| { |
| "model_name": "ear-vae", |
| "description": "εar-VAE: perceptually aligned, phase-aware music reconstruction autoencoder (VAE-GAN). This file is the repository manifest and the canonical entry point for programmatic loading; per-variant architecture configs live under config/.", |
| "paper": "https://arxiv.org/abs/2509.14912", |
| "license": "apache-2.0", |
| "default_variant": "ear_vae_v2_48k", |
| "variants": { |
| "ear_vae_44k": { |
| "sample_rate": 44100, |
| "io_channels": 2, |
| "latent_dim": 64, |
| "downsampling_ratio": 1024, |
| "config": "config/model_config.json", |
| "weights": "pretrained_weight/ear_vae_44k.pyt" |
| }, |
| "ear_vae_v2_48k": { |
| "sample_rate": 48000, |
| "io_channels": 2, |
| "latent_dim": 64, |
| "downsampling_ratio": 960, |
| "config": "config/ear_vae_v2.json", |
| "weights": "pretrained_weight/ear_vae_v2_48k.pyt" |
| } |
| } |
| } |
|
|
