loopcutter 0.3.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- loopcutter-0.3.0/LICENSE +21 -0
- loopcutter-0.3.0/PKG-INFO +664 -0
- loopcutter-0.3.0/README.md +622 -0
- loopcutter-0.3.0/pyproject.toml +55 -0
- loopcutter-0.3.0/setup.cfg +4 -0
- loopcutter-0.3.0/src/loopcutter/__init__.py +13 -0
- loopcutter-0.3.0/src/loopcutter/analyse.py +153 -0
- loopcutter-0.3.0/src/loopcutter/audio_io.py +46 -0
- loopcutter-0.3.0/src/loopcutter/cli.py +768 -0
- loopcutter-0.3.0/src/loopcutter/cut.py +197 -0
- loopcutter-0.3.0/src/loopcutter/grid.py +205 -0
- loopcutter-0.3.0/src/loopcutter/keydetect.py +77 -0
- loopcutter-0.3.0/src/loopcutter/keyreport.py +236 -0
- loopcutter-0.3.0/src/loopcutter/keys.py +346 -0
- loopcutter-0.3.0/src/loopcutter/manifest.py +245 -0
- loopcutter-0.3.0/src/loopcutter/markers.py +175 -0
- loopcutter-0.3.0/src/loopcutter/model.py +114 -0
- loopcutter-0.3.0/src/loopcutter/naming.py +95 -0
- loopcutter-0.3.0/src/loopcutter/onsets.py +117 -0
- loopcutter-0.3.0/src/loopcutter/prep.py +188 -0
- loopcutter-0.3.0/src/loopcutter/provenance.py +64 -0
- loopcutter-0.3.0/src/loopcutter/snap.py +215 -0
- loopcutter-0.3.0/src/loopcutter/stems.py +197 -0
- loopcutter-0.3.0/src/loopcutter/tags.py +135 -0
- loopcutter-0.3.0/src/loopcutter/timing.py +144 -0
- loopcutter-0.3.0/src/loopcutter/trackdb.py +77 -0
- loopcutter-0.3.0/src/loopcutter/verify.py +200 -0
- loopcutter-0.3.0/src/loopcutter/workspace.py +119 -0
- loopcutter-0.3.0/src/loopcutter.egg-info/PKG-INFO +664 -0
- loopcutter-0.3.0/src/loopcutter.egg-info/SOURCES.txt +50 -0
- loopcutter-0.3.0/src/loopcutter.egg-info/dependency_links.txt +1 -0
- loopcutter-0.3.0/src/loopcutter.egg-info/entry_points.txt +2 -0
- loopcutter-0.3.0/src/loopcutter.egg-info/requires.txt +25 -0
- loopcutter-0.3.0/src/loopcutter.egg-info/top_level.txt +1 -0
- loopcutter-0.3.0/tests/test_analyse.py +101 -0
- loopcutter-0.3.0/tests/test_cli.py +538 -0
- loopcutter-0.3.0/tests/test_cut.py +196 -0
- loopcutter-0.3.0/tests/test_grid.py +136 -0
- loopcutter-0.3.0/tests/test_keys.py +180 -0
- loopcutter-0.3.0/tests/test_markers.py +147 -0
- loopcutter-0.3.0/tests/test_model.py +26 -0
- loopcutter-0.3.0/tests/test_oneshot_xfade.py +82 -0
- loopcutter-0.3.0/tests/test_onsets.py +49 -0
- loopcutter-0.3.0/tests/test_prep.py +152 -0
- loopcutter-0.3.0/tests/test_provenance.py +174 -0
- loopcutter-0.3.0/tests/test_snap.py +215 -0
- loopcutter-0.3.0/tests/test_stems.py +234 -0
- loopcutter-0.3.0/tests/test_tags.py +52 -0
- loopcutter-0.3.0/tests/test_timing.py +94 -0
- loopcutter-0.3.0/tests/test_trackdb.py +137 -0
- loopcutter-0.3.0/tests/test_verify.py +123 -0
- loopcutter-0.3.0/tests/test_workspace.py +46 -0
loopcutter-0.3.0/LICENSE
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
1
|
+
MIT License
|
|
2
|
+
|
|
3
|
+
Copyright (c) 2026 Aaron Smiles
|
|
4
|
+
|
|
5
|
+
Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
6
|
+
of this software and associated documentation files (the "Software"), to deal
|
|
7
|
+
in the Software without restriction, including without limitation the rights
|
|
8
|
+
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
9
|
+
copies of the Software, and to permit persons to whom the Software is
|
|
10
|
+
furnished to do so, subject to the following conditions:
|
|
11
|
+
|
|
12
|
+
The above copyright notice and this permission notice shall be included in all
|
|
13
|
+
copies or substantial portions of the Software.
|
|
14
|
+
|
|
15
|
+
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
16
|
+
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
17
|
+
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
18
|
+
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
19
|
+
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
20
|
+
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
|
21
|
+
SOFTWARE.
