Correct Azimuth when capturing Monoural Cassette Recordings

I am often asked to make CD’s from cassette tapes that were recorded in mono. More often than not I find the channels skewed out of time alignment due to azimuth error in the recorder. This is much more offensive in the sound when the two channels are mixed because there’s more time skew between the channels than across the width of an individual channel’s track on the tape. If you copy from only one channel, you avoid this effect but you lose about 3 dB in signal to noise ratio. Best solution is to capture the tape in stereo, then split to mono, go to a spot near the middle of the recording and align the tracks. This generally must be done separately for Side A and Side B of the cassette. Combining the two tracks as mono preserves as much signal to noise ratio as you’re going to get before applying Noise Removal. I do the Noise Removal separately on the left and right tracks because typically they weren’t quite in balance and the high frequency rolloffs don’t quite match. The results usually justify the time spent.

  • DickN

<<<the high frequency rolloffs don’t quite match.>>>

Can you apply average high frequency correction based on knowing the head skew value?


As you imply near the end of your post, azimuth error implies not only time “skew” between the L and R channels, but HF rolloff in each channel as well. IMO best practice is to adjust the azimuth of the playback machine to match what was recorded. This will simultaneously correct both problems, and you won’t have to go through the many steps you describe. If you do decide to further tweak the recording, that process will be easier as the tweaks should be smaller.

You might want to make a “calibration” cassette before messing with the azimuth. Use Audacity to record a 10 kHz tone with length of a minute or two. Using a new, high-quality cassette record this tone at -20 dB without Dolby. Now you have a standard that you can use to return your deck to a “normal” azimuth setting.

– Bill

Yes, I agree tweaking my deck’s azimuth to match the recording would be the ideal fix. Call me lazy, but the mono mix exhibits conspicuous comb-filtering long before the individual channels do. Moreover, if the HF rolloff were due mainly to the azimuth error, the two channels would still be reasonably matched. I suspect it’s usually a combination of a dirty record head and a poor quality cassette. I note your point that matching my azimuth to the tape would reduce the amount of fiddling I have to do later, and agree with that too.

Koz - Did the above answer your question re: applying an averaged HF correction?

I advised de-skewing the channels near the middle of the tape for a reason: Some recorder/cassette combinations produce a skew that varies from start to finish of a tape side. One suspect here is the tension sensing probe that some low-end decks use to detect end-of-tape, but that’s a guess - I haven’t done any research. My thinking is if the probe is off-center or tilted, or if it pushes the tape against the cassette housing and the housing isn’t perfectly flat in that spot, it strains the tape nonuniformly and can skew the tape in azimuth. Then if the tape tension varies with radius on the supply spindle, the skew will vary.

I should add three points here:

1: Listening to or examining the result after de-skewing in the middle will tell you whether you ought to divide the side into multiple clips and de-skew them individually.

2: If the skew is fluctuating as the supply spindle turns there’s no point in combining the channels at all - you’ll still get that mini-flanging effect throughout, which is more annoying than the hiss.

3: Combining the channels has another benefit besides 3 dB improvement in s/n ratio: Minor tape dropouts that aren’t due to mechanical damage are often not coincident in both channels. A 6 dB dropout is not as bad as a 60 dB dropout, and if you combine the channels after fixing the ones that can be fixed you’ll have fewer of them.

BTW, I’ve found that a single frequency is not the best test signal for tweaking azimuth, even though that’s what standard test tapes have. Sometimes the central peak occurs at slightly different settings in the two channels (maybe head mfg tolerances, who knows?). If I’m using a standard test tape (used to have one, lost track of it years ago) I split the difference with the max amplitude adjustment and then put a scope on the output and matched the phase of the two channels. Back then, I had a basic single-channel scope so I used the Lissajous test: Put one channel on vertical and the other on horizontal and adjust for a diagonal line on the screen sloping up and to the right. If you don’t have a scope, you might have a way to measure the differential voltage across the two channels and null that out. Monoural white noise, however, provides a very sensitive test. When the two channels are combined, the azimuth can be set quickly and very accurately by listening - no need even to meter the output. Besides, it just makes sense to be tweaking out the effect to which the ear is most sensitive, using a test signal that maximizes it. If you prefer instrumentation and have a scope, you can still use the Lissajous test on white noise and you don’t have to search for the global max among multiple peaks.

As this spreadsheet shows, the HF loss slope becomes quite steep once the loss exceeds about 4 dB. The -4dB point occurs in each channel at about the frequency where the two channels cancel in the sum (before de-skewing, that is), assuming no other contributing factor such as dirty head, bad tape-to-head contact, etc. Compensating via equalization is therefore feasible only if the azimuth error is constant. It will significantly increase the HF hiss. Since the tape hiss is primarily due to the granularity of the recording medium, it is unaffected by the azimuth error in the recording. A sample of post-compensation noise should therefore remain valid as long as the compensation is unchanged, so it shouldn’t matter much whether noise removal is applied before or after the compensation.

In the spreadsheet, K is the ratio of Track Width to Track center-to-center spacing. I assume no space between the L & R tracks and so let K=1. I haven’t been able to find information about the actual track dimensions on cassette format. If anyone supplies this information, I’ll revise the spreadsheet. I uploaded it as an .xls so both Excel and Open Office can open it.

x may be thought of as normalized frequency, although it’s really the ratio of the head’s aperture distance to the wavelength on the tape. The aperture distance is how far the tape travels while the head is still picking up a single flux change recorded on the tape.

Suppose, for example, the time offset DeltaT between the channels is 100 microseconds. Then assuming K=1, x = F * 10^-4, and F = 10 KHz corresponds to x=1 (attenuation in each channel is infinite). At 5 KHz, the channels cancel in the sum, and x = 0.5. This is the -4dB point in each channel.

All of the above is theoretical and untested, at least by me.
Tape Azimuth Response.xls (10 KB)

Why not to create an Audacity Plug-In to correct the “Phase Correlation” between the Left and Right audio channels? ???
This is one of the “missing things” I get testing Audacity.

I haven’t used my spreadsheet in a few years now. If you can live with adjusting your cassette deck’s azimuth in real-time while capturing the playback, there is an excellent way to do it which works equally well for stereo recordings provided there is at least one source in the mix that was originally in-phase between the two channels. I’ve found this program extremely useful - it contains a variety of virtual test instruments including a function generator, spectrum analyzer and oscilloscope. The spectrum analyzer part can compute the cross-correlation of the two channels vs frequency, and any signal component which is co-phased between the channels contributes to the display a normally flat horizontal line at zero phase. Any skew between the channels yields a curve away from zero, or even multiple cycles of phase rotation if the skew is severe. You can adjust the azimuth while observing the display. I now consider this tool indispensable for copying cassettes because I can usually keep the azimuth error down to less than a sample period while capturing. It’s quite revealing just how bad even prerecorded (i.e., commercial) cassettes are, and the worst part is the skew is constantly varying with the rotation of the spindles and over the length of the tape. Of course, you can’t really compensate as fast as the spindles are rotating, but you can correct the average as it drifts over the length of the tape. Resolution is a fraction of a sample time at 44,100 Hz.

The program is at I’m still using an old version and just now discovered it got a data management update last year which might fix some issues with Windows 7 and Vista. I’ll have to try it!