User:Mjb/Ripping vinyl

Here are some random notes about my vinyl ripping process.

Physical connections

For vinyl, unless you have a modern turntable (phonograph) with USB output, you need to connect the analog outputs of the turntable to a preamp. The preamp boosts and properly EQs the very weak input signal coming from the turntable so that it's at the standard "line level" expected by other devices.

Examples of preamps:

• a receiver with inputs labeled phono
• an old DJ mixer with inputs labeled phono
• a modern phono-to-USB box.

If the preamp doesn't have USB output, then you need to connect the preamp's line-level outputs (e.g. "tape out", "rec out", or "monitor") to the audio "line in" inputs on your computer.

If you are ripping tape instead of vinyl, you don't need a preamp; tape decks have line out connectors which can go directly to your line in.

All of these connections use standard "RCA"-type patch cables, the same kind used by TVs, VCRs, video game consoles, and consumer audio gear since the 1950s. You do not need to buy expensive cables, but it does help if they are well-made and have good shielding, otherwise you risk them picking up EM interference. If you keep hearing the sound of a local radio station or noises from cell phones, you probably need better cables. Otherwise, they are fine. In my experience, 4 out of 5 random cheap cables are perfectly OK.

DJ mixer connections

I am using a DJ mixer as my preamp. In addition to the crossfader and input level controls, it has 3 tone controls and a mono/stereo switch, plus an internal amplifier with a volume knob. The mixer has two line outs. The "rec" output bypasses the mixer's internal amp, while the "amp" output carries the processed audio. So I have the rec out going to my computer's soundcard, and the amp out going to my external sound system.

When ripping, I don't use the external sound system at all; I don't want any loud noises or bass vibrations feeding back into the needle.

I also try to ensure the mixer is only outputting sound from the one phono source. The crossfader is all the way to one side, the slider for the phono input level is set as high as it can go without distorting (depends on the loudness of the record, but usually around 70%-80% of maximum), and all the input levels of the other channels are at zero. More on this in the testing section below.

Recording software

To do the recording, you need audio editing & recording software, a.k.a. a wave editor. I use Adobe Audition, which has a lot of good features but is not free. A lot of people use a free alternative called Audacity. If you were to use Audacity, it requires some extra configuration to disable the default addition of "dither" (noise like tape hiss) to all the files you save with it, so be careful with that.

Configure the software to get its input from the appropriate device, and to record in stereo.

Input levels

Some decks, preamps, mixers, and computer soundcards may have input volume (gain) controls which needs to be adjusted to ensure the signal is as strong as it can be without overloading and distorting.

These devices also may have built-in dynamic range compressors/limiters which protect from overloading but flatten out the sound. My computer's soundcard, for example, begins flattening the input when the level gets within 3 dB of the maximum, so I try to make sure the loudest parts don't exceed that level.

Know where your OS's audio controls are and what they all do. Disable or set to zero any special audio enhancement features of your hardware; this includes stereo/mono switches and tone (bass/mid/treble) controls.

Testing my input levels

In Windows, in the Sound control panel, Recording tab, (Realtek High Definition Audio) Line In Properties, Levels tab, the Line In slider can set the input gain from 0 to 100. What do these numbers mean?

Here is how I figured it out:

My soundcard only supports 16-bit recording. If I have Audition record at 24-bit, the lowest 8 bits of each sample are random numbers, just noise (I guess from the Windows mixer), so I record at 16-bit.

With no incoming signal (DJ mixer powered off) and this slider at 0, the background noise is extremely low. Sample values range only from -1 to +1, with the average RMS of my line noise coming out to about -99.3 dB (as measured by Audition's Amplitude Statistics window with "0 dB = FS Sine Wave" and "Account for DC" checked, plus "Window Width: 50 ms").

As the slider goes up from zero, the level jumps, but the higher it goes, the smaller the jump. Here are the average RMS levels for various slider settings, and the sample value min/max. The last two columns are for when the mixer and turntable are powered on, but nothing is playing. These values will vary slightly depending on line noise conditions.