|
|
@@ -0,0 +1,664 @@
|
|
|
1
|
+
Metadata-Version: 2.4
|
|
2
|
+
Name: loopcutter
|
|
3
|
+
Version: 0.3.0
|
|
4
|
+
Summary: Cut sample-accurate loops out of full tracks, driven by a manifest, for live loop-based sets
|
|
5
|
+
Author: Aaron Smiles
|
|
6
|
+
License-Expression: MIT
|
|
7
|
+
Project-URL: Repository, https://github.com/aaronlsmiles/loopcutter
|
|
8
|
+
Project-URL: Issues, https://github.com/aaronlsmiles/loopcutter/issues
|
|
9
|
+
Keywords: audio,loops,sampling,dj,ableton,live-performance
|
|
10
|
+
Classifier: Development Status :: 4 - Beta
|
|
11
|
+
Classifier: Environment :: Console
|
|
12
|
+
Classifier: Intended Audience :: End Users/Desktop
|
|
13
|
+
Classifier: Operating System :: MacOS
|
|
14
|
+
Classifier: Operating System :: POSIX :: Linux
|
|
15
|
+
Classifier: Programming Language :: Python :: 3
|
|
16
|
+
Classifier: Programming Language :: Python :: 3.12
|
|
17
|
+
Classifier: Topic :: Multimedia :: Sound/Audio :: Analysis
|
|
18
|
+
Classifier: Topic :: Multimedia :: Sound/Audio :: Editors
|
|
19
|
+
Requires-Python: >=3.12
|
|
20
|
+
Description-Content-Type: text/markdown
|
|
21
|
+
License-File: LICENSE
|
|
22
|
+
Requires-Dist: numpy>=1.24
|
|
23
|
+
Requires-Dist: soundfile>=0.12
|
|
24
|
+
Requires-Dist: soxr>=0.3
|
|
25
|
+
Requires-Dist: mutagen>=1.47
|
|
26
|
+
Provides-Extra: analysis
|
|
27
|
+
Requires-Dist: beat-this>=1.1; extra == "analysis"
|
|
28
|
+
Requires-Dist: librosa>=0.10; extra == "analysis"
|
|
29
|
+
Provides-Extra: markers
|
|
30
|
+
Requires-Dist: pyrekordbox>=0.4; extra == "markers"
|
|
31
|
+
Requires-Dist: serato-tools>=4.0; extra == "markers"
|
|
32
|
+
Provides-Extra: verify
|
|
33
|
+
Requires-Dist: librosa>=0.10; extra == "verify"
|
|
34
|
+
Provides-Extra: stems
|
|
35
|
+
Requires-Dist: audio-separator[cpu]>=0.47; extra == "stems"
|
|
36
|
+
Requires-Dist: audioread>=3.0; extra == "stems"
|
|
37
|
+
Provides-Extra: sheets
|
|
38
|
+
Requires-Dist: openpyxl>=3.1; extra == "sheets"
|
|
39
|
+
Provides-Extra: dev
|
|
40
|
+
Requires-Dist: pytest>=7.0; extra == "dev"
|
|
41
|
+
Dynamic: license-file
|
|
42
|
+
|
|
43
|
+
# loopcutter
|
|
44
|
+
|
|
45
|
+
Cut sample-accurate loops out of full tracks, driven by a plain-text manifest.
|
|
46
|
+
|
|
47
|
+
loopcutter is for live sets built from many short loops played at once, in the
|
|
48
|
+
spirit of Richie Hawtin's DE9 mixes. When eight loops run together for a
|
|
49
|
+
minute, a loop that is three samples too long drifts audibly. loopcutter
|
|
50
|
+
derives every loop's length from its tempo and bar count and rounds once. It
|
|
51
|
+
finds each track's beat grid, puts every start on a beat, and checks every
|
|
52
|
+
file before it calls a run a success.
|
|
53
|
+
|
|
54
|
+
The manifest is the source of truth. Keep it in version control and every
|
|
55
|
+
loop on disk can be rebuilt from one CSV file.
|
|
56
|
+
|
|
57
|
+
## Features
|
|
58
|
+
|
|
59
|
+
- **Exact lengths.** `samples = round(bars × beats_per_bar × sample_rate × 60 / bpm)`,
|
|
60
|
+
rounded once at the end, never taken from a second timecode.
|
|
61
|
+
- **A beat grid per track.** `scan` fits a constant-tempo grid to the beats
|
|
62
|
+
that [beat_this](https://github.com/CPJKU/beat_this) detects, finds which
|
|
63
|
+
beat is the one, and moves the grid onto the audio's own attacks.
|
|
64
|
+
- **Mark loops by ear.** Set memory loops in rekordbox (or cues and saved loops
|
|
65
|
+
in Serato) on a waveform, then `import` them. Each start is snapped onto the
|
|
66
|
+
grid after the DJ app's own timing offset is removed, and every shift is
|
|
67
|
+
reported.
|
|
68
|
+
- **Click-free edges.** The start moves back to the nearest zero crossing
|
|
69
|
+
within 2 ms, never forward, so the attack is never clipped. The end moves
|
|
70
|
+
with it, so the length never changes.
|
|
71
|
+
- **Roll variations.** One manifest row can produce 4, 2, 1 and ½-bar versions
|
|
72
|
+
from the same start, ready to stack in a clip launcher for beat-division rolls.
|
|
73
|
+
- **Checks that can fail.** Every file is checked for exact length, sample
|
|
74
|
+
rate, headroom against its source, silence and DC offset. The loop's own
|
|
75
|
+
audio is checked for where its attacks sit against its beats, and its tempo
|
|
76
|
+
against the analysis.
|
|
77
|
+
- **Tagged and filed.** AIFF loops carry BPM, key and stem tags (ID3v2.3) and
|
|
78
|
+
are filed by stem and tempo band, ready for a sample browser.
|
|
79
|
+
- **Traceable.** Every loop's name ends with a short provenance code
|
|
80
|
+
(`[LC-1a2b3c4d]`), its tags say which recording it came from and where, and
|
|
81
|
+
`provenance.csv` next to the loops lists them all. A loop used in a set, a
|
|
82
|
+
collected project or a sampler can be traced back to its source, which is
|
|
83
|
+
what you need when it comes to clearing samples.
|
|
84
|
+
- **Stems and one-shots.** `stems` separates a master into drums, bass,
|
|
85
|
+
vocals and more at the master's own rate, and stem rows are cut from them.
|
|
86
|
+
One-shot rows cut a phrase or stab from a start to an end.
|
|
87
|
+
- **Seamless tonal loops.** An optional pre-roll crossfade blends a loop's
|
|
88
|
+
tail into the audio just before its start, so a pad wraps without a dip.
|
|
89
|
+
- **Session key planning.** From a list of keys, works out which one to three
|
|
90
|
+
session keys cover the most loops within a transposition limit, and the
|
|
91
|
+
exact shift each loop needs.
|
|
92
|
+
|
|
93
|
+
## Status
|
|
94
|
+
|
|
95
|
+
Version 0.3. The beat grid, the DJ-app import, stems, one-shots and the
|
|
96
|
+
checks are in place, tested, and proven on a real library. See
|
|
97
|
+
[Roadmap](#roadmap) for what's next and [Known issues](#known-issues).
