Commit
•
5e2a4f2
1
Parent(s):
e977ad8
Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,109 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
language:
|
4 |
+
- en
|
5 |
+
tags:
|
6 |
+
- hearing loss
|
7 |
+
- challenge
|
8 |
+
- signal processing
|
9 |
+
- source separation
|
10 |
+
- lyrics intelligibility
|
11 |
+
- audio
|
12 |
+
- audio-to-audio
|
13 |
+
library_name: asteroid
|
14 |
+
---
|
15 |
+
# Cadenza Challenge: CAD2-Task1
|
16 |
+
|
17 |
+
A NonCausal Lyrics/Accompaniment separation model for the CAD2-Task1 baseline system.
|
18 |
+
|
19 |
+
* Architecture: ConvTasNet (Kaituo XU) with multichannel support (Alexandre Defossez).
|
20 |
+
* Parameters:
|
21 |
+
* B: 256
|
22 |
+
* C: 2
|
23 |
+
* H: 512
|
24 |
+
* L: 20
|
25 |
+
* N: 256
|
26 |
+
* P: 3
|
27 |
+
* R: 4
|
28 |
+
* X: 10
|
29 |
+
* audio_channels: 2
|
30 |
+
* causal: false
|
31 |
+
* mask_nonlinear: relu
|
32 |
+
* norm_type: gLN
|
33 |
+
* training:
|
34 |
+
* sample_rate: 44100
|
35 |
+
* samples_per_track: 64
|
36 |
+
* segment: 5.0
|
37 |
+
* aggregate: 2
|
38 |
+
* batch_size: 4
|
39 |
+
* early_stop: true
|
40 |
+
* epochs: 200
|
41 |
+
|
42 |
+
|
43 |
+
## Dataset
|
44 |
+
The model was trained on the training split of the MUSDB18-HQ dataset.
|
45 |
+
|
46 |
+
## How to use
|
47 |
+
|
48 |
+
```
|
49 |
+
from tasnet import ConvTasNetStereo
|
50 |
+
model = ConvTasNetStereo.from_pretrained(
|
51 |
+
"cadenzachallenge/ConvTasNet_LyricsSeparation_NonCausal"
|
52 |
+
).cpu()
|
53 |
+
```
|
54 |
+
|
55 |
+
## Results
|
56 |
+
|
57 |
+
| Track | Vocals (SDR) | Accompaniment (SDR) |
|
58 |
+
|:------|:------------:|:---------:|
|
59 |
+
| Al James - Schoolboy Facination | 15.006 | 7.893 |
|
60 |
+
| AM Contra - Heart Peripheral | 21.616 | 8.014 |
|
61 |
+
| Angels In Amplifiers - I'm Alright | 18.152 | 7.742 |
|
62 |
+
| Arise - Run Run Run | 28.328 | 7.471 |
|
63 |
+
| Ben Carrigan - We'll Talk About It All Tonight | 20.789 | 5.218 |
|
64 |
+
| BKS - Bulldozer | 24.61 | 4.017 |
|
65 |
+
| BKS - Too Much | 22.342 | 9.79 |
|
66 |
+
| Bobby Nobody - Stitch Up | 20.559 | 9.151 |
|
67 |
+
| Buitraker - Revo X | 21.887 | 5.843 |
|
68 |
+
| Carlos Gonzalez - A Place For Us | 19.358 | 5.231 |
|
69 |
+
| Cristina Vane - So Easy | 17.977 | 10.211 |
|
70 |
+
| Detsky Sad - Walkie Talkie | 18.544 | 8.603 |
|
71 |
+
| Enda Reilly - Cur An Long Ag Seol | 20.317 | -0.04 |
|
72 |
+
| Forkupines - Semantics | 23.6 | 7.771 |
|
73 |
+
| Georgia Wonder - Siren | 15.564 | 5.524 |
|
74 |
+
| Girls Under Glass - We Feel Alright | 27.57 | 4.542 |
|
75 |
+
| Hollow Ground - Ill Fate | 22.336 | 9.244 |
|
76 |
+
| James Elder & Mark M Thompson - The English Actor | 19.278 | 5.524 |
|
77 |
+
| Juliet's Rescue - Heartbeats | 21.802 | 8.428 |
|
78 |
+
| Little Chicago's Finest - My Own | 6.498 | 5.774 |
|
79 |
+
| Louis Cressy Band - Good Time | 23.391 | 9.418 |
|
80 |
+
| Lyndsey Ollard - Catching Up | 20.467 | 9.685 |
|
81 |
+
| M.E.R.C. Music - Knockout | 12.641 | 9.479 |
|
82 |
+
| Moosmusic - Big Dummy Shake | 17.662 | 10.024 |
|
83 |
+
| Motor Tapes - Shore | 18.09 | 6.043 |
|
84 |
+
| Mu - Too Bright | 16.715 | 10.079 |
|
85 |
+
| Nerve 9 - Pray For The Rain | 24.028 | 7.481 |
|
86 |
+
| PR - Happy Daze | 45.118 | 0.37 |
|
87 |
+
| PR - Oh No | 9.312 | 0.459 |
|
88 |
+
| Punkdisco - Oral Hygiene | 23.343 | 7.736 |
|
89 |
+
| Raft Monk - Tiring | 17.245 | 3.306 |
|
90 |
+
| Sambasevam Shanmugam - Kaathaadi | 19.432 | 10.086 |
|
91 |
+
| Secretariat - Borderline | 22.836 | 6.729 |
|
92 |
+
| Secretariat - Over The Top | 26.406 | 8.413 |
|
93 |
+
| Side Effects Project - Sing With Me | 17.073 | 12.129 |
|
94 |
+
| Signe Jakobsen - What Have You Done To Me | 16.439 | 9.999 |
|
95 |
+
| Skelpolu - Resurrection | 27.398 | 0.665 |
|
96 |
+
| Speak Softly - Broken Man | 21.053 | 6.772 |
|
97 |
+
| Speak Softly - Like Horses | 13.648 | 7.353 |
|
98 |
+
| The Doppler Shift - Atrophy | 20.36 | 3.394 |
|
99 |
+
| The Easton Ellises - Falcon 69 | 11.352 | 6.494 |
|
100 |
+
| The Easton Ellises (Baumi) - SDRNR | 13.906 | 2.433 |
|
101 |
+
| The Long Wait - Dark Horses | 21.987 | 7.209 |
|
102 |
+
| The Mountaineering Club - Mallory | 27.447 | 12.238 |
|
103 |
+
| The Sunshine Garcia Band - For I Am The Moon | 18.684 | 11.287 |
|
104 |
+
| Timboz - Pony | 22.435 | 4.517 |
|
105 |
+
| Tom McKenzie - Directions | 26.247 | 10.277 |
|
106 |
+
| Triviul feat. The Fiend - Widow | 12.671 | 9.977 |
|
107 |
+
| We Fell From The Sky - Not You | 26.073 | 5.367 |
|
108 |
+
| Zeno - Signs | 18.11 | 7.174 |
|
109 |
+
| **Total (median over frames, median over tracks)** | **5.249** | **11.155** |
|