Views
No views yet
vocal_gender (ACE-Step)1from src.steering import SteerableACEModel, CAASteeringController
2
3model = SteerableACEModel(device="cuda")
4model.pipeline.load()
5ctrl = CAASteeringController.from_pretrained("lukasz-staniszewski/ace-step-rfm-vocal-gender", alpha=20.0)
6
7with model.steer(ctrl):
8 audio = model.generate(
9 prompt="instrumental music", lyrics="[inst]",
10 audio_duration=10.0, infer_step=30, manual_seed=0,
11 )1{
2 "method": "rfm",
3 "concept": "vocal_gender",
4 "lyrics": "",
5 "num_cfg_passes": 2,
6 "save_all_cfg_passes": true,
7 "audio_duration": 30.0,
8 "num_inference_steps": 30,
9 "seed": 10,
10 "device": "cuda",
11 "save_dir": "steering_vectors/rfm",
12 "guidance_scale_text": 0.0,
13 "guidance_scale_lyric": 0.0,
14 "guidance_scale": 5.0,
15 "guidance_interval": 1.0,
16 "guidance_interval_decay": 0.0,
17 "rfm_iters": 30,
18 "n_components": 12,
19 "hyperparam_samples": 100,
20 "val_frac": 0.2
21}