  0 -99.3   -1/+1    -94.4   -2/+2
  1 -94.2   -2/+2    -90.5   -3/+3
  2 -91.4   -3/+3    -88.5   -4/+4
  3 -89.9   -4/+4    -87.5   -5/+5
  4 -87.5   -5/+5    -83.1   -8/+8
  5 -85.8   -7/+7    -85.2   -8/+9
  6 -84.1   -7/+9    -81.7  -10/+10
  7 -83.6   -8/+10   -79.9  -11/+12
 50 -68.2  -50/+64   -64.4  -69/+75
100 -62.4 -102/+127  -57.3 -139/+165

I suspect the positive skew is just due to my electrical line noise being a little stronger in one channel.

So, for this particular soundcard, the numbers are how much you want to amplify the input to the ADC, on a logarithmic scale. That is, zero does not change the input level at all, 1 boosts it about 5 dB, 2 boosts it about 8 dB, 3 boosts it about 10 dB, and so on, with 100 boosting it about 37 dB. So unless you are dealing with abnormally quiet signals, you want to keep this control set very low. However, I found that it's not adding any more noise at higher levels (e.g. when I record at 0 and amplify by 37 dB, the noise is exactly the same as recording at 100).

As I mentioned, I have to keep the DJ mixer's phono input level at about 7 out of 10, or the bass hits on a fairly loud drum & bass 12" will start to clip, regardless of the input gain level. This suggests that the mixer's sliders can amplify the input from the turntable, not just attenuate it.

With the mixer's slider at 7, and the mixer's volume knob at the halfway point, these bass hits peak at 0 dB on the mixer's VU meters. VU meters are a crude analog metering system showing averages, not instantaneous peaks, and where zero is roughly -18 dBFS. Indeed, with the soundcard's input gain at 0, Audition shows the loudest instantaneous peaks are about -14 dBFS, although most of them are down around -18 to -22.

Conclusion: If I set my mixer's master volume at the midpoint, I can use 0 dB on the mixer's VU meter as an ideal peak for what's going out of the mixer. On the computer, I settled on setting the Line In input gain control at 3, which gives about a 10 dB boost and thus makes the actual peaks be about -3 or -4 dBFS in Audition. This will probably be between 3 and 6 on the control, depending on the loudness of the record.

Why -3 or -4 dBFS? Because if peaks are going over -3 dBFS, they might be getting compressed. Most soundcards have built-in dynamic range limiters to protect their circuitry from excessive input voltages. You can see them in action if you crank up the volume coming into the soundcard, and it just squashes everything into the 0 to -3 dBFS range and never clips.

Update: If I configure Audition to use ASIO input (after installing ASIO4ALL, which provides ASIO wrappers to WDM APIs to bypass the Windows mixer), the input gain control doesn't do anything. However, I get a signal equivalent to setting the control to 6. Why? I don't know.

Sample rate and bit depth

The sample rate and bit depth to capture must be chosen. Higher values use more space.

Set the same values in both the software and your OS's sound hardware controls; otherwise it may refuse to work or it may just silently convert to what you asked for in the software.

The choice of sample rate affects the audio frequency (pitch) range. The upper limit will be slightly under half the sample rate, e.g. if you sample at 96 kHz then you will get pitches up to about 48 kHz. Most records have nothing but noise above 30 kHz, human hearing maxes out at 20 kHz, and FM radio maxes out at 16 kHz. So for pretty much anything you ever record, the standard 44.1 kHz or 48 kHz are good sample rates to use. 48 kHz is standard for audio synced with video, and 44.1 kHz is the only rate supported by audio CDs. So basically you just choose 44.1 or 48. People who misunderstand the science behind it tend to think higher sample rates must be some kind of improvement, but it really just wastes space.

