What Does Cassette Noise Look Like? (July 2022)
As of this writing, I have converted over 200 cassettes in my business, about half commercially produced and half personal recordings. Among those commercially produced, there’s a consistent, though not universal, pattern of analogue signal at the beginning of the audio file created in the conversion process. The waveform pattern is depicted in Figure 1, where the horizontal axis is time, and the vertical axis is amplitude (left channel on top, right channel on the bottom). This screen shot has four sections created by chopping up a digitized cassette file, each section four seconds long, and containing signal originating from:
-
My sound system (cassette deck, amplifier, sound card, and interconnects) after starting the digitizing software but before pressing the deck’s Play button.
-
The cassette’s leader tape.
-
The cassette’s magnetic tape before the music begins.
-
The cassette’s magnetic tape after the music begins.
Figure 1. Waveform of system, leader, tape, and music amplitude.
While the music in the last section is readily apparent, it’s difficult to discern from Figure 1 what, if anything, appears in the first three sections. (When working, I have the advantage of seeing this fully expanded on a 32” monitor). I included this figure to provide a sense of the amplitude of the signals in the first three sections—which represent sources of broadband noise—compared to the music section. Figure 2 shows the same 16 seconds magnified by zooming in 24 dB vertically. The noise in the system section remains barely detectible, but the leader and tape noise in the middle two sections is more pronounced.
Figure 2. Waveform of system, leader, tape, and music amplitude at +24 dB zoom.
There’s more to be learned from these sections. Figure 3 is a spectrogram of the same 16-second mini-track. The horizontal axis remains time; the vertical axis is frequency; and the colors reflect amplitude (low to high: black, dark to medium blue, dark to light yellow, white).
Figure 3. Spectogram of system, leader, tape, and music frequencies.
The following can be gleaned from Figure 3:
-
System: There's a dark blue strip at the bottom of the graph in the first section, though barely visible in this screenshot. It represents very-low-frequency and very-low-amplitude noise.
-
Leader: The second section has mostly low-to-mid-frequency (up to ~500 Hz) and very-high-frequency (10-20 KHz) noise. Both are low-amplitude.
-
Tape: The third section—after the cassette has played past the leader but before the music begins—is similar to the leader section. There are low-to-mid-frequency and very-high-frequency components in the noise, but more frequencies appear (wider blue bands), and the amplitude is greater compared to the leader (brighter blues and the appearance of yellow in the very low frequencies at the bottom of the graph).
-
Music: The final section has more frequencies present, most at much higher amplitude (brighter color)—not surprising, since this is what music is supposed to look like in a spectogram. But note the band of dark blue at the top of the music section's graph (roughly in the 13-20 KHz range). It appears identical to the dark blue band at the top of the tape section, indicating the presence of tape noise in the music.
I routinely remove system noise. Even though it’s unlikely to be noticeable to the listener, there’s no sense in leaving identifiable noise in what will be delivered to a customer when it’s straightforward to remove.
Although leader signal is captured in the conversion process, the leader is irrelevant to what will be saved from the cassette. I delete it before any further processing of the audio file.
In many cases, a cassette’s music will sound fine without further noise reduction since the amplitude of the music typically overpowers the noise. However, tape noise often remains audible in very soft passages or rests within the music, as well as in the gaps between tracks—despite its comparatively low amplitude. If it’s audible, I prefer to remove it.
Figure 4 is the spectrogram following removal of the system and tape noises from the tape and music sections. The dark blue at the top of the music section’s graph has almost disappeared; what remains are the music’s high harmonic frequencies. (That’s what would happen in a perfect world, but in practice, noise removal isn’t that simple. Removing the high-frequency noise matching the noise sampled from a quiet segment may also change the timbre of the music by removing too much of the high-frequency signal that’s actually part of the music. Iterative testing of noise removal parameter settings is often required to remove as much noise as possible while preserving as much of the music as possible.)
Figure 4. Spectogram with system and tape noise removed from the tape and music sections.
You may have noticed some low-frequency “blobs” remaining in the tape section in Figure 4. This residual noise can result from irregularities in the tape or, more likely, electromagnetic or environmental factors that have affected the tape’s oxide layer. If that residual noise is evident in the intertrack gaps, it can be removed after the tracks are split.
My cassette deck has the options of applying Dolby B and C de-emphasis, and I invoke that capability when a cassette is labeled as having been created with Dolby noise reduction. (See What Are Those Codes on Cassettes? for a discussion of noise reduction on cassettes.) If there are no markings, Dolby de-emphasis may still be useful in removing high-frequency hiss (the top part of what appears in the Tape sections in Figure 3). The risk is that doing so will also remove the music’s high frequencies that are supposed to be there. Typically, I listen to parts of the cassette with and without Dolby B and Dolby C de-emphasis, and I decide which approach is the best for knocking out tape noise while preserving the fidelity of the music. If I bypass Dolby de-emphasis, removing tape hiss can be accomplished by identifying a sample of tape-only noise from the beginning or end of the cassette or in an intertrack gap and removing the sample’s noise pattern from the music, as I have described above.
At the beginning of this article, I mentioned that about half of my cassette conversions were of personal recordings. While removal of system noise and deletion of leader capture are identical to what I do with professionally produced cassettes, treatment of tape noise varies widely depending on the history of the cassette. Cassettes with one-time recordings are usually straightforward, but those with newer recordings on previously recorded tape and those that are copies of another cassette or other medium present more challenges. Often there is no blank tape between the leader and the beginning of recorded audio; users occasionally and unwittingly begin recording on the leader before it has been transported beyond the record head. Similarly, users often continue recording until the cassette has run out of tape and stops. Moreover, different parts of the tape may have different recording histories, and each segment will need to be addressed separately. Noise removal for personally recorded cassettes must be conducted on a case-by-case basis. (For further information, see Technical Challenges Posed by Personal Recording and Content Challenges Posed by Personal Recordings.)
Back to Paul’s Blog & Contents