Cosmik Debris Live: the hi-hat audit

The question

One Songsterr tab, one narrow question. Tab s1154774, titled Cosmik Debris Live, is an AI-generated transcription uploaded by DavidxHanes in March 2025. Its drum track writes 1,053 closed hi-hats, 81 open hi-hats and 24 pedal hi-hats. This page asks whether that hi-hat writing describes what the drummer played.

The tab carries its own answer key. Songsterr stores the video each tab is synced to, and s1154774 names three. The first is the reference recording, so the audio was pulled and measured against the notation.

The audio, and which performance this is

The recording carries usable high-frequency content. Averaged across the whole track the 6–14 kHz band sits 29 to 34 dB below peak, and the hi-hat lives there. A hat verdict is physically possible on this source, which is why the question was worth asking.

Method, stated ahead of the results

1. Read the tab's notation from Songsterr's own JSON, unauthenticated, at dqsljvtekg760.cloudfront.net/1154774/1540143/<hash>/3.json.
2. Pull the synced video's audio to WAV, 44.1 kHz.
3. Separate the drum stem with Demucs htdemucs, so guitar and vocal sibilance stop contaminating the hat band. On the raw mix a hat detector returns 840 onsets in 60 seconds, which is nonsense.
4. Detect cymbal events from spectral flux in 7–14 kHz on the stem alone.
5. Separate open from closed by gap-ring ratio: the high-frequency energy still ringing at the end of the gap, divided by this hit's peak. A closed hat is choked by the pedal and dies. An open hat rings into the gap.
6. Validate every detector against a null, then trust it.

The hi-hat verdict on the mixed stem superseded

The tab marks certain blocks as entirely open and others as entirely closed. If those calls carry information, the blocks must differ in the audio. They do not.

The distribution behind that test is worth stating plainly. Ring ratio across 2,081 hat events is unimodal. It does not split into a choked cluster and a ringing cluster. On this source, after separation, closed and open hi-hat are not cleanly separable by any threshold. That is a limit of the recording as much as a finding about the tab, and it is reported here rather than hidden.

Four structural failures that need no alignment

1. The opening is impossible

Bars 1 to 4 write 81 snare hits, 14 crashes and 4 kicks in 11 seconds, with zero hi-hat. The pattern is kick, snare and crash together, then five bare snares, repeated. That is 99 drum notes in 11 seconds.

TAB bars 1-4, claimed: 99 notes in 11.0 s
AUDIO, actual first 11 s: 26 onsets of any kind
AUDIO, densest 11 s anywhere: 64 onsets (begins t=239 s)

The claim exceeds the densest eleven seconds in the entire performance. No sync offset repairs this, because no such window exists anywhere in the recording. The audio's opening is quieter still: one onset at 0.023 s, then nothing until 2.171 s.

2. Every pedal hi-hat sits in one block

All 24 pedal-hat notes in the tab fall inside bars 129 to 146. The other 141 bars contain none. A drummer's left foot does not switch on for eighteen bars and off for the rest of a song.

3. The open-hat sections are all-or-nothing

Bars 123 to 125 carry 23 open hats and zero closed. Bars 45 to 65 carry 225 closed and zero open. Real shuffle playing mixes them continuously. Blocks this pure are a classifier holding a state.

4. The hat line is thin, and in one place it vanishes withdrawn

Across 30-second windows the tab writes a median of 0.60 hat notes for every cymbal event detected in the audio. Between 349 and 380 seconds it writes five, all of them open, against 176 detected events.

The alignment test, with its nulls

Before any of the above can be called a transcription error, one question has to be settled. Does this tab track this performance at all. The kick was used to answer it, because the kick is the easiest drum to detect and it is independent of the hat question.

The kick detector was validated first. Across 24 windows of 20 seconds, all 24 show a statistically significant kick period, and the median local period of 788 ms agrees with the independent high-frequency autocorrelation estimate of 784 ms. The detector works.

The global test was run first and was unfair, so it was rerun. The tab holds a flat tempo. The band drifts from 72.6 to 82.4 BPM, and any transcription with a flat clock slides out of a live performance within a minute. The local test lets the offset float freely by up to eight seconds in each fifteen-second window. That is the most generous test available, and the verdict survives it.

The clock, precisely

The tab writes 120 BPM in 12/8 for 135 bars, which is a felt beat of 80 BPM. The performance sits at 76.1. The tab therefore runs about 5 percent fast, and it computes to 460.2 seconds against 486.4 of audio, a 5.4 percent shortfall. Those two figures agree, which is how the scattered short bars are explained: the 9/8 and the three 11/8 bars are dropped beats absorbing accumulated drift. They are not meter.

The second transcription

Songsterr holds exactly two Cosmik Debris tabs. There is no independent transcription of this live performance on the site.

The studio tab is a control worth having. It computes to 4:19.5 against a 4:14 album track, so its tempo map is sound, and its opening bar is a playable ride shuffle. Its open hats are 2.0 percent of the hat family. The live tab's are 7.0 percent.

What exists anywhere else, checked by search

A web search for a Cosmik Debris drum transcription returns commercial sheet music from Hal Leonard's Apostrophe (') book, Virtual Sheet Music, Sheet Music Plus, Sheet Music Direct and Stanton's. Every one of them is guitar tablature. The only dedicated drum transcription any search surfaced is Songsterr's own s412177. So the absence of a second drum transcription of this live performance is a fact about the world, checked beyond Songsterr's catalogue, and it is the reason s1154774 has had no correction applied to it.

The search did settle the drum chair. Ralph Humphrey was in the band from 1973 and played the Cosmik Debris session at Bolic Sound on 1973-05-29. Chester Thompson joined as a second drummer in 1974, so a 1974 live Cosmik Debris can carry two drummers at once. That matters for any future attempt to transcribe this recording, because a dual-drummer stage mix will not reduce to one hi-hat lane.

An independent source on the meter

A German musicological analysis of Apostrophe (') held in the local Zappa drum-source archive gives Cosmik Debris a duration of 4:16 and states its meter as 6/8 and 4/4. It sets out the basic rhythms, the blues schema behind the guitar solo and the strophe-refrain form. Its 6/8 is compatible with notating the shuffle in 12/8, since one 12/8 bar is two of 6/8. Nothing in it supports 9/8 or 11/8 bars.

The answer, from the isolated hat lane

The question was whether the hi-hat writing in this tab describes what the drummer played. On the mixed drum stem the honest reply was that the measurement could not resolve it, p = 0.80 on a unimodal distribution. On the isolated hat lane, across the complete 486.4 seconds and 1,091 hat onsets, it resolves.

The earlier p = 0.80 was not wrong, it was blind. A drum stem carrying kick, snare, toms, ride and crash alongside the hat cannot answer a question about hi-hat articulation, and the unimodal ring-ratio was that blindness showing up as noise. Splitting the kit six ways turned a non-result into a signed one.

And the same lanes settle the ride

The tab's largest ride section, 348.5 to 377.1 seconds carrying 109 ride notes, is the closest it comes. The ride lane peaks there at RMS 0.02053, its loudest anywhere, so the drummer does lean toward the ride across that stretch. The hat still leads it 1.52 to 1. The tab hears the move and overstates it into an exclusive ride passage.

The isolated kit, and what it corrects

The verdict above rests on a Demucs drum stem holding the whole kit at once. DrumSep MDX23C splits that into six lanes, kick snare toms hh ride crash, so the hi-hat can be measured on its own. The first 60-second block is analysed here. It covers the tab's bars 1 to 20, which includes the opening this page called impossible.