Bit depth determines the dynamic range (number of possible volume levels, basically) and also the amount and volume level of "quantization noise" (raspy, buzzy, distorted sound) resulting from the rounding of fractional levels to whole numbers. Modern soundcards can only do 16-bit or 24-bit sampling, and at these levels, quantization noise is not an issue unless you are recording the sound of butterfly wings flapping in an anechoic chamber. An empty record groove or blank tape will have a noise floor at around the 9 to 12 bit level, and all the music will be well within that range, so 16-bit is fine and is standard for CD. Again, people think higher must be better, and 24 would give you more room to use very low input levels or capture the sound of those butterfly wings, but for recording from vinyl or tape it's just wasting space. (You may also see 32-bit floating-point sampling is an option in your software. This doesn't hurt, but it is not higher quality; the hardware is delivering, at most, 24-bit, and the software is just saving it in a way that removes some limitations which are of no consequence to you.)

Recording and editing

Once you have set up everything, you just hit your software's Record button, and start playing your record or tape. Stop when you're done, and save your work in progress.

After you have captured some audio in this way, it probably needs some editing:

• splitting into separate files
• conversion to mono if the input was mono
• trim the ends; maybe fade in and fade out
• adjust the overall volume
• adjust the left/right balance by adjusting volume or using EQ in one channel
• apply a small amount of EQ to correct for deficiencies in the recording chain (maybe a slight treble boost)
• reduce noise (see next section)

To keep the sound true to the original, I avoid the temptation to use "remastering" types of effects like heavy EQ, stereo expansion, or any kind of dynamic range compression.

Reel-to-reel notes

When recording from reel-to-reel tape, I found that many of the tapes are recorded with 1 mono track per side, but my stereo player plays both sides at the same time, with the left channel being the front side and the right channel being the back side, playing in reverse.

This saves a lot of time since both sides play at the same time, but it means I have to use my software to reverse the right channel and save it separately from the left, and to make each file be centered mono. When doing this, I may also change the sample rate, depending on the content. I use the software's spectrogram view to visually see what frequencies are in the signal, and I decide whether to resample to 32, 24, 16, or even 8 kHz in some cases. If there's nothing but noise above 9 kHz, for example, I'll resample to 24 kHz so that 12 kHz will be the upper limit. Reducing the sample rate cuts out the extra hiss and makes the audio take up much less disk space. Resampling is safe to do in the software I use, but some software does it poorly and messes up the sound. http://src.infinitewave.ca/ is a good site to see which apps do a better job.

78 notes

78 RPM records are usually made of brittle shellac instead of vinyl. Almost without exception, they are in mono and have a very wide groove. A normal, relatively tiny stereo needle made for vinyl will sound terrible when you play a 78 with it.

To play a 78 properly, your turntable needs a special cartridge and needle, like this: http://www.amazon.com/Audio-Technica-AT-MONO3-SP-Moving-Cartridge/dp/B0002ERE30/ (There are some cheaper ones available, but they are not properly made; they read the rough surface at the bottom of the groove and add that noise to the music.)

Ideally, your turntable also needs to be able to spin at 78 RPM, although it will work just as well if you use 45 RPM and play with the sample rate. You can sample at 45/78ths of the sample rate you want, and then tell your audio editing software to pretend it's the rate you wanted. For example, play the record at 45 RPM, sample it at 27692 Hz (this means relying on your software to do good conversion from the hardware's rate), then make your editor interpret the sample rate as 48000 Hz. This is not the same thing as sample rate conversion.

Despite their high rotation speed, most shellac 78s were recorded with no frequencies above 10 kHz (1926–1937) or below 200 Hz and above 4–6 kHz (before 1926). Your phono preamp applies the RIAA EQ curve for vinyl LPs, but 78s all used different curves. They also were not necessarily cut at exactly 78.0 RPM—in fact, most weren't! There is a lot of research online about this, but unfortunately for any given record, there is not going to be a lot of agreement about what the precise playback speed and correction EQ really should be, and you may prefer the sound with a more aggressive EQ than what is prescribed.