|
|
98
|
+
|
|
99
|
+
## Install
|
|
100
|
+
|
|
101
|
+
Needs Python 3.12 or later. Install the `loopcutter` command with
|
|
102
|
+
[pipx](https://pipx.pypa.io) (or `uv tool install`), choosing the extras you
|
|
103
|
+
need:
|
|
104
|
+
|
|
105
|
+
```bash
|
|
106
|
+
pipx install "loopcutter[analysis,markers,verify]"
|
|
107
|
+
loopcutter --help
|
|
108
|
+
```
|
|
109
|
+
|
|
110
|
+
To work on the code instead:
|
|
111
|
+
|
|
112
|
+
```bash
|
|
113
|
+
git clone https://github.com/aaronlsmiles/loopcutter.git
|
|
114
|
+
cd loopcutter
|
|
115
|
+
python3 -m venv .venv
|
|
116
|
+
source .venv/bin/activate
|
|
117
|
+
pip install -e ".[dev,analysis,markers,verify]"
|
|
118
|
+
pytest -q
|
|
119
|
+
```
|
|
120
|
+
|
|
121
|
+
Optional extras:
|
|
122
|
+
|
|
123
|
+
| Extra | Adds | For |
|
|
124
|
+
|---|---|---|
|
|
125
|
+
| `analysis` | beat_this, librosa | `scan`: beat grids, attack phase |
|
|
126
|
+
| `markers` | pyrekordbox, serato-tools | `import` and the DJ-app grid cross-check |
|
|
127
|
+
| `verify` | librosa | the `beat_alignment` and measured-tempo checks in `cut` |
|
|
128
|
+
| `sheets` | openpyxl | reading XLSX key lists |
|
|
129
|
+
| `stems` | audio-separator, onnxruntime, audioread | the `stems` command |
|
|
130
|
+
|
|
131
|
+
beat_this runs on PyTorch and uses the Apple GPU when there is one; its model
|
|
132
|
+
weights download on first use.
|
|
133
|
+
|
|
134
|
+
Audio is read with libsndfile, which handles WAV, AIFF, FLAC and MP3. `prep`
|
|
135
|
+
reads AAC, M4A and anything else through [ffmpeg](https://ffmpeg.org/), so
|
|
136
|
+
install it if your sources include those. Key cross-checks use
|
|
137
|
+
[keyfinder-cli](https://github.com/EvanPurkhiser/keyfinder-cli) when it is on
|
|
138
|
+
your `PATH`.
|
|
139
|
+
|
|
140
|
+
## Workspace
|
|
141
|
+
|
|
142
|
+
Your masters, analysis and loops are private working data, so they live in a
|
|
143
|
+
workspace outside this repository:
|
|
144
|
+
|
|
145
|
+
```bash
|
|
146
|
+
loopcutter init ~/loops-workspace
|
|
147
|
+
cd ~/loops-workspace
|
|
148
|
+
```
|
|
149
|
+
|
|
150
|
+
`init` refuses to create a workspace inside a git repository that has a
|
|
151
|
+
remote, so private audio can't end up pushed by accident. Every command finds
|
|
152
|
+
the workspace the way git finds a repository, by walking up from the current
|
|
153
|
+
directory to the folder holding `loopcutter.toml`.
|
|
154
|
+
|
|
155
|
+
| Folder | Holds |
|
|
156
|
+
|---|---|
|
|
157
|
+
| `masters/` | Lossless working copies, one per source, all at one rate. |
|
|
158
|
+
| `analysis/` | `tracks.csv` (one row per track) and `beats/`, the cached beats. |
|
|
159
|
+
| `manifests/` | Your manifests. |
|
|
160
|
+
| `reports/` | Scan reports and manifest backups. |
|
|
161
|
+
| `loops/aiff/`, `loops/wav/` | Cut loops, filed as `<stem>/<tempo band>/`. |
|
|
162
|
+
| `stems/` | Separated stems. |
|
|
163
|
+
|
|
164
|
+
`loopcutter.toml` settings:
|
|
165
|
+
|
|
166
|
+
| Key | Default | What it does |
|
|
167
|
+
|---|---|---|
|
|
168
|
+
| `[audio] sample_rate` | `48000` | The rate of every master. A cut never resamples. |
|
|
169
|
+
| `[audio] subtype` | `"PCM_24"` | The masters' sample format. |
|
|
170
|
+
| `[audio] headroom_db` | `3.0` (0 if absent) | Gain taken off every master, so overs from decoding and resampling loud sources fit instead of clipping. Changing it only affects masters made afterwards. |
|
|
171
|
+
| `[prep] replaced_dir` | `"masters/.replaced"` | Where a copy of a master built from a lossy file is kept when a lossless copy replaces it. Point it outside `masters/` if you drag that whole folder into a DJ app. |
|
|
172
|
+
| `[sources] paths` | `[]` | Folders or files that `prep` reads when given none. |
|
|
173
|
+
| `[snap] max_shift_ms` | `60` | The furthest a start may be moved onto the grid. |
|
|
174
|
+
| `[marking] app` | `"rekordbox"` | What `import` reads when `--from` isn't given: `rekordbox`, `rekordbox-xml` or `serato`. |
|
|
175
|
+
| `[marking] playlist` | `""` | The rekordbox playlist that holds your masters. |
|
|
176
|
+
|
|
177
|
+
## Quick start
|
|
178
|
+
|
|
179
|
+
1. Make masters and analyse them:
|
|
180
|
+
|
|
181
|
+
```bash
|
|
182
|
+
loopcutter prep ~/Music/sources
|
|
183
|
+
loopcutter scan
|
|
184
|
+
```
|
|
185
|
+
|
|
186
|
+
2. Mark loops in rekordbox (see [Marking loops in DJ software](#marking-loops-in-dj-software))
|
|
187
|
+
and import them, or write a manifest by hand. [manifests/example.csv](https://github.com/aaronlsmiles/loopcutter/blob/main/manifests/example.csv)
|
|
188
|
+
shows the common columns.
|
|
189
|
+
|
|
190
|
+
3. Check that every row resolves, then cut:
|
|
191
|
+
|
|
192
|
+
```bash
|
|
193
|
+
loopcutter cut manifests/session.csv --dry-run
|
|
194
|
+
loopcutter cut manifests/session.csv
|
|
195
|
+
```
|
|
196
|
+
|
|
197
|
+
4. Spot-check an output. The bar count must come back whole:
|
|
198
|
+
|
|
199
|
+
```bash
|
|
200
|
+
loopcutter inspect "loops/aiff/full/125-129/Some Artist - Some Track [A1][128][4bar][8A].aiff" --bpm 128
|
|
201
|
+
# at 128.0 BPM that is 16.0000 beats (4.0000 bars)
|
|
202
|
+
```
|
|
203
|
+
|
|
204
|
+
If it reads 3.9996 bars, something upstream is using timecode arithmetic.
|
|
205
|
+
|
|
206
|
+
Prove the pipeline on a three-row manifest first. Load those loops into your
|
|
207
|
+
DAW, set them looping and listen for a full minute, because drift and seam
|
|
208
|
+
clicks only show on repeat. Then scale up. A batch that fails on row 280 has
|
|
209
|
+
wasted a lot of time that three rows would have saved.
