README.md

---
license: apache-2.0
language:
- en
tags:
- hearing loss
- challenge
- signal processing
- source separation
- lyrics intelligibility
- audio
- audio-to-audio
library_name: asteroid
---
# Cadenza Challenge: CAD2-Task1

A NonCausal Lyrics/Accompaniment separation model for the CAD2-Task1 baseline system.

* Architecture: ConvTasNet (Kaituo XU) with multichannel support (Alexandre Defossez).
* Parameters:
  * B: 256
  * C: 2
  * H: 512
  * L: 20
  * N: 256
  * P: 3
  * R: 4
  * X: 10
  * audio_channels: 2
  * causal: false
  * mask_nonlinear: relu
  * norm_type: gLN
* training:
  * sample_rate: 44100
  * samples_per_track: 64
  * segment: 5.0
  * aggregate: 2
  * batch_size: 4
  * early_stop: true
  * epochs: 200


## Dataset
The model was trained on the training split of the MUSDB18-HQ dataset.

## How to use

```
from tasnet import ConvTasNetStereo
model = ConvTasNetStereo.from_pretrained(
    "cadenzachallenge/ConvTasNet_LyricsSeparation_NonCausal"
).cpu()
```

## Results 

| Track | Vocals (SDR) | Accompaniment (SDR) |
|:------|:------------:|:---------:|
| Al James - Schoolboy Facination | 15.006 | 7.893 | 
| AM Contra - Heart Peripheral | 21.616 | 8.014 | 
| Angels In Amplifiers - I'm Alright | 18.152 | 7.742 | 
| Arise - Run Run Run | 28.328 | 7.471 | 
| Ben Carrigan - We'll Talk About It All Tonight | 20.789 | 5.218 | 
| BKS - Bulldozer | 24.61 | 4.017 | 
| BKS - Too Much | 22.342 | 9.79 | 
| Bobby Nobody - Stitch Up | 20.559 | 9.151 | 
| Buitraker - Revo X | 21.887 | 5.843 | 
| Carlos Gonzalez - A Place For Us | 19.358 | 5.231 | 
| Cristina Vane - So Easy | 17.977 | 10.211 | 
| Detsky Sad - Walkie Talkie | 18.544 | 8.603 | 
| Enda Reilly - Cur An Long Ag Seol | 20.317 | -0.04 | 
| Forkupines - Semantics | 23.6 | 7.771 | 
| Georgia Wonder - Siren | 15.564 | 5.524 | 
| Girls Under Glass - We Feel Alright | 27.57 | 4.542 | 
| Hollow Ground - Ill Fate | 22.336 | 9.244 | 
| James Elder & Mark M Thompson - The English Actor | 19.278 | 5.524 | 
| Juliet's Rescue - Heartbeats | 21.802 | 8.428 | 
| Little Chicago's Finest - My Own | 6.498 | 5.774 | 
| Louis Cressy Band - Good Time | 23.391 | 9.418 | 
| Lyndsey Ollard - Catching Up | 20.467 | 9.685 | 
| M.E.R.C. Music - Knockout | 12.641 | 9.479 | 
| Moosmusic - Big Dummy Shake | 17.662 | 10.024 | 
| Motor Tapes - Shore | 18.09 | 6.043 | 
| Mu - Too Bright | 16.715 | 10.079 | 
| Nerve 9 - Pray For The Rain | 24.028 | 7.481 | 
| PR - Happy Daze | 45.118 | 0.37 | 
| PR - Oh No | 9.312 | 0.459 | 
| Punkdisco - Oral Hygiene | 23.343 | 7.736 | 
| Raft Monk - Tiring | 17.245 | 3.306 | 
| Sambasevam Shanmugam - Kaathaadi | 19.432 | 10.086 | 
| Secretariat - Borderline | 22.836 | 6.729 | 
| Secretariat - Over The Top | 26.406 | 8.413 | 
| Side Effects Project - Sing With Me | 17.073 | 12.129 | 
| Signe Jakobsen - What Have You Done To Me | 16.439 | 9.999 | 
| Skelpolu - Resurrection | 27.398 | 0.665 | 
| Speak Softly - Broken Man | 21.053 | 6.772 | 
| Speak Softly - Like Horses | 13.648 | 7.353 | 
| The Doppler Shift - Atrophy | 20.36 | 3.394 | 
| The Easton Ellises - Falcon 69 | 11.352 | 6.494 | 
| The Easton Ellises (Baumi) - SDRNR | 13.906 | 2.433 | 
| The Long Wait - Dark Horses | 21.987 | 7.209 | 
| The Mountaineering Club - Mallory | 27.447 | 12.238 | 
| The Sunshine Garcia Band - For I Am The Moon | 18.684 | 11.287 | 
| Timboz - Pony | 22.435 | 4.517 | 
| Tom McKenzie - Directions | 26.247 | 10.277 | 
| Triviul feat. The Fiend - Widow | 12.671 | 9.977 | 
| We Fell From The Sky - Not You | 26.073 | 5.367 | 
| Zeno - Signs | 18.11 | 7.174 | 
| **Total (median over frames, median over tracks)** | **5.249** | **11.155** |