• Victor acoustic (1909–1926) = 76.59 RPM; 45/76.59*48000=28202
• Tops (1947–) = probably 78.26 RPM; 45/78.26*48000=27600

Here's my procedure so far:

• with soundcard configured for 16-bit, 48000 Hz (its maximum), have audio software record at rate from list above, 32-bit float
• apply inverse RIAA EQ
• interpret sample rate as 48000
• normalize to -3 dB
• trim ends
• further EQ and resampling not yet figured out

Reducing and removing noise

Electrical noise

One kind of noise I've had to deal with is electrical "ground loop" hum and/or buzzing which manifests as spikes at multiples of 60 Hz.

Another kind of noise is interference from poorly shielded audio cables and related wiring, including the circuitry inside your computer; it manifests as random whines, clicks, buzzes that correspond to things happening with your hard drive or on your screen, or even as the sound of an analog radio station because the cabling is creating an antenna.

It is better to try to eliminate all of these kinds of noise in your physical setup rather than with noise reduction features of software; otherwise your attempts to reduce it are likely to damage the music. It can be frustrating to try to figure out the cause of noise, let alone to get rid of it, but things that can help are a different soundcard or RCA cables, adding an "isolator" in your cable modem line, making sure everything is plugged in correctly and well-grounded, or simply ensuring that your computer and audio gear is all plugged into the same household electrical circuit.

When recording from tape, I don't worry about the extra noise coming from the player's circuitry or the background noise caused by the magnetic particles on the tape. Sometimes, though, there will be an annoying ground loop hum recorded on the tape. I use my audio editor to reduce or remove this hum by using an FFT filter (a kind of EQ) to emulate a set of notch filters at 60, 120, 180, 240, and 300 Hz. It often takes several tries before I get acceptable results; I have to be careful not to try to make the filters too steep. Often, it's best to just try to reduce the noise so it's not so distracting, as opposed to completely removing it.

Declick, decrackle, depop

For vinyl I usually have to do a lot of "declicking", which is using the editor to remove or reduce the audibility of ticks and pops caused by scratches and crud in the record groove.

Declicking can be very fast and automatic, but can also audibly hurt the music because the declicker can't tell the difference between a noise that's not supposed to be there and musical sounds with similarly spiky waveforms or fast attacks, like the sound of a trumpet or the beginning of a tap of a cymbal. So I tend to declick in a more manual fashion by finding the clicks myself and applying an automatic declicker to just each spot of interest, or sometimes even manually moving samples around or excising unrepairable sections entirely. I have it down to a science, but it takes about 3 times as long as the play time of each song to fully clean up just that one song.

Manual declicking

In Audition, I use the spectrogram view (Shift+D) to see the clicks that produce broadband noise; they're vertical stripes that are easily spotted. In the waveform, I select the affected samples (usually in one channel only, using up/down arrow keys) and Ctrl+U them to get a near-perfect fix.

Bassier pops are harder to spot, but are easily heard and you eventually get an eye for them at the bottom of the spectrogram. When I find one, I select the "lump" in the waveform plus a fair bit beyond on each side of it, then apply a bass-rolloff EQ, or Ctrl+U, or both, listening to the result to make sure it's acceptable. Before doing a Ctrl+U I often adjust the top edge of the selection in the spectrogram view so that only the lower frequencies will be affected.

When the pop is in both channels, it's often out-of-phase, i.e. the lump goes up over the zero line in one channel, and down below it in the other. For little ones it doesn't matter, but the louder ones will repair better if the channels stay in phase. So I select the affected samples in one channel, invert them to match the other channel (usually matching the one that forms a peak rather than a valley) and then Ctrl+U both at once. On rare occasion, I'll select individual samples or spans of a few samples and use the volume control to move them into more favorable positions. (If only they'd just give us a pencil tool like I used to use in Sound Forge!)

