Commit
•
bef9f58
1
Parent(s):
da3a87d
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,9 +1,113 @@
|
|
| 1 |
---
|
|
|
|
|
|
|
|
|
|
| 2 |
tags:
|
| 3 |
-
-
|
| 4 |
-
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 5 |
---
|
|
|
|
| 6 |
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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 |
+
widget:
|
| 13 |
+
- src: >-
|
| 14 |
+
https://github.com/sigsep/open-unmix-pytorch/blob/fb672c9584997c2b05e148eeaa65b4c23ed4693b/tests/data/test.wav
|
| 15 |
+
example_title: Test
|
| 16 |
---
|
| 17 |
+
# Cadenza Challenge: CAD2-Task1
|
| 18 |
|
| 19 |
+
A Causal Lyrics/Accompaniment separation model for the CAD2-Task1 baseline system.
|
| 20 |
+
|
| 21 |
+
## Parameters
|
| 22 |
+
|
| 23 |
+
* Architecture: ConvTasNet (Kaituo XU) with multichannel support (Alexandre Defossez).
|
| 24 |
+
* Parameters:
|
| 25 |
+
* B: 256
|
| 26 |
+
* C: 2
|
| 27 |
+
* H: 512
|
| 28 |
+
* L: 20
|
| 29 |
+
* N: 256
|
| 30 |
+
* P: 3
|
| 31 |
+
* R: 4
|
| 32 |
+
* X: 10
|
| 33 |
+
* audio_channels: 2
|
| 34 |
+
* causal: true
|
| 35 |
+
* mask_nonlinear: relu
|
| 36 |
+
* norm_type: cLN
|
| 37 |
+
* training:
|
| 38 |
+
* sample_rate: 44100
|
| 39 |
+
* samples_per_track: 64
|
| 40 |
+
* segment: 4.0
|
| 41 |
+
* aggregate: 1
|
| 42 |
+
* batch_size: 4
|
| 43 |
+
* early_stop: true
|
| 44 |
+
* epochs: 200
|
| 45 |
+
|
| 46 |
+
|
| 47 |
+
## Dataset
|
| 48 |
+
The model was trained on the training split of the MUSDB18-HQ dataset.
|
| 49 |
+
|
| 50 |
+
## How to use
|
| 51 |
+
|
| 52 |
+
```
|
| 53 |
+
from tasnet import ConvTasNetStereo
|
| 54 |
+
model = ConvTasNetStereo.from_pretrained(
|
| 55 |
+
"cadenzachallenge/ConvTasNet_LyricsSeparation_Causal"
|
| 56 |
+
).cpu()
|
| 57 |
+
```
|
| 58 |
+
|
| 59 |
+
## Results
|
| 60 |
+
|
| 61 |
+
| Track | Vocals (SDR) | Accompaniment (SDR) |
|
| 62 |
+
|:------|:------------:|:---------:|
|
| 63 |
+
| Al James - Schoolboy Facination | 5.733 | 8.049 |
|
| 64 |
+
| AM Contra - Heart Peripheral | 5.887 | 12.691 |
|
| 65 |
+
| Angels In Amplifiers - I'm Alright | 5.901 | 9.124 |
|
| 66 |
+
| Arise - Run Run Run | 5.208 | 14.868 |
|
| 67 |
+
| Ben Carrigan - We'll Talk About It All Tonight | 2.676 | 9.919 |
|
| 68 |
+
| BKS - Bulldozer | 1.523 | 11.488 |
|
| 69 |
+
| BKS - Too Much | 7.005 | 11.087 |
|
| 70 |
+
| Bobby Nobody - Stitch Up | 6.518 | 11.303 |
|
| 71 |
+
| Buitraker - Revo X | 4.242 | 13.763 |
|
| 72 |
+
| Carlos Gonzalez - A Place For Us | 3.882 | 7.57 |
