Home
I use the following equipment to digitize my records and get them on the web:
The Owl 1 pre-amp is a special piece of equipment for playback of 78s. It has equalization , rumble filtering, and hiss filtering. The equalization is important because, prior to the development of the RIAA standard, record manufacturers all used different equalization curves, and prior to electrical recording (which came in around 1925), they used no equalization at all. The Owl 1 allows one to set the equalization differently for each record. I could do the equalization using the Sound Forge software, but it's easier to use the pre-amp, and doing it this way might even produce better results.
I use a home-made switch box to switch the inputs and outputs of my sound card, Owl 1 unit, and Harmon/Kardon pre-amp. In one position, the switch routes the output of the Owl 1 to the input of the sound card, and the output of the sound card to the input of my pre-amp. In the other position, the switch routes the output of the Owl 1 to the input of the pre-amp, and the output of the pre-amp to the input of the sound card. This allows me to use the pre-amp as a volume control for the input to the sound card when I am recording (the Owl 1 has no volume control), and to listen through the pre-amp when I am playing sounds through the sound card.
I do not use the rumble and hiss filters when I digitize records. The reason is that the digital noise reduction in Sound Forge has less effect on the music. I do use the rumble and hiss filters when playing records for listening rather than for recording.
The truncated elliptical styli help with the sound of worn records. Often, when records are damaged through the mis-use of steel needles, the damage is mostly in the bottoms of the grooves. Truncated styli don't reach all the way to the bottoms of the grooves, so they can avoid the damaged parts. The different sizes of styli are important because the groove sizes are not the same on all records (later records tend to have narrower grooves), and because you can sometimes avoid the damaged portions of grooves by switching styli (the different sizes ride in different parts of the grooves).
There are two types of noise reduction in the Sound Forge plug-in: pop and tick elimination, and broad-band noise (hiss) reduction. One can't use either indiscriminately without hurting the sound. The pop and tick eliminator works by looking for spikes in the waveform, and either interpolating them away or replacing them with other waveforms. If you use this feature indiscriminately, the software can interpret signal as noise. When this happens, you get chattering or burping noises. I get around this by applying the tick eliminator to only a few seconds of the recording at a time, and listening afterwards to tell whether any problems were introduced. If there are problems, I undo the operation and tell the software to stop at each spike and let me decide whether to eliminate it - I can usually tell by visual inspection whether the spike is signal or noise. This can be tedious, but it is necessary to get good results.
Hiss reduction in Sound Forge works by giving the software a section of the recording where there is all noise and no signal (I use the lead-in groove for this). The software analyzes the noise and forms a frequency distribution curve. You then tell it to draw an envelope some number of decibels over the noise distribution. When you apply noise reduction, it analyzes the frequency distribution of the recording. Any frequency whose amplitude falls below the envelope is reduced in amplitude by some number of decibels that you get to specify. If you use this feature indiscriminately, you can get some pretty nasty ringing or whistling noises. I think what happens is that some noise frequencies will "poke through" the envelope, and these isolated frequencies will not be reduced in amplitude. After playing with the software for a few weeks, I figured out that I could avoid the problem by increasing the number of points in the envelope to 4096, and by raising the envelope six decibels above the default provided by the software. My theory is that this allows the envelope to more closely match the frequency distribution of the noise, and raises the envelope farther above the noise distribution, making it less likely that spikes will "poke up" through the envelope. To increase the number of points in the envelope, the software requires that you increase the number of FFT points, as well. I have found that increasing the FFT points above 4096 causes the hiss to surge in and out, and softens attacks and decays. I have not figured out a way around these problems.
I should note that I am very happy with my Sound Forge software. It is easy to use, has good features, and is robust. Also, the manuals are well-written.
You need a fast CPU to use the hiss reduction feature of Sound Forge satisfactorily. It does a lot of number crunching. On my system, with a 350 MHz Pentium II, it takes a couple of minutes to apply hiss reduction to a three minute recording. Since the algorithm is CPU-bound, the speed of the software is proportional to the speed of the CPU.
I use MP3 and RealMedia audio compression to make the audio files smaller. Without compression, no reasonably patient person would wait for the files to transfer across a normal phone line. CD quality audio files take up 10 Mb per minute. 78 RPM records are monaural, so we can reduce that to 5 Mb per minute. So, a three minute recording would take up 15 Mb, which would take about 38 minutes to download with a 56K modem. With 11:1 compression, a three minute recording takes only about 1.4 Mb, which would take three to four minutes to download with a 56K modem. MP3 and RealMedia compression is variable - the more compression you use, the worse it sounds. I have determined through listening tests that I can't compress more than 11:1 without hurting the sound significantly.
I do not use streaming audio because it doesn't sound good. I have listened to streaming audio over a modem line, and it sounds pretty terrible. It might be OK for voice communications, but not for music.
Here are some sound clips that demonstrate the effects of the various stages of noise reduction. I have made the clips available in RealMedia and MP3. The RealMedia clips are a lot clearer than the MP3 clips, but they also have more high-frequency distortion.
The record I used for the sound clips is I'm Just Wild About Animal Crackers by the California Ramblers on Brunswick. I chose this record because it has a fair amount of noise.
First, let's see how the record sounds with different styli. Here it is with the 3.3 mil stylus, in RealMedia format and MP3 format. It sounds pretty bad.
Here's the same portion of the record with a 2.5 mil stylus, in RealMedia format and MP3 format. It sounds much better.
Here's the same portion of the record with a 2.0 mil stylus, in RealMedia format and MP3 format. This sounds a little better than the 2.5 mil stylus. Let's go with this one.
Now let's listen to the same clip after pop and tick elimination have been applied, in RealMedia format and MP3 format.
After the pops and ticks have been eliminated, we apply hiss reduction. Here are the results of 20 dB of hiss reduction, in RealMedia format and MP3 format.
Finally, we cut off all frequencies above 7.5 kHz using the parametric equalizer in Sound Forge, to get rid of hiss in areas of the frequency spectrum where there is no signal. Here are the results in RealMedia format and MP3 format.
NOTE: I once had an example here of noise reduction done incorrectly, using the record Black Rat Swing by Son Joe. I have removed this example, because the new version of the noise reduction software makes it obsolete.