For the massive pops, sometimes some surgery is needed, copying a bit of the waveform from the other channel or from an undamaged section, and pasting or mix-pasting it over the pop. In extreme cases, when the pop is in both channels and has no musical content (it's just a bright solid line in the spectrogram), I'll just consider it a skip, the needle getting bounced around, and I'll cut it out completely, shortening the length of the file. These are rare in my own rips, because if the needle is getting tossed and slammed against the groove walls, the problem is likely with the turntable setup and should be fixed there.

When there's too many minor clicks to clean up manually, I'll also use Audition's declicker, or Izotope's (slightly better IMHO), with very light settings, then undo/redo (Ctrl+Z, Ctrl+Y) repeatedly to see the waveform differences and make sure it didn't "fix" instruments with sharp attacks. And of course I also listen. Sometimes I'll have to run the declicker with more aggressive settings on one beat or bar at a time, selecting everything in between the sharp attacks that I want to preserve.

Before doing any of that, I make sure the channels have no DC offset, and have rolled off the deep bass (30 Hz and below, I think) on an even slope in order to reduce tonearm resonance, rumble, and through-the-floor noise. I've read that declickers work better without extraneous bass. I also try to get the channels balanced. Depending on what stylus I use, the left channel often has more bass than the right, so I use EQ to make them more even. (I've been unable to figure out a way to correct it in the turntable setup.)

After declicking, if there's grounding hum or other tonal electrical interference that's audible and distracting, I'll use general noise reduction to reduce the noise by 20 dB, but really I prefer not to do this because the problem can be resolved externally. I don't worry about the background whoosh of the vinyl. It's just not worth the risk of trying to remove.

Normalization

I used to normalize each track on a single or EP to have 0 dBFS peaks, thinking it didn't make much difference, and before I learned about "inter-sample overs" and (more importantly) ReplayGain.

Around 2004, I realized that it's normal for different sides of a single or EP to be mastered at different overall volume levels. So now I try to preserve the relative volume levels of the tracks, never adjusting the volume of one track without adjusting all the other tracks on the same side by the same amount. Basically it means treating each side as a separate "album", and that goes for ReplayGain scanning, too.

As for overall volume, I generally aim for whatever will produce a ReplayGain value within a couple of dB of zero.

Fixing wrong-speed recordings

Sometimes someone (possibly me) will accidentally record a 33 RPM side at 45 RPM, or vice-versa. The best way to fix this is to re-record it! But you can fix it digitally in Audition, too.

Speed up 33 to 45

If you are speeding up the rip (it was recorded at 33, but you want to pretend you recorded it at 45), I think the best way is to do this: Edit > Interpret Sample Rate ... multiply current sample rate by 1.35, e.g. 44100 becomes 59535. Then do File > Save As and under Sample Type choose the original sample rate (also, Advanced: Quality 100%, [X] Pre/Post Filter), and leave the bit depth Same As Source, Dithering Disabled.

Slow down from 45 to 33

If you are slowing it down from 45 to 33, then the method depends on the original sample rate.

If the original sample rate is 96000 or 192000, you can use the same method as above, but divide instead of multiply. For example: Edit > Interpret Sample Rate ... for 96000 you enter 71111.111111111111111. Then you must reduce the sample rate to 44100 or 48000 when saving.

This doesn't work as well when the original sample rate is low, because when you interpret e.g. 44100 as 32666.66666666667, you lose frequencies above ~16333 Hz, which will make an audible difference. So for slowing down a recording that's at a 44100 or 48000 sample rate, you should use an effect:

Effects > Time and Pitch > Stretch and Pitch (process)

• Algorithm: iZotope Radius, Precision: High
• Stretch and Pitch: Stretch 135%, [X] Lock Stretch and Pitch Shift
• Advanced: [ ] Solo Instrument or Voice [ ] Preserve Speech Characteristics