The earlier pass cut the tape into 100 equal 2:15 slices on a blind clock grid. A slice boundary landed wherever the timer said, so songs were cut in half and every false start, tuning pause and bit of chatter stayed in. That answered a request for tenths and it did not answer the real ask.
The detector treats a false start as a take the band tried again. It compares each take head-to-head against the takes that follow using chroma-sequence similarity with lag alignment, so the same pitch classes have to appear in the same order. Duration and loudness alone would mislabel almost everything, since talking is loud and a rehearsal room is never silent.
| Stage | Result |
|---|---|
| Takes segmented from the tape | 83 |
| First pass, default threshold | 31 kept, 17 survived re-check |
| Stricter pass, score 70 / min 60s | 20 kept, 13 survived re-check |
| Head trims applied | 6 files |
| Long take split into songs | 1 take into 3 |
| Final | 22 songs, 21 verified clean |
| Verdict | Count | Meaning |
|---|---|---|
| FRAGMENT | 46 | A stab, a riff check, a count-in |
| NOODLE | 9 | No steady pulse, tuning or playing over talk |
| WORKSHOP | 5 | The band drilling one section repeatedly |
| FALSE_START | 3 | A real attempt, abandoned, replayed later |
Timecodes are positions in the original 3:45:27 tape, so any song can be found in the master.
| # | Tape start | Length | Play | Stops | Check | File |
|---|---|---|---|---|---|---|
| 1 | 0:29:52 | 0:01:15 | 94.7% | 0 | clean | 01_tape0-29-52_len1m15s.wav |
| 2 | 0:34:28 | 0:04:07 | 93.0% | 1 | clean | 02_tape0-34-28_len4m07s.wav |
| 3 | 0:43:16 | 0:03:42 | 99.1% | 0 | clean | 03_tape0-43-16_len3m42s.wav |
| 4 | 0:47:02 | 0:03:39 | 97.7% | 1 | clean | 04_tape0-47-02_len3m39s.wav |
| 5 | 0:55:27 | 0:05:03 | 95.4% | 2 | clean | 05_tape0-55-27_len5m03s.wav |
| 6 | 1:06:14 | 0:10:25 | 97.5% | 2 | clean | 06_tape1-06-14_len10m25s.wav |
| 7 | 1:21:10 | 0:04:20 | 98.1% | 1 | clean | 07_tape1-21-10_len4m20s.wav |
| 8 | 1:26:58 | 0:04:22 | 90.1% | 3 | clean | 08_tape1-26-58_len4m22s.wav |
| 9 | 1:31:34 | 0:05:01 | 97.9% | 1 | clean | 09_tape1-31-34_len5m01s.wav |
| 10 | 1:36:56 | 0:04:39 | 99.0% | 1 | clean | 10_tape1-36-56_len4m39s.wav |
| 11 | 1:43:27 | 0:09:38 | 97.8% | 0 | clean | 11_tape1-43-27_len9m38s.wav |
| 12 | 1:53:19 | 0:01:14 | 94.0% | 0 | clean | 12_tape1-53-19_len1m14s.wav |
| 13 | 1:55:00 | 0:09:16 | 96.5% | 4 | clean | 13_tape1-55-00_len9m16s.wav |
| 14 | 2:11:22 | 0:02:27 | 82.9% | 0 | 83% play | 14_tape2-11-22_len2m27s.wav |
| 15 | 2:14:06 | 0:03:04 | 99.3% | 0 | clean | 15_tape2-14-06_len3m04s.wav |
| 16 | 2:19:27 | 0:08:19 | 95.1% | 4 | clean | 16_tape2-19-27_len8m19s.wav |
| 17 | 2:30:21 | 0:08:20 | 99.7% | 0 | clean | 17_tape2-30-21_len8m20s.wav |
| 18 | 2:38:47 | 0:20:12 | 99.1% | 0 | clean | 18_tape2-38-47_len20m12s.wav |
| 19 | 2:59:08 | 0:06:39 | 99.4% | 0 | clean | 19_tape2-59-08_len6m39s.wav |
| 20 | 3:06:46 | 0:24:42 | 99.2% | 1 | clean | 20_tape3-06-46_len24m42s.wav |
| 21 | 3:33:27 | 0:02:12 | 93.9% | 0 | clean | 21_tape3-33-27_len2m12s.wav |
| 22 | 3:35:48 | 0:03:50 | 97.7% | 1 | clean | 22_tape3-35-48_len3m50s.wav |
The speech confirmation returned nothing. All 83 takes came back UNKNOWN with zero contradictions. The room is silent between takes on this tape, so whisper had no talk to transcribe. That check gave no independent confirmation here and is not counted as a pass.
Song 14 reads 83% above the room floor and is the one file still flagged. Its gap analysis found zero stops over 1.2 seconds, so the low reading is a quiet passage rather than chatter. Kept on that evidence.
Songs 17, 18 and 19 came out of one 35.5 minute block. Two independent methods agreed on where to cut: silence gaps at 8:24 and 28:42, and a tempo analysis run on 2026-08-18 that found boundaries at 8:50 and 28:41. The cuts sit on the gap edges, so no music was clipped.
Song 20 is a single 24:42 jam and was left whole. It holds one 1.6 second gap across the entire span, so there is no evidence it contains separate songs.
Six keepers began with an aborted intro welded to the front. Each was cut to the end of the last early gap so the file starts on the real downbeat.
| Song | Trimmed | Why |
|---|---|---|
| 6 | 29.3s | Four separate stops before the take settled |
| 8 | 12.5s | Stop at 11s |
| 10 | 19.1s | Stop at 18s |
| 12 | 7.2s | Stop at 5s |
| 15 | 9.4s | Head air plus a stop at 8s |
| 17 | 4.1s | Dead air at head |