|
| 73 |
+
| Cristina Vane - So Easy | 7.477 | 12.126 |
|
| 74 |
+
| Detsky Sad - Walkie Talkie | 6.214 | 9.47 |
|
| 75 |
+
| Enda Reilly - Cur An Long Ag Seol | 7.329 | 11.51 |
|
| 76 |
+
| Forkupines - Semantics | 4.556 | 11.228 |
|
| 77 |
+
| Georgia Wonder - Siren | 3.165 | 7.622 |
|
| 78 |
+
| Girls Under Glass - We Feel Alright | 3.176 | 11.677 |
|
| 79 |
+
| Hollow Ground - Ill Fate | 5.67 | 14.987 |
|
| 80 |
+
| James Elder & Mark M Thompson - The English Actor | 4.014 | 8.834 |
|
| 81 |
+
| Juliet's Rescue - Heartbeats | 5.317 | 13.101 |
|
| 82 |
+
| Little Chicago's Finest - My Own | 4.409 | 5.378 |
|
| 83 |
+
| Louis Cressy Band - Good Time | 5.903 | 10.918 |
|
| 84 |
+
| Lyndsey Ollard - Catching Up | 7.812 | 10.793 |
|
| 85 |
+
| M.E.R.C. Music - Knockout | 5.663 | 7.815 |
|
| 86 |
+
| Moosmusic - Big Dummy Shake | 7.081 | 12.772 |
|
| 87 |
+
| Motor Tapes - Shore | 1.745 | 8.775 |
|
| 88 |
+
| Mu - Too Bright | 5.518 | 12.242 |
|
| 89 |
+
| Nerve 9 - Pray For The Rain | 5.685 | 11.674 |
|
| 90 |
+
| PR - Happy Daze | -2.89 | 37.274 |
|
| 91 |
+
| PR - Oh No | 0 | 8.987 |
|
| 92 |
+
| Punkdisco - Oral Hygiene | 5.044 | 16.173 |
|
| 93 |
+
| Raft Monk - Tiring | 2.119 | 8.977 |
|
| 94 |
+
| Sambasevam Shanmugam - Kaathaadi | 7.51 | 9.801 |
|
| 95 |
+
| Secretariat - Borderline | 5.068 | 9.195 |
|
| 96 |
+
| Secretariat - Over The Top | 6.278 | 13.556 |
|
| 97 |
+
| Side Effects Project - Sing With Me | 9.637 | 11.222 |
|
| 98 |
+
| Signe Jakobsen - What Have You Done To Me | 6.884 | 9.656 |
|
| 99 |
+
| Skelpolu - Resurrection | 0.053 | 8.272 |
|
| 100 |
+
| Speak Softly - Broken Man | 3.743 | 13.497 |
|
| 101 |
+
| Speak Softly - Like Horses | 4.339 | 7.233 |
|
| 102 |
+
| The Doppler Shift - Atrophy | 2.47 | 12.58 |
|
| 103 |
+
| The Easton Ellises - Falcon 69 | 2.507 | 8.137 |
|
| 104 |
+
| The Easton Ellises (Baumi) - SDRNR | 1.463 | 8.136 |
|
| 105 |
+
| The Long Wait - Dark Horses | 4.784 | 10.964 |
|
| 106 |
+
| The Mountaineering Club - Mallory | 9.015 | 13.26 |
|
| 107 |
+
| The Sunshine Garcia Band - For I Am The Moon | 8.341 | 12.1 |
|
| 108 |
+
| Timboz - Pony | 2.698 | 12.415 |
|
| 109 |
+
| Tom McKenzie - Directions | 7.305 | 15.07 |
|
| 110 |
+
| Triviul feat. The Fiend - Widow | 6.409 | 7.938 |
|
| 111 |
+
| We Fell From The Sky - Not You | 3.661 | 11.403 |
|
| 112 |
+
| Zeno - Signs | 5.291 | 10.178 |
|
| 113 |
+
| **Total (median over frames, median over tracks)** | **5.249** | **11.155** |
|