|
|
210
|
+
|
|
211
|
+
## The manifest
|
|
212
|
+
|
|
213
|
+
| Column | Required | Notes |
|
|
214
|
+
|---|---|---|
|
|
215
|
+
| `source` | yes | Path to the full track, normally its master. |
|
|
216
|
+
| `label` | yes | Your cue or ID reference, such as `A1` or `16.2`. |
|
|
217
|
+
| `bars` | yes | Loop length in bars. |
|
|
218
|
+
| `bpm` | yes | The track's tempo. |
|
|
219
|
+
| `start` | one of | `M:SS.mmm`, `H:MM:SS.mmm` or bare seconds. |
|
|
220
|
+
| `downbeat` + `start_bar` | one of | A precise downbeat time plus a 1-based bar number counted from it. |
|
|
221
|
+
| `kind` | no | `loop` (the default) or `oneshot`. |
|
|
222
|
+
| `end` | one-shots | Where a one-shot ends. A one-shot needs `source`, `label`, `start` and `end`, not `bars` or `bpm`. |
|
|
223
|
+
| `xfade_ms` | no | Pre-roll crossfade length for this loop; overrides `--xfade-ms`. |
|
|
224
|
+
| `track_id` | no | The track's row in `tracks.csv`. `resolve` fills `source`, `bpm` and `key` from it. |
|
|
225
|
+
| `snap` | no | `beat` or `bar` asks `resolve` to move `start` onto the grid. Must be empty before `cut`. |
|
|
226
|
+
| `variations` | no | Extra bar lengths from the same start, such as `"2,1,0.5"`. |
|
|
227
|
+
| `beats_per_bar` | no | Defaults to 4. |
|
|
228
|
+
| `artist`, `track` | no | Used to build the output filename. |
|
|
229
|
+
| `stem` | no | Written into the filename and the tags, and picks the library folder. |
|
|
230
|
+
| `key` | no | Written into the filename and the tags, and read by the key planner. |
|
|
231
|
+
| `notes` | no | Ignored by the tool. For you. |
|
|
232
|
+
|
|
233
|
+
Give either `start` or `downbeat` plus `start_bar`, never both. The loader
|
|
234
|
+
rejects rows that give both, rows with no start, rows that still say
|
|
235
|
+
`snap=beat` or `snap=bar`, and manifests whose rows would produce the same
|
|
236
|
+
filename, and it names the row number every time. Relative `source` paths
|
|
237
|
+
resolve against the current directory. Use `--audio-root` to point somewhere
|
|
238
|
+
else.
|
|
239
|
+
|
|
240
|
+
For rows you type by hand, give a `track_id`, a rough `start` and
|
|
241
|
+
`snap=beat`, then run `loopcutter resolve manifests/set.csv`. It backs the
|
|
242
|
+
manifest up to `reports/`, fills the missing columns from `tracks.csv`, snaps
|
|
243
|
+
each start and reports how far it moved. A start further than `max_shift_ms`
|
|
244
|
+
from the nearest beat is refused, and the row is named along with the beats
|
|
245
|
+
and bar either side of it.
|
|
246
|
+
|
|
247
|
+
## Getting the start right
|
|
248
|
+
|
|
249
|
+
`cut` cuts exactly where the manifest says. It never moves a start by more
|
|
250
|
+
than the 2 ms zero-crossing search, and it never runs a model. Placement
|
|
251
|
+
happens earlier, in `import` and `resolve`, where every move is reported.
|
|
252
|
+
|
|
253
|
+
- **Mark on the master.** DJ apps decode MP3s with different padding, so a cue
|
|
254
|
+
read off an MP3 can sit tens of milliseconds from the same moment in the
|
|
255
|
+
audio this tool decodes. Marking on the lossless masters that `prep` makes
|
|
256
|
+
removes that offset. For marks made on the original files, `scan --app`
|
|
257
|
+
measures the app's offset per track and `import` removes it.
|
|
258
|
+
- **Give milliseconds.** A time typed to the whole second lands anywhere up to
|
|
259
|
+
half a second from the downbeat, which at 125 BPM is the wrong beat about
|
|
260
|
+
half the time.
|
|
261
|
+
- **Get the tempo exact.** A 4-bar loop cut at 134 BPM from a 136 BPM track is
|
|
262
|
+
105 ms too long on every repeat. `scan` measures each track's tempo; the
|
|
263
|
+
`bpm_match` check catches a manifest that disagrees.
|
|
264
|
+
|
|
265
|
+
## Marking loops in DJ software
|
|
266
|
+
|
|
267
|
+
1. **`loopcutter prep`** makes 24-bit masters at the workspace rate from your
|
|
268
|
+
sources. Existing masters are not overwritten, because Mixed In Key or a
|
|
269
|
+
DJ app may have written tags to them; the one exception is below. A changed
|
|
270
|
+
source is reported instead; delete the master to rebuild it.
|
|
271
|
+
2. **Import the masters into rekordbox** as one playlist, name it in
|
|
272
|
+
`loopcutter.toml` under `[marking] playlist`, and fix any grid whose first
|
|
273
|
+
beat is wrong.
|
|
274
|
+
3. **Mark memory loops** with quantise on. Hot cues are ignored, so they stay
|
|
275
|
+
free for performing. Comment each loop with its label, then its stem, then
|
|
276
|
+
any roll lengths: `A1 bass 2,1` is loop A1, cut from the bass stem, with
|
|
277
|
+
extra 2 and 1-bar versions. A memory cue with no loop gets the default
|
|
278
|
+
length of 4 bars. Close rekordbox before importing.
|
|
279
|
+
4. **`loopcutter scan --app rekordbox`** analyses each master and compares the
|
|
280
|
+
grid with rekordbox's.
|
|
281
|
+
5. **`loopcutter import --from rekordbox --playlist NAME`** reads the marks,
|
|
282
|
+
snaps them and writes a resolved manifest to `manifests/`, reporting the
|
|
283
|
+
median and range of the moves. Serato users run `import --from serato`;
|
|
284
|
+
`--from rekordbox-xml --xml FILE` reads a rekordbox XML export instead of
|
|
285
|
+
the database. Marks that can't be placed are skipped and named, and the
|
|
286
|
+
command exits 1 so they can't be missed.
|
|
287
|
+
6. **Review the manifest.** Every row's `notes` say where the mark was and how
|
|
288
|
+
far it moved.
|
|
289
|
+
7. **`loopcutter cut`** the manifest.
