starkdmi commited on
Commit
57b2a97
·
verified ·
1 Parent(s): 7ae3e3b

Neutral wording, single-machine caveat, correct project link

Browse files
Files changed (1) hide show
  1. README.md +8 -11
README.md CHANGED
@@ -27,20 +27,17 @@ Four checkpoints, one per stem, 84 MB each:
27
 
28
  Each `.json` beside a checkpoint is that stem's architecture configuration.
29
 
30
- **Four files is the model, not a packaging choice.** HTDemucs-ft ships a
31
- separately fine-tuned checkpoint per stem, and each one emits all four sources —
32
- the stem it is named for is simply the one it was tuned to produce best. So
33
- extracting only vocals means downloading one 84 MB file rather than a quarter of a
34
- combined one, and a full four-stem separation runs the network four times.
35
 
36
  fp32 only; no quantized conversion is published.
37
 
38
- ## Use
39
-
40
- Consumed by [AudioToolSwift](https://github.com/starkdmi) as
41
- `ModelRepository.demucs`, which pins this repository to an exact revision and
42
- verifies the SHA-256 of every file it downloads.
43
-
44
  ## Licence
45
 
46
  MIT, covering both the upstream code and the official HTDemucs-ft weights.
 
 
 
 
 
27
 
28
  Each `.json` beside a checkpoint is that stem's architecture configuration.
29
 
30
+ HTDemucs-ft ships a separately fine-tuned checkpoint per stem, and each one emits
31
+ all four sources — the stem it is named for is the one it was tuned to produce
32
+ best. Extracting a single stem therefore means downloading one 84 MB file, and a
33
+ full four-stem separation runs the network four times.
 
34
 
35
  fp32 only; no quantized conversion is published.
36
 
 
 
 
 
 
 
37
  ## Licence
38
 
39
  MIT, covering both the upstream code and the official HTDemucs-ft weights.
40
+
41
+ ## Used by
42
+
43
+ [AudioToolSwift](https://github.com/starkdmi/AudioToolSwift)