Hiss as a signature#
Last week you measured one copy against a source. This week two copies share something, and the question is what that entitles you to say.
A population with no records#
Through the 1970s and 80s, cassette duplication carried most of the recorded music in South Asia, outside the industry that thought it owned the business. Small operators dubbed from whatever they had — usually somebody else’s dub.
From Peter Manuel, Cassette Culture (1993), introduction and chapter 3.
Two copies, W and X#
Both carry a raised noise floor.
Related?
No. Every analogue dub raises the noise floor. A property of the process, not of any particular history.
Two copies, X and Y#
Both carry a 40ms dropout, left channel, at 00:12.400.
A single physical event at a fixed location. Two dubs producing it independently would be a coincidence nobody should accept without argument.
Four explanations for a shared feature#
- It is in the source. Rule this out first, every time.
- Every copy acquired it separately, because the process does that.
- Same equipment. Shared equipment is not shared ancestry.
- They descend from a copy that already had it. Only now.
The one worth slowing down for#
Frequency response is a property of a machine and its settings.
Two dubs made on the same worn deck will share a rolloff above 12 kHz whether or not they have anything to do with each other.
It produces exactly the pattern shared ancestry produces.
What makes the difference#
Not rarity, exactly.
Independent occurrence is implausible — the feature is specific and arbitrary enough that producing it twice by accident would be a coincidence.
A frequency-response curve does not have that property; a one-off dropout at a fixed timestamp does.
A claim about mechanisms, not statistics#
Proposing a relationship asserts that you cannot think of a plausible mechanism that would produce the feature twice independently.
State that assumption. Someone may think of one.
What this week does not do#
You are pairing copies, not building a tree.
With four copies and one diagnostic feature you can support one relationship — not how many copies sat between, in what order, or whether the intermediate still exists.
Week 6 builds the tree. This week earns one edge of it.
Before Wednesday#
Open the week 2 pack: feature-table.csv and
features.txt. Ten minutes.
Workshop: Hiss as a signature — one source and four copies, six features, and a pair you have to propose and then attack.
Text walkthrough#
For a reader who cannot use the deck view: cassette duplication in South Asia through the 1970s and 80s produced a population of copies with no record of what came from what, which is the situation this course reads objects instead of records for. Whenever two copies share a feature, four explanations compete: it may already be in the source, every copy may have acquired it separately because the copying process does that to everything, both copies may have been made on the same equipment (a shared deck produces a shared frequency-response rolloff without any shared ancestry at all), or the copies may genuinely descend from an intermediate that already had it. A raised noise floor shared by two copies fails all but the second explanation. A 40-millisecond dropout in one channel at one timestamp, shared by two copies, is different: it is a single event at a fixed location, and independent occurrence in two separate dubs would be an implausible coincidence, which is what makes it evidence for a shared ancestor rather than for the process or the equipment. With four copies and one such feature this only supports pairing two of them — it says nothing about how many intermediates sat between them, which week 6’s larger witness set is needed to address.