|
|
290
|
+
|
|
291
|
+
## Commands
|
|
292
|
+
|
|
293
|
+
### `init`
|
|
294
|
+
|
|
295
|
+
```bash
|
|
296
|
+
loopcutter init [DIR]
|
|
297
|
+
```
|
|
298
|
+
|
|
299
|
+
Creates a workspace with its folders and a default `loopcutter.toml`. Running
|
|
300
|
+
it again keeps your edits.
|
|
301
|
+
|
|
302
|
+
### `prep`
|
|
303
|
+
|
|
304
|
+
```bash
|
|
305
|
+
loopcutter prep [FILES OR FOLDERS ...]
|
|
306
|
+
```
|
|
307
|
+
|
|
308
|
+
Decodes each source once, resamples it once, takes `headroom_db` off, and
|
|
309
|
+
writes a 24-bit AIFF master, adding a row to `tracks.csv`. Folders are
|
|
310
|
+
searched recursively, skipping hidden folders and the workspace's own masters.
|
|
311
|
+
Masters are named after the source file, so a lossless
|
|
312
|
+
file and a lossy copy with the same name make the same master: the lossless
|
|
313
|
+
file wins, and the lossy one is skipped with a note (names are
|
|
314
|
+
compared as a case-insensitive filesystem sees them, an `.m4a` counts as
|
|
315
|
+
lossless only if it holds ALAC, and the two must be the same length to within
|
|
316
|
+
0.25 s). A master already built from a lossy file is rebuilt when a lossless
|
|
317
|
+
copy of the same track appears; a different track that shares a name is
|
|
318
|
+
reported as CHANGED and left alone. The rebuild happens beside the masters
|
|
319
|
+
and swaps in under the old master's name only when it's complete: a copy of
|
|
320
|
+
the old master is kept in `replaced_dir`, its tags (except DJ-app cue data)
|
|
321
|
+
are copied onto the new one, your `override_*` columns are kept, and the rest
|
|
322
|
+
of the track's analysis is left for `scan`. Re-analyse the track in your DJ
|
|
323
|
+
app too, because its cues were placed on the old audio. Any other pair of sources that would
|
|
324
|
+
make the same master is refused before anything is written. Loud MP3s
|
|
325
|
+
decode above full scale; the headroom keeps those peaks, and any sample still
|
|
326
|
+
over is clipped and counted.
|
|
327
|
+
|
|
328
|
+
### `scan`
|
|
329
|
+
|
|
330
|
+
```bash
|
|
331
|
+
loopcutter scan [TRACK_ID ...] [--app serato|rekordbox] [--force]
|
|
332
|
+
```
|
|
333
|
+
|
|
334
|
+
Analyses each master once: tempo, bar phase, attack phase, key, and with
|
|
335
|
+
`--app` the DJ app's tempo and offset. It writes them to `tracks.csv` with a
|
|
336
|
+
flag for anything doubtful:
|
|
337
|
+
|
|
338
|
+
| Flag | Means |
|
|
339
|
+
|---|---|
|
|
340
|
+
| `grid-fit` | Under 90% of the detected beats fit one steady grid: a tempo change, a long break or a swung rhythm. Snapping uses the whole-track grid wherever it still fits the beats around a mark, and a local grid only where one fits clearly better (a real tempo change). |
|
|
341
|
+
| `bar-phase` | The downbeats disagree about which beat is the one. |
|
|
342
|
+
| `phase` | The attacks don't agree on where the beat sits. Only rekordbox marks made on the master are kept as they are. |
|
|
343
|
+
| `key` | The key tag and keyfinder disagree. |
|
|
344
|
+
| `app-bpm` | The DJ app's tempo differs by more than 0.05 BPM. |
|
|
345
|
+
| `half-beat` | The DJ app's grid sits half a beat from ours. |
|
|
346
|
+
| `app-offset` | On a lossless file, the app's grid sits more than 20 ms from ours. |
|
|
347
|
+
|
|
348
|
+
`override_bpm` and `override_phase_ms` in `tracks.csv` are yours: `scan` never
|
|
349
|
+
touches them, and snapping uses them when set. Tracks already analysed are
|
|
350
|
+
skipped unless you pass `--force`. A report of each run goes in `reports/`.
|
|
351
|
+
|
|
352
|
+
### `import`
|
|
353
|
+
|
|
354
|
+
```bash
|
|
355
|
+
loopcutter import [--from rekordbox|rekordbox-xml|serato] [FILES ...] [--playlist NAME] [--xml FILE] [--out FILE] [--max-shift-ms N]
|
|
356
|
+
```
|
|
357
|
+
|
|
358
|
+
Turns marked loops into a resolved manifest. `--from` defaults to `[marking]
|
|
359
|
+
app`. An existing file is never overwritten.
|
|
360
|
+
|
|
361
|
+
### `resolve`
|
|
362
|
+
|
|
363
|
+
```bash
|
|
364
|
+
loopcutter resolve MANIFEST [--max-shift-ms N]
|
|
365
|
+
```
|
|
366
|
+
|
|
367
|
+
Fills rows from `tracks.csv` and snaps starts marked `snap=beat` or
|
|
368
|
+
`snap=bar`, after backing the manifest up.
|
|
369
|
+
|
|
370
|
+
### `stems`
|
|
371
|
+
|
|
372
|
+
```bash
|
|
373
|
+
loopcutter stems TRACK_ID ... [--engine audio-separator|demucs] [--model NAME]
|
|
374
|
+
```
|
|
375
|
+
|
|
376
|
+
Separates each master with [audio-separator](https://github.com/nomadkaraoke/python-audio-separator)
|
|
377
|
+
(Demucs `htdemucs_6s` by default: drums, bass, vocals, guitar, piano and
|
|
378
|
+
other), or with the Demucs command line as a fallback. Separators work at
|
|
379
|
+
44.1 kHz, so each stem is resampled to the master's rate and written to
|
|
380
|
+
`stems/<track_id>/<stem>.aiff`, and the stems' sum is proven to line up with
|
|
381
|
+
the master to the sample before any loop is cut from them. Raw output is
|
|
382
|
+
cached, so a second run is instant.
