---
license: cc-by-nc-sa-4.0
---

You can try listening to the performance of this model [here](https://huggingface.co/Sucial/Aspiration_Mel_Band_Roformer/tree/main/example_audio)

How to use the model?<br>
Try it with [ZFTurbo's Music-Source-Separation-Training](https://github.com/ZFTurbo/Music-Source-Separation-Training)

Description: The model is used to separate aspiration, which will be useful for mixing to some mixrs.<br>
Instruments: aspiration, other<br>
Dataset: My own datasets(171 songs for training and 17 songs for validation).<br>
Metrics: Based on the SDR of 17 songs for validation.<br>
Finetuned from: `model_mel_band_roformer_ep_3005_sdr_11.4360.ckpt`<br>
Configs: [config_aspiration_mel_band_roformer.yaml](./config_aspiration_mel_band_roformer.yaml)

Model: [aspiration_mel_band_roformer_sdr_18.9845.ckpt](./aspiration_mel_band_roformer_sdr_18.9845.ckpt)<br>
Epoch: 123<br>
Instr SDR aspiration: 9.8554<br>
Instr SDR other: 28.1136<br>
SDR Avg: 18.9845<br>

Model: [aspiration_mel_band_roformer_less_aggr_sdr_18.1201.ckpt](./aspiration_mel_band_roformer_less_aggr_sdr_18.1201.ckpt)<br>
Epoch: 27<br>
Instr SDR aspiration: 9.0704<br>
Instr SDR other: 27.1699<br>
SDR Avg: 18.1201<br>

Training logs:
![image](./training_logs.png)