistent results than NR over the entire recording.

The second problem is that the hum apparently isn't very even over the length of the second cassette. In some parts, after NR, there is still significant hum audible even between the speech. Trying to chase the levels from moment to moment is just far too much work for a whole batch of 42 minute recordings, at least at my rate of pay (zero).

Between my initial post and my reading of Steve's reply I continued to experiment. I got fairly reasonable results by using the notch filter just for 60Hz, then switching to the parametric equalizer. It has a 60Hz hum preset into which I substituted 180 and 300 Hz.

This gave the same poor results as the notch filter, but improved significantly when I reduced the slope, widening the cut. This seems consistent with Steve's comment about adjacent frequencies being responsible for the unpleasant hollow echo effect. Anyway, with a combination of filters plus NR, I managed to make the recording fairly listenable rather than headache inducing.

This is great info AndyH, Thank You for posting it!  

I've described exactly how this happens before, but I can't presently find the page with it on. It's all to do with signals in the filter skirts - these have a very rapid rate of phase change if the filter slope is steep, which it will be on the notch filters. One or both of two things can happen; you end up with a very 'hollow' sound around this point, because of the phase changes, and/or you can excite these frequencies into 'ringing' - which at 60Hz will sound like a slightly different pitch hum, of course. And if the signal comes and goes, so will this hum - which is essentially modulated by the signal.

Solution no. 2 works with speech, and is a bit of a fiddle, although it can work very well if you are lucky, and are fortunate with the original signals.  You need from the recording a sample of the LF room ambience with no talking, and no hum. On most recordings, this is quite easy to arrange, because getting the hum cleanly out of silence doesn't introduce any more artefacts. Then, you do a gentle bass roll-off on the speech content, and get rid of all the LF that's causing a problem. Normally this would sound as though something was missing (because it was!), but mixing the silent ambience, carefully looped, back in with the filtered signal can often remove that 'incomplete' feeling completely. But it does need to be a pretty clean sample to work well, with nothing that repeats in it.

Then, you do a gentle bass roll-off on the speech content, and get rid of all the LF that's causing a problem.

If this case the problem was with the 180Hz and 300Hz. The 60Hz was easy to remove with the notch filter and, if done by itself, caused no problems. Do the 180 and 300Hz components fall into that LF part one gets rid of, or does this solution only work for 50/60Hz? It seems like there is quite a bit more than just hum and room ambience by the time you get to those frequencies.

Then, you do a gentle bass roll-off on the speech content, and get rid of all the LF that's causing a problem.

If this case the problem was with the 180Hz and 300Hz. The 60Hz was easy to remove with the notch filter and, if done by itself, caused no problems. Do the 180 and 300Hz components fall into that LF part one gets rid of, or does this solution only work for 50/60Hz? It seems like there is quite a bit more than just hum and room ambience by the time you get to those frequencies.

It does tend to work better with female voices than male ones, admittedly - and as I said, you do have to be a bit lucky with the original material anyway. But sometimes, even just rolling the bass off gently and replacing it with a hum-free masking signal can help - it all depends on how clean you want the result to be. Sometimes doing a gentle hum removal that isn't complete, but doesn't have artefacts, and adding a hum-free mask can help too. Experimentation is invariably the way forward here.