|
|
383
|
+
|
|
384
|
+
### `cut`
|
|
385
|
+
|
|
386
|
+
```bash
|
|
387
|
+
loopcutter cut MANIFEST [options]
|
|
388
|
+
```
|
|
389
|
+
|
|
390
|
+
| Option | Default | What it does |
|
|
391
|
+
|---|---|---|
|
|
392
|
+
| `--out DIR` | `loops/<format>` in a workspace, else `loops` | Output directory. |
|
|
393
|
+
| `--layout` | `library` in a workspace, else `flat` | `library` files loops as `<stem>/<tempo band>/`. |
|
|
394
|
+
| `--tag`, `--no-tag` | on for AIFF in a workspace | Writes BPM, key and stem tags. |
|
|
395
|
+
| `--tracks FILE` | the workspace's `tracks.csv` | Analysis used for the tempo check. |
|
|
396
|
+
| `--audio-root DIR` | current directory | Base for relative `source` paths. |
|
|
397
|
+
| `--format` | `aiff` | `aiff`, `wav` or `flac`. |
|
|
398
|
+
| `--subtype` | `PCM_24` | Any libsndfile subtype. |
|
|
399
|
+
| `--snap-ms` | `2.0` | Zero-crossing search radius (backwards only). |
|
|
400
|
+
| `--fade-ms` | `0.5` | Edge fade length. |
|
|
401
|
+
| `--trim-db` | `0.0` | Gain for every loop, for example `-1.0`. |
|
|
402
|
+
| `--xfade-ms` | `0.0` | Crossfades each loop's tail into the audio before its start (a row's `xfade_ms` wins). |
|
|
403
|
+
| `--check-bpm` | off | Measures the tempo from the audio. Automatic for analysed tracks. |
|
|
404
|
+
| `--lenient` | off | Reports `beat_alignment` failures as warnings, for unsteady material. |
|
|
405
|
+
| `--dry-run` | off | Resolves every row without writing audio. |
|
|
406
|
+
| `--code`, `--no-code` | on in a workspace | Ends each name with its provenance code. |
|
|
407
|
+
|
|
408
|
+
32-bit float isn't the default because many CDJs won't play it.
|
|
409
|
+
|
|
410
|
+
**Provenance.** The code is `LC-` plus 8 hex of a hash of what the loop is:
|
|
411
|
+
track, label, length, stem and kind, but not the start, so re-marking a loop
|
|
412
|
+
keeps its code. Tagged loops also carry `TXXX:PROVENANCE_CODE` and
|
|
413
|
+
`TXXX:SOURCE` (artist and track, the source file, the start, bars at BPM, stem
|
|
414
|
+
and label), and every cut updates `provenance.csv` in the output folder, one
|
|
415
|
+
row per code. Re-cutting a row that was cut before codes existed replaces the
|
|
416
|
+
old, uncoded file.
|
|
417
|
+
|
|
418
|
+
### `provenance`
|
|
419
|
+
|
|
420
|
+
```bash
|
|
421
|
+
loopcutter provenance MANIFEST [MANIFEST ...] [--keep-names] [--dry-run]
|
|
422
|
+
```
|
|
423
|
+
|
|
424
|
+
Adds codes, `SOURCE` tags and `provenance.csv` rows to loops that are already
|
|
425
|
+
cut, without cutting them again. By default it renames each file to its coded
|
|
426
|
+
name; `--keep-names` tags and indexes only. Renaming moves files that a set or
|
|
427
|
+
sampler project may point at, so run it before the loops are in use, or
|
|
428
|
+
re-point the set afterwards.
|
|
429
|
+
|
|
430
|
+
### `inspect`
|
|
431
|
+
|
|
432
|
+
```bash
|
|
433
|
+
loopcutter inspect FILE [--bpm N]
|
|
434
|
+
```
|
|
435
|
+
|
|
436
|
+
Prints sample rate, channels, format, length, and the length in beats and bars
|
|
437
|
+
at the given tempo.
|
|
438
|
+
|
|
439
|
+
### `keys`
|
|
440
|
+
|
|
441
|
+
See [Session key planning](#session-key-planning).
|
|
442
|
+
|
|
443
|
+
## Output files
|
|
444
|
+
|
|
445
|
+
Files are named `Artist - Track [label][bpm][bars][key][stem].aiff`, for
|
|
446
|
+
example `Some Artist - Some Track [A1][128][4bar][8A][bass].aiff`. The brackets
|
|
447
|
+
make a library easy to sort and search in any browser. In a workspace they are
|
|
448
|
+
filed by stem and tempo band, such as `loops/aiff/bass/125-129/`; loops without
|
|
449
|
+
a stem go under `full/`, and one-shots under `<stem>/oneshots/`, named
|
|
450
|
+
`Artist - Track [label][oneshot].aiff`.
|
|
451
|
+
|
|
452
|
+
A row that names a `stem` and a `track_id` is cut from that track's separated
|
|
453
|
+
stem, while its timing is still checked on the full mix, where the attacks
|
|
454
|
+
are. `cut` names any stem that hasn't been separated yet. A `stem` row without
|
|
455
|
+
a `track_id` is cut from its `source` as given, with a note, so point `source`
|
|
456
|
+
at the stem file if you separated it yourself.
|
|
457
|
+
|
|
458
|
+
AIFF loops are tagged with BPM, key, stem (as the grouping), title, artist and
|
|
459
|
+
a comment holding the Camelot key and label. WAV loops are left untagged by
|
|
460
|
+
default, because some hardware samplers reject the tag chunk; pass `--tag` to
|
|
461
|
+
tag them anyway.
|
|
462
|
+
|
|
463
|
+
Ableton Live doesn't read the BPM in a filename. It assumes a clip is 1, 2, 4,
|
|
464
|
+
8 or 16 bars long and sets the clip's tempo from its length, so loops of those
|
|
465
|
+
lengths land on tempo because the lengths are exact. For any other length,
|
|
466
|
+
such as ½-bar rolls or 3-bar loops, `cut` prints a note: set that clip's tempo
|
|
467
|
+
by hand.
|
|
468
|
+
|
|
469
|
+
To make two copies of the same cuts, run the manifest twice: once as AIFF for
|
|
470
|
+
your DAW, and once with `--format wav` for a hardware sampler.
|
|
471
|
+
|
|
472
|
+
### Roll variations
|
|
473
|
+
|
|
474
|
+
A `variations` cell such as `"2,1,0.5"` turns one row into several loops that
|
|
475
|
+
share a start. Put them in one clip-launcher column, for example Ableton's
|
|
476
|
+
Session view played from a Launchpad. Set the clips to Legato launch so
|
|
477
|
+
switching keeps phase, and pressing down the column gives a roll that stays
|
|
478
|
+
locked. Each one is a real file cut to an exact sample count.
|
|
479
|
+
|
|
480
|
+
### Sample rates and tiling
|
|
481
|
+
|
|
482
|
+
Sample counts are whole numbers, so a sub-bar variation doesn't always divide
|
|
483
|
+
its parent exactly. Whether it does depends on the tempo as well as the rate:
|
|
484
|
+
|
|
485
|
+
| Rate | Common tempos where half bars tile exactly |
|
|
486
|
+
|---|---|
|
|
487
|
+
| 48 kHz | 120, 125, 128, 144, 150, 160 BPM |
|
|
488
|
+
| 44.1 kHz | 120, 125, 126, 135, 140, 144, 147, 150, 160 BPM |
|
|
489
|
+
|
|
490
|
+
Neither rate is always better. At other tempos a variation drifts by a few
|
|
491
|
+
samples per cycle: up to 4 for a half bar against 4 bars, and up to 8 against
|
|
492
|
+
8 or 16 bars, under 0.2 ms. `cut` reports each drifting variation set
|
|
493
|
+
and, when the other rate would be exact, names it. DAWs that warp clips
|
|
494
|
+
re-sync them every cycle anyway. loopcutter never resamples at cut time, so
|
|
495
|
+
the rate is chosen once, when `prep` makes the masters.
|
|
496
|
+
|
|
497
|
+
## Verification
|
|
498
|
+
|
|
499
|
+
An agent can't hear the output, and neither can a batch run, so every file is
|
|
500
|
+
checked before a run reports success. Every check has a test proving it can
|
|
501
|
+
fail.
|
|
502
|
+
|
|
503
|
+
- **`sample_count`, `sample_rate`:** the file is exactly the declared length at
|
|
504
|
+
the source's rate.
|
|
505
|
+
- **`peak_headroom`:** the loop is no louder than its source, after any
|
|
506
|
+
`--trim-db`. A master that already peaks at full scale passes; a trim that
|
|
507
|
+
would push a loop past full scale fails.
|
|
508
|
+
- **`not_silent`, `dc_offset`.**
|
|
509
|
+
- **`beat_alignment`:** finds the strongest attack within 60 ms of each of the
|
|
510
|
+
loop's beats, in the source around the loop (the full mix, for a stem).
|
|
511
|
+
- *Placement* is where those attacks sit. It passes from −3 to +8 ms, warns
|
|
512
|
+
out to −10 or +15 ms, and fails beyond. An attack before the start means
|
|
513
|
+
the start cuts into it.
|
|
514
|
+
- *Drift* is how the attacks move from the loop's first beat to its last,
|
|
515
|
+
which is what a wrong tempo looks like. It warns above 4 ms and fails
|
|
516
|
+
above 10 ms.
|
|
517
|
+
- When under half the beats share a steady attack (breaks, pads, swung
|
|
518
|
+
material), it reports *not judged* rather than guessing.
|
|
519
|
+
- `--lenient` turns its failures into warnings.
|
|
520
|
+
- One-shots have no beats to align and no tempo to match, so they get the
|
|
521
|
+
length, rate, headroom, silence and DC checks only.
|
|
522
|
+
- **`bpm_match`:** how far the loop's end lands from where the next bar
|
|
523
|
+
begins. Against the tempo `scan` proposed it fails above 2 ms. On a
|
|
524
|
+
`grid-fit` track, and with `--check-bpm` on an unanalysed one, it measures
|
|
525
|
+
the tempo from the audio instead and fails above 5 ms; on a `grid-fit`
|
|
526
|
+
track a half- or double-time tempo fails outright.
|
|
527
|
+
|
|
528
|
+
A failed check names the file and the reason, and the run exits with an error.
|
|
529
|
+
|
|
530
|
+
## Session key planning
|
|
531
|
+
|
|
532
|
+
Key detection is unreliable on short loops. Analyse the full source tracks
|
|
533
|
+
instead: `scan` reads the key your DJ software or Mixed In Key wrote to each
|
|
534
|
+
master, cross-checks it with keyfinder-cli, and carries it into the manifest's
|
|
535
|
+
`key` column. Every slice inherits its parent's key.
|
|
536
|
+
|
|
537
|
+
Then work out what to build the set around:
|
|
538
|
+
|
|
539
|
+
```bash
|
|
540
|
+
loopcutter keys library.xlsx --histogram
|
|
541
|
+
loopcutter keys library.xlsx --keys 3 --semitones 3 -v
|
|
542
|
+
loopcutter keys library.xlsx --sweep
|
|
543
|
+
```
|
|
544
|
+
|
|
545
|
+
`keys` reads CSV, XLSX or a manifest, using a column called `key`, `camelot`
|
|
546
|
+
or `initial key`, so exports from Mixed In Key, Serato and rekordbox work as
|
|
547
|
+
they are. It accepts Camelot (`8A`) and named keys (`Am`, `F# minor`, `Bb`).
|
|
548
|
+
It rejects Open Key rather than misreading it, because Open Key numbers sit
|
|
549
|
+
five places away from Camelot's.
|
|
550
|
+
|
|
551
|
+
The model is transposition, not harmonic mixing: a loop is usable in a session
|
|
552
|
+
key if you can pitch it there within the limit. It picks the key that covers
|
|
553
|
+
the most loops, then the best key for what's left, and reports the exact
|
|
554
|
+
semitone shift for each loop.
|
|
555
|
+
|
|
556
|
+
| Flag | Default | What it does |
|
|
557
|
+
|---|---|---|
|
|
558
|
+
| `--keys N` | 3 | How many session keys to plan for. |
|
|
559
|
+
| `--semitones N` | 3 | Transposition limit either way, 0 to 6. |
|
|
560
|
+
| `--comfort N` | 3 | Shifts beyond this are flagged for audition in the verbose listing. |
|
|
561
|
+
| `--sweep` | off | Tabulates coverage at every limit from 0 to 6. |
|
|
562
|
+
| `--relative` | off | Counts relative major/minor as a free match. |
|
|
563
|
+
| `--histogram` | off | Prints the key distribution as a bar chart. |
|
|
564
|
+
| `--sheet NAME` | first sheet | Worksheet to read in a multi-sheet XLSX. |
|
|
565
|
+
| `-v`, `--verbose` | off | Lists every loop with its Camelot key and shift. |
|
|
566
|
+
|
|
567
|
+
Read the sweep's "keys to cover all" column first. Relaxing the limit by one
|
|
568
|
+
semitone can drop you from three session keys to two, which changes how a set
|
|
569
|
+
is built far more than a few percent of coverage does. The best session key
|
|
570
|
+
is often not the most common one: a key between two clusters can reach more
|
|
571
|
+
material than the biggest cluster.
