Limits of filtering
Everyone has the hope that you can recover a poor recording using filtering. Let's take a look at what you can practically do
Waxwings are made for recording - a fantastic trilling sound as the flock stays together, these Scandinavian visitors occasionally brighten up that dreary part of the year past Christmas but before the weather starts to turn up. The only problem with the ones in this recording is that they showed no taste at all, descending on a B&Q in a peripheral industrial estate right next to the city substation which transforms the high voltage carried in on the pylons to a lower voltage. The site is next to a busy A route carrying hundreds of trucks to the port of Felixstowe.
Even with a Telinga parabolic dish, the original recording is pretty rotten, with bags of hum from the substation, and a fair amount of truck noise. I dipped on these last year, so I was definitely going to get a recording of these, and early on Sunday morning they showed up.
Waxwings, Suffolk, 5 Mar 2006, Telinga
The hum is not picked up electrically - it is the mechanical noise emitted by the substation transformers. Naturally, when making the recording I tried to place my back to the substation to reduce the hum, but the waxwings chose a tree about 30m from the substation. So this is the first thing I tried to filter out, using a 50Hz comb filter, and this was reasonably successful, because it takes out narrow frequency bands.
after a 50Hz comb filter
Better - this is the sound recording equivalent of the birder's 'record shot' - this clip reminds me of a lovely crisp March Sunday morning when waxwings came to town, and you can easily ID the sound as waxwings, but the record has obvious gross faults - the sound of all that traffic.
The stock answer to traffic noise is a high-pass filter - in layman's terms you turn the bass down. Let's give that a go.
after a 550Hz high pass filter
If you listen to the waxwings, you notice that they are reasonably high-pitched, and looking at a sonogram they don't go much below 3000Hz. The track above with the 550Hz filter is okay, but there's still an awful lot of traffic noise. Let's be more aggressive and filter at 2500Hz
after a 2550Hz high pass filter
Yuck. This just sounds ugly and unreal to me. The reason is that the traffic noise has low frequency parts, made of the rumble of heavy vehicles travelling over an uneven road plus engine noise, and a high frequency part, mainly from the tyre noise on the concrete. That extends high in frequency, and because we know very well what traffic noise sounds like, then you filter it heavily it just sounds very wrong. On the waveform you do see the waxwing trills stand out more when the low frequency rubbish is filtered out, but the ear is smarter than that as is already doing this job on the original. Filtering out huge swathes of the frequency range rarely works well, because few interference sources are confined to just one small part of the spectrum.
This recording did have one such source, the hum, and filtering just that out was reasonably successful. It did make the traffic rumble sound a tad tubey in comparison with the original, but a good compromise to my ears for getting rid of the hum. For me the second version of this recording (after a 50Hz comb filter) works - the birds are in perspective with their city environment. After that, there isn't much more I can do to improve this, other than catch the waxwings next year! I can try and filter out everything but the waxwings' trilling, but the result sounds unnatural. This demonstrates the principle -
the first law of signal processing:
Signal quality once lost can never be regained