Update README.md
Browse files
README.md
CHANGED
@@ -11,14 +11,11 @@ datasets:
|
|
11 |
- MUSDB18-HQ
|
12 |
metrics:
|
13 |
- SDR
|
14 |
-
- SIR
|
15 |
-
- SAR
|
16 |
---
|
17 |
|
18 |
-
# Sampling-frequency-independent (SFI) Conv-TasNet
|
19 |
-
|
20 |
-
|
21 |
-
It was proposed in [our IEEE/ACM Trans. ASLP paper](https://doi.org/10.1109/TASLP.2022.3203907).
|
22 |
This model was trained with 32 kHz-sampled data but works well with untrained sampling frequencies (e.g., 8, 16 kHz).
|
23 |
|
24 |
# License
|
@@ -42,6 +39,3 @@ Please cite the following paper.
|
|
42 |
# Contents
|
43 |
- Four trained models (seed=40,42,44,47)
|
44 |
- Evaluation results (json files obtained with the museval library)
|
45 |
-
|
46 |
-
# Results on vocal ensemble separation
|
47 |
-
|
|
|
11 |
- MUSDB18-HQ
|
12 |
metrics:
|
13 |
- SDR
|
|
|
|
|
14 |
---
|
15 |
|
16 |
+
# Sampling-frequency-independent (SFI) Conv-TasNet trained with the MUSDB18-HQ dataset for music source separation.
|
17 |
+
This model was proposed in [our IEEE/ACM Trans. ASLP paper](https://doi.org/10.1109/TASLP.2022.3203907) and works well with untrained sampling frequencies by using sampling-frequency-independent convolutional layers with the frequency domain filter design.
|
18 |
+
It was trained by Tomohiko Nakamura using [the codebase](https://github.com/TomohikoNakamura/sfi_convtasnet)).
|
|
|
19 |
This model was trained with 32 kHz-sampled data but works well with untrained sampling frequencies (e.g., 8, 16 kHz).
|
20 |
|
21 |
# License
|
|
|
39 |
# Contents
|
40 |
- Four trained models (seed=40,42,44,47)
|
41 |
- Evaluation results (json files obtained with the museval library)
|
|
|
|
|
|