|
|
572
|
+
|
|
573
|
+
## Stems and seamless loops
|
|
574
|
+
|
|
575
|
+
One 4-bar window across six stems gives six loops you can EQ and launch apart:
|
|
576
|
+
run `loopcutter stems "<track_id>"` for the track, then give the rows a
|
|
577
|
+
`stem`. Comment a rekordbox memory loop `A1 bass` and `import` fills it in.
|
|
578
|
+
|
|
579
|
+
Edge fades stop a loop clicking, but on a sustained pad they dip to silence
|
|
580
|
+
at every wrap. `--xfade-ms 20` (or a row's `xfade_ms`) instead fades the
|
|
581
|
+
loop's tail into the audio just before its start, so the wrap plays exactly
|
|
582
|
+
what the source played there. The length doesn't change, and because the two
|
|
583
|
+
sides are similar audio, the linear blend never exceeds the source's peak.
|
|
584
|
+
|
|
585
|
+
## Known issues
|
|
586
|
+
|
|
587
|
+
- **rekordbox's database format changes between versions.** If
|
|
588
|
+
`import --from rekordbox` can't read yours, export the collection as XML
|
|
589
|
+
from rekordbox and use `--from rekordbox-xml --xml FILE`.
|
|
590
|
+
- **Tempo-changing tracks get local grids.** A track whose beats don't fit one
|
|
591
|
+
steady grid is flagged `grid-fit`. Each mark is snapped to the whole-track
|
|
592
|
+
grid where that still fits the nearby beats, and otherwise to a grid fitted
|
|
593
|
+
to the 64 beats around it, provided that grid fits clearly better and its
|
|
594
|
+
tempo isn't a half- or double-time reading. A mark where neither holds is
|
|
595
|
+
refused. On a tempo-changing track the `tracks.csv` tempo is only an
|
|
596
|
+
average, so `cut` checks those loops against the tempo measured from the
|
|
597
|
+
audio.
|
|
598
|
+
- **Odd bar lengths need their tempo set in Live by hand** (see
|
|
599
|
+
[Output files](#output-files)).
|
|
600
|
+
- **Starts in beatless passages can't be judged.** `beat_alignment` reports
|
|
601
|
+
*not judged* there, so a hand-typed start in a breakdown relies on the grid
|
|
602
|
+
alone.
|
|
603
|
+
|
|
604
|
+
## Roadmap
|
|
605
|
+
|
|
606
|
+
- An audition page: each master's waveform with the cutter's own bar lines,
|
|
607
|
+
where a click previews the exact loop and a key accepts it into a manifest.
|
|
608
|
+
- Ableton integration: the right tempo for odd-length and half-bar clips, and
|
|
609
|
+
a generated Live Set with each loop family in one Launchpad column.
|
|
610
|
+
|
|
611
|
+
## Development
|
|
612
|
+
|
|
613
|
+
```bash
|
|
614
|
+
pytest -q # unit tests: synthetic audio only, no model weights
|
|
615
|
+
pytest -q -m slow # the real beat detector and ffmpeg
|
|
616
|
+
```
|
|
617
|
+
|
|
618
|
+
```
|
|
619
|
+
src/loopcutter/
|
|
620
|
+
model.py shared data types: Grid, TrackRecord, Marker, SnapResult
|
|
621
|
+
onsets.py the one onset detector that the grid, the analysis and the checks share
|
|
622
|
+
timing.py sample arithmetic, no file I/O
|
|
623
|
+
workspace.py finding and creating workspaces
|
|
624
|
+
audio_io.py reading any audio file (libsndfile, else ffmpeg)
|
|
625
|
+
prep.py lossless working masters
|
|
626
|
+
grid.py beat-grid fit, bar phase, attack phase, local grids
|
|
627
|
+
keydetect.py key from tags, cross-checked with keyfinder-cli
|
|
628
|
+
trackdb.py tracks.csv
|
|
629
|
+
analyse.py scan: one analysis per master
|
|
630
|
+
markers.py reading marks and grids from rekordbox and Serato
|
|
631
|
+
snap.py app-offset correction, snapping, marks to manifest rows
|
|
632
|
+
manifest.py CSV to LoopSpec; all validation lives here
|
|
633
|
+
cut.py extraction, zero-crossing snap, edge fades
|
|
634
|
+
verify.py objective checks on every output
|
|
635
|
+
tags.py ID3v2.3 tags on AIFF and WAV loops
|
|
636
|
+
naming.py output filename convention and library layout
|
|
637
|
+
keys.py key parsing and session-cover arithmetic, no file I/O
|
|
638
|
+
keyreport.py spreadsheet loading and report rendering
|
|
639
|
+
stems.py separation, resampling to the master, alignment proof
|
|
640
|
+
cli.py command-line entry point
|
|
641
|
+
```
|
|
642
|
+
|
|
643
|
+
The code keeps a few rules:
|
|
644
|
+
|
|
645
|
+
- Loop length always comes from tempo and bar count, rounded once.
|
|
646
|
+
`tests/test_timing.py` guards this against a well-meant "simplification".
|
|
647
|
+
- Analysis proposes and `cut` disposes: `scan` writes `tracks.csv`, only
|
|
648
|
+
`import` and `resolve` move a start (reporting every move), and `cut` never
|
|
649
|
+
moves a start or runs a model.
|
|
650
|
+
- Shared types go in `model.py`, and every part that looks for attacks uses
|
|
651
|
+
`onsets.py`, so the grid and the checks can't disagree about where a beat is.
|
|
652
|
+
- `timing.py` and `keys.py` do arithmetic only, so they stay testable without
|
|
653
|
+
audio fixtures.
|
|
654
|
+
- Failures are loud and name the manifest row. The tool never writes a short,
|
|
655
|
+
silent or wrong-length file.
|
|
656
|
+
- Nothing is fixed silently: no trimming to fit, no resampling at cut time, no
|
|
657
|
+
padding.
|
|
658
|
+
|
|
659
|
+
## Licence
|
|
660
|
+
|
|
661
|
+
MIT. See [LICENSE](https://github.com/aaronlsmiles/loopcutter/blob/main/LICENSE).
|
|
662
|
+
|
|
663
|
+
Loops cut from commercial releases are for your own practice and performance.
|
|
664
|
+
Releasing anything built from them needs the rights